Network World
networkworld.com.web.brid.gy
Network World
@networkworld.com.web.brid.gy
Network World provides news and analysis of enterprise data center technologies, including networking, storage, servers and virtualization.

[bridged from https://networkworld.com/ on the web: https://fed.brid.gy/web/networkworld.com ]
Cisco identifies vulnerability in ISE network access control devices
The latest flaw in Cisco Systems Identity Services Engine (ISE), which could expose sensitive information to an attacker, requires rotation of credentials as well as installation of a patch to correct, says an expert. Cisco ISE is a network access control platform that enforces access policy and manages endpoints. There have been more critical holes in Cisco products, acknowledged Paddy Harrington, a senior analyst at Forrester Research, and this one does need a threat actor with administrative privileges to execute and get read access to sensitive information. “However,” he advised senior infosec leaders with Cisco ISE servers, “don’t let these things hang around.” Before patching, he said, admins should: * rotate ISE credentials for those with existing and approved access; * ensure only those who need access have credentials; * reduce the number of devices that can access the ISE server; * patch as soon as it’s possible to take the server offline. In its notice to customers, Cisco says a vulnerability [CVE-2026-20029] in the licensing features of ISE and Cisco ISE Passive Identity Connector (ISE-PIC) could allow an authenticated remote attacker with administrative privileges to gain access to sensitive information. It isn’t clear why this is called a licensing feature vulnerability. Cisco didn’t respond by deadline when asked for an explanation. The advisory, which describes the problem as of medium criticality, with a CVSS score of 4.9, says the vulnerability is due to improper parsing of XML that is processed by the web-based management interface of Cisco ISE and Cisco ISE-PIC. Johannes Ullrich, dean of research at the SANS Institute, said, “Most likely, this is an XML External Entity vulnerability.” External entities, he explained, are an XML feature that instructs the parser to either read local files or access external URLs. In this case, an attacker could embed an external entity in the license file, instructing the XML parser to read a confidential file and include it in the response. This is a common vulnerability in XML parsers, he said, typically mitigated by disabling external entity parsing. An attacker would be able to obtain read access to confidential files like configuration files, he added, and possibly user credentials. Ullrich also said an ISE administrator may have access to a lot of the information, but they should not have access to user credentials. The Cisco advisory says an attacker could exploit this vulnerability by uploading a malicious file to the application: “A successful exploit could allow the attacker to read arbitrary files from the underlying operating system that could include sensitive data that should otherwise be inaccessible even to administrators. To exploit this vulnerability, the attacker must have valid administrative credentials.” Cisco said proof-of-concept exploit code is available for this vulnerability, but so far the company isn’t aware of any malicious use of the hole. These days, admin credentials aren’t hard to get, Harrington noted. The “dirty secret that few people want to talk about is across IT and security operations there are so many systems that are left with default credentials.” That’s particularly common, he said, with devices behind a firewall, such as network access control servers, because admins think because they are inside the network they can’t be touched by external hackers. But lots of credentials can be scooped up in compromises of applications where Cisco admins might have stored passwords. **Related content:Cisco warns of three critical ISE vulnerabilities** Coincidentally, today researchers at SCORadar released an analysis of data thefts in 2025. Among other things, it notes that credential theft hit a new high last year. A total of 388 million credentials were stolen from the ten most affected platforms, including Facebook, Google, and Roblox.
www.networkworld.com
January 9, 2026 at 6:27 AM
The state of Enterprise Linux for networking
The open-source Linux operating system has emerged to be the foundation for cloud and networking across industries. Enterprise Linux distributions form the core of modern networking setups. They deliver reliable, secure platforms for data centers, cloud systems, edge devices and telecom networks. While there is a core open-source Linux kernel that is at the heart of the operating system, there are many different vendor distributions. Beyond just the kernel there are also different ways that distributions handle networking. The modern landscape is dominated by distributions that emphasize long-term support, compatibility and specialized features for telecommunications and general networking. There are both general-purpose operating systems that have networking capabilities as well as purpose-built network operating systems (NOS) available. This article examines the key players where they come from, what they offer and how they differentiate on networking capabilities. ## Red Hat Enterprise Linux (RHEL) Red Hat Enterprise Linux (commonly referred to as RHEL) is arguably the first enterprise Linux distribution to have hit the market in the early 2000s. The basic idea behind the ‘Enterprise’ designation originally was all about long term support and stability. Red Hat was acquired by IBM in 2019 for $34 billion and has continued as an operating unit of IBM ever since. **Latest release.** RHEL 10 was released on May 25, 2025 and will be supported until 2035. **Standout features** : RHEL 10 integrates the Lightspeed AI set of tools that aim to help improve overall system management and administration. The update also has **post-** quantum cryptography (PQC) support, optimized cloud integration and improved SELinux for security. RHEL also integrates the Podman container platform which is Red Hat’s competitive offering to the Docker containers used by other vendors. It’s also worth noting that Red Hat’s OpenShift is a leading Kubernetes cloud-native distribution and RHEL is often the primary base, providing optimizations and easier integration**.** ##### **Key networking features:** * **Encrypted DNS.** Support for DNS over TLS (DoT) and DNS over HTTPS (DoH) * **DPDK**. Support for the Data Plane Development Kit (DKPK) enables Network Function Virtualization (NFV). * **Netavark.** Podman Container Networking which is RHEL’s default for container uses the Netavark system instead of the Container Network Interface (CNI) that is common on other distributions. * **NetworkManager** , RHEL 10 uses the open-source NetworkManager as its standard tool for network configuration. * **Real-Time.** There is an available Real-Time kernel that is useful for telco and mission-critical industries for more deterministic processing times ##### **Other RHEL-based Linux distributions** Of note, there are multiple distributions that are largely based on RHEL in the market today as well. Among them are Oracle Linux which enables users to swap in an Oracle Unbreakable Linux kernel, which can offer potential advantages. The Alma Linux and Rocky Linux distributions got their starts from Red Hat’s CentOS community version of RHEL. Oracle, Alma and Rocky Linux all adhere to the OpenELA (Open Enterprise Linux Association) specifications to align with RHEL for approximate feature compatibility. ## SUSE Linux Enterprise Server (SLES) SUSE Linux got its start in Europe and now serves organizations around the world. The company is currently privately-held and has a vibrant open-source community. **Latest release:** SUSE Linux Enterprise Server 16.0 became generally available on November 4, 2025 and will be supported until November 30, 2038. **Standout features:** AI integration is a standout feature with built-in model context protocol (MCP) host for agentic AI applications. SUSE also has integrated its Adaptable Linux Platform that eliminates dependency hell by decoupling applications from the host. ##### **Key networking features:** * **NetworkManager.** SLES had previously relied on the ‘wicked’ networking stack but as of SLES 16 has standardized on NetworkManager. * **NFTables firewall framework:** Modern packet filtering replacement for IPTables with simplified syntax and improved performance * **KEA DHCP server:** Next-generation DHCP implementation replacing ISC DHCP, * **Cockpit web-based network management:** Remote network administration through browser interface with real-time monitoring and configuration capabilities ## Canonical Ubuntu ** **Ubuntu got its start in 2004, originally largely based on the popular Debian community Linux distribution. Canonical, the lead commercial vendor behind Ubuntu has emphasized telco and networking use cases and the distribution is widely deployed in communications infrastructure. **Latest release**. The latest release is Ubuntu 25.10 though that isn’t a long term supported release. The most recent long term support (LTS) release is Ubuntu 24.04, which was released in April 2024. The next Ubuntu LTS is set to debut in April 2026 with a 12-year support lifespan. **Standout features.** Ubuntu is known for its easy to use desktop interface and it has strong Kubernetes integrations with the platform’s MicroK8s system. The Metal-as-a-Service (MaaS) system enables simplified bare-metal provisioning and the Juju orchestration system can help administrators with application deployments. ##### **Key networking features:** * **Netplan.** Ubuntu stands out with its Netplan tools which is a YAML markup based declarative configuration tool that abstracts away the complexity of the systemd-networkd foundation. * **Open vSwitch.** SupportsIPv4/IPv6, DHCP and SDN deployments. * **SR-IOV (Single Root I/O Virtualization).** Enables a single network interface card (NIC) to appear as multiple system for a virtual machine (VM). ## Dedicated Linux-based networking operating systems Beyond the general-purpose enterprise Linux platform that are well-suited for networking, there are also a series of purpose-built Linux-based networking operating systems as well. Among the most widely deployed are: ### SONiC (Software for Open Networking in the Cloud) Microsoft launched SONiC in 2016 and it became a Linux Foundation project in 2022. SONiC is a Debian Linux based open network operating system for switches that is largely hardware-agnostic and modular. SONiC is used by a growing list of both established and startup networking vendors to provide a base NOS. **Latest release.** SONiC 4.5 was released in May 2025 and will be community-supported until at least October 2026. Each commercial vendor that provides its own SONiC-based system can extend and support the platform for longer periods of time. ##### Key networking features: * **MCLAG**(Multi-Chassis Link Aggregation) at Layer 2 and Layer 3. * **Weighted ECMP** (Weighted Equal Cost Multipath) for fine-grained traffic engineering. * BGP, OSPF, BFD, and IS-IS routing protocol support via FRRouting ### Nvidia Cumulus Linux Cumulus Linux was an early pioneer in the NOS space back in 2010. The company was acquired by Nvidia a decade later in 2020. Like SONiC it is a switch optimized NOS. **Latest release:** Cumulus Linux 5.15 is the most recent version, though it is not a long term supported release. Cumulus Linux 5.11 which debuted in 2024 is an LTS and is supported until 2027. ##### **Key networking features:** * **Unnumbered Interfaces:** Simplified IP approach for BGP and OSPF, requiring only one template for leaf and spine nodes * **Redistribute Neighbor (RDNBR**): Enables VM and host mobility with layer-3 discovery * **Prescriptive Topology Manager (PTM)** : Verifies connections and resolves issues efficiently * **Nvidia User Experience (NVUE)** : Full CLI object model enabling advanced programmability _Support periods may vary based on specific editions and commercial agreements. Visit vendor websites for detailed information._ Foundry
www.networkworld.com
January 9, 2026 at 12:48 AM
Holes in Veeam Backup suite allow remote code execution, creation of malicious backup config files
Veeam says that four vulnerabilities could allow a person with certain oversight roles for its flagship Backup & Replication suite to do serious damage to – but not destroy – a backup database. The company has already issued a patch for the bugs, which, it says, should be applied immediately. The worst of the vulnerabilities, CVE-2025-59470, carries a criticality score of 9 and would allow a threat actor “to do something nefarious,” said Rick Vanover, Veeam’s vice-president of product strategy. But he emphasized that, because of the immutable nature of the backup, data can’t be destroyed. The issue: Veeam discovered that a person with the role of Backup Admin, Backup Operator, or Tape Operator status in unpatched version 13 of the suite (versions 13.0.1.180 and earlier) have more permissions than they should. The patch corrects that. Specifically, the flaws addressed are: * CVE-2025-59470 (with a CVSS score of 9) allows a Backup or Tape Operator to perform remote code execution (RCE) as the Postgres user by sending a malicious interval or order parameter; * CVE-2025-59469 (with a severity score of 7.2) allows a Backup or Tape Operator to write files as root; * CVE-2025-55125 (with a severity score of 7.2) allows a Backup or Tape Operator to perform remote code execution (RCE) as root by creating a malicious backup configuration file; * CVE-2025-59468 (with a severity score of 6.7) allows a Backup Administrator to perform remote code execution (RCE) as the Postgres user by sending a malicious password parameter. The patch to version 13.0.1.1071 will be an “easy installation” that won’t be disruptive, Vanover said. As of Tuesday afternoon, Veeam hadn’t received reports of exploitation, he added. “The good news is, if a Veeam server is broken, we can create a new server right away – presumably with this patch installed – import the backups and carry on. The core data is completely unimpacted by this,” Vanover said. “The worst type of thing would be the [backup] environment isn’t working right or the Postgres database is messed up on the Veeam server, so jobs might not behave in a way one might expect.” In these cases, admins using the Veeam One monitoring management suite would get an alert if, for example, a job was unable to connect to the backup server or backup jobs were failing. The four vulnerabilities being patched are less severe than some because an attacker, internal or external, would need valid credentials for the three specific roles, noted Johannes Ullrich, dean of research at the SANS Institute. On the other hand, he added, backup systems like Veeam are targets for attackers, in particular those who inject ransomware, who often attempt to erase backups. “Backup systems should be regularly audited to ensure that access rights, such as those mentioned in this vulnerability, are properly managed and only accessible to users who actually need them,” he said. “Authentication credentials should be reviewed to ensure they comply with the respective standards.” Kellman Meghu, principal security architect at Canadian-based risk management firm DeepCove Cybersecurity, said the worry is how the vulnerabilities could be used by a threat actor to get root privileges to the backup, “which is the worst it can get as far as compromise. From the sounds of the exploit, just being able to update a config file could be the avenue for executing malicious commands at the highest privileges.” Admins who can’t patch quickly, or who have been running unpatched versions for any length of time, should first audit all config files and operations to ensure there have been no changes to the config files or execution of additional unexpected actions. Alerts should be set for every backup process run, so it is closely monitored until the suite can be patched. “Keep in mind,” he added, “if you do see unusual behavior, it is a sign that there is a malicious actor or inside threat operating, and you would need to take a holistic incident response.”
www.networkworld.com
January 9, 2026 at 12:47 AM
Lenovo unveils purpose-built AI inferencing servers
Lenovo Group Ltd. has introduced a range of new enterprise-level servers designed specifically for AI inference tasks. The servers are part of Lenovo’s Hybrid AI Advantage lineup, a family of inferencing devices. Nvidia has captured the training space, where large language models (LLMs) are generated, but the inferencing space, where the LLMs are put to work doing things like answering questions and making decisions, is wide open with no clear leader. But it is growing fast. Futurum Group estimates the global AI inference infrastructure market will grow from $5.0 billion in 2024 to $48.8 billion by 2030, for a six-year CAGR of 46.3%. Lenovo says that moving from training to action turns the significant capital committed to AI into tangible business return, and invaluable competitive gain. Its new AI Inferencing suite executes AI workloads across an organization’s cloud, data center, and edge to wherever they deliver the greatest value. “Enterprises today need AI that can turn massive amounts of data into insight the moment it’s created,” said Ashley Gorakhpurwalla, executive vice president at Lenovo and president of Lenovo Infrastructure Solutions Group in a statement. “With Lenovo’s new inferencing-optimized infrastructure, we are giving customers that real-time advantage—transforming massive amount of data into instant, actionable intelligence that fuels stronger decisions, greater security, and faster innovation.” The first server is the Lenovo ThinkSystem SR675i, a high-end server featuring AMD Eypc server CPUs and Nvidia Blackwell GPUs and is built to handle large language models at scale and speed up simulations in sectors like healthcare, manufacturing, and finance. There is also the Lenovo ThinkSystem SR650i, which offers high-density GPU computing power for faster AI inference and is intended for easy installation in existing data centers to work with existing systems. Finally, there is the Lenovo ThinkEdge SE455i for smaller, edge locations such as retail outlets, telecom sites, and industrial facilities. Its compact design allows for low-latency AI inference close to where data is generated and is rugged enough to operate in temperatures ranging from -5°C to 55°C. All of the servers include Lenovo’s Neptune air- and liquid-cooling technology and are available through the TruScale pay-as-you-go pricing model. In addition to the new hardware, Lenovo introduced new AI Advisory Services with AI Factory Integration. This service gives access to professionals for identifying, deploying, and managing best-fit AI Inferencing servers. It also launched Premier Support Plus, a service that gives professional assistance in data center management, freeing up IT resources for more important projects.
www.networkworld.com
January 7, 2026 at 6:32 PM
AWS stealthily raises GPU prices by 15 percent
Amazon Web Services (AWS) raised the prices of its GPU instances for machine learning by around 15 percent this weekend, without warning, reports The Register. The price increase applies in particular to EC2 Capacity Blocks for ML, where, for example, the cost of the p5e.48xlarge instance rose from $ 34.61 to $ 39.80 per hour. AWS had previously announced that their prices would be updated in January, but did not indicate in which direction. In the past, the company has almost always lowered prices or changed pricing models rather than raising them directly. Even for larger customers with fixed discounts, the increase means that their actual costs will increase anyway, as discounts are calculated on list prices. AWS justifies the increase itself with “changing supply and demand patterns”. This article originally appeared on Computer Sweden, for further reporting, see “AWS hikes prices for EC2 Capacity Blocks amid soaring GPU demand.” More AWS news: * AWS finally moves to simplify multicloud operations with Google * With AI Factories, AWS aims to help enterprises scale AI while respecting data sovereignty * AWS launches ‘Capabilities by Region’ to simplify planning for cloud deployments[ ](https://www.infoworld.com/article/4105220/aws-finally-listened-to-its-customers.html)
www.networkworld.com
January 7, 2026 at 3:30 PM
Samsung warns of memory shortages driving industry-wide price surge in 2026
Samsung Electronics warned that memory chip shortages will drive price increases across the electronics industry in 2026, with the world’s largest memory manufacturer acknowledging that even its vast production capacity cannot insulate its own products from the surge, a signal that enterprise IT buyers face unavoidable cost increases regardless of vendor choice. Wonjin Lee, president and head of global marketing at Samsung, told Bloomberg in an interview that the company expects memory chip shortages to affect pricing industry-wide. “In 2026, there’s going to be issues around semiconductor supplies, and it’s going to affect everyone, not just Samsung,” Lee said. “I think it’s an industry-wide reality that we’re going to see some supply issues.” Samsung’s inability to insulate its own product lines from price pressures represents a shift in market dynamics, according to Manish Rawat, semiconductor analyst at TechInsights. “Memory manufacturers once functioned as shock absorbers for the tech ecosystem, using scale, inventory discipline, and long-term contracts to provide pricing and supply predictability,” Rawat said. “Samsung’s inability to cushion volatility despite its unmatched capacity indicates a market in disequilibrium.” ## AI infrastructure drains conventional memory supply The disequilibrium stems from manufacturers’ reallocation of production capacity toward high-bandwidth memory for AI data centers. HBM commands higher margins than conventional DRAM, driving Samsung, SK Hynix, and Micron to shift capacity away from traditional enterprise and consumer memory products. “Demand for memory is strong, driven by ongoing AI investments,” said Kanishka Chauhan, senior principal analyst at Gartner. “High Bandwidth Memory, which is used in AI applications, demands stronger pricing compared to traditional and legacy memory, leading vendors to allocate more production capacity to HBM.” HBM production for AI accelerators consumes approximately three times the wafer capacity of standard DRAM per gigabyte, according to a Micron executive, forcing memory makers to reallocate production away from consumer and enterprise products. As a result, the supply of both traditional and legacy DRAM has decreased for industrial and PC customers, particularly those with lower volume requirements, Chauhan said. SK Hynix reported during its October earnings call that its HBM, DRAM, and NAND capacity is “essentially sold out” for 2026, while Micron recently exited the consumer memory market entirely to focus on enterprise and AI customers. ## Enterprise hardware costs surge The supply constraints have translated directly into sharp price increases across enterprise hardware. Samsung raised prices for 32GB DDR5 modules to $239 from $149 in September, a 60% increase, while contract pricing for DDR5 has surged more than 100%, reaching $19.50 per unit compared to around $7 earlier in 2025. DRAM prices have already risen approximately 50% year to date and are expected to climb another 30% in Q4 2025, followed by an additional 20% in early 2026, according to Counterpoint Research. The firm projected that DDR5 64GB RDIMM modules, widely used in enterprise data centers, could cost twice as much by the end of 2026 as they did in early 2025. Gartner forecast DRAM prices to increase by 47% in 2026 due to significant undersupply in both traditional and legacy DRAM markets, Chauhan said. ## Procurement leverage shifts to hyperscalers The pricing pressures and supply constraints are reshaping the power dynamics in enterprise procurement. For enterprise procurement, supplier size no longer guarantees stability. “As supply becomes more contested in 2026, procurement leverage will hinge less on volume and more on strategic alignment,” Rawat said. Hyperscale cloud providers secure supply through long-term commitments, capacity reservations, and direct fab investments, obtaining lower costs and assured availability. Mid-market firms rely on shorter contracts and spot sourcing, competing for residual capacity after large buyers claim priority supply. “This imbalance creates a dual constraint for the mid-market: higher input costs and longer delivery timelines,” Rawat said. “Both directly limit their ability to scale infrastructure, deploy new workloads, or innovate at pace.” Samsung has announced plans to build a new memory production line at its Pyeongtaek, South Korea plant, but mass production will not begin until 2028. For enterprises, Samsung’s warning indicates that memory constraints will affect IT procurement strategies in 2026 and beyond, requiring organizations to account for supply availability and cost volatility in hardware planning. Samsung, SK Hynix, Micron, and Nvidia did not immediately respond to requests for comment.
www.networkworld.com
January 7, 2026 at 3:32 PM
Lenovo-Nvidia partnership targets faster AI infrastructure rollouts
Lenovo is pitching a faster path to enterprise AI infrastructure, pairing its liquid-cooled systems and networking with Nvidia platforms to deliver what it calls “AI cloud gigafactories” designed to reduce deployment timelines from months to weeks. The announcement, made at CES in Las Vegas, reflects growing pressure on enterprises to build AI infrastructure faster than traditional data center build cycles allow, even as networking, power, and cooling constraints continue to slow deployments. In a statement, Lenovo said the program focuses on speeding time to first token for AI cloud providers by simplifying the deployment of large-scale AI infrastructure through pre-integrated systems and deployment support. ## Progress in pre-integrated systems Analysts say the promise of deploying AI cloud infrastructure in weeks reflects progress in pre-integrated systems, but caution that most enterprise deployments still face practical constraints. “The AI data center rollout has been hindered by a lack of a clear business case, supply chain challenges, and insufficient internal system integration and engineering capability,” said Lian Jye Su, chief analyst at Omdia. He said that Lenovo’s claim is plausible because of its partnership with Nvidia and the use of a pre-validated, modular infrastructure solution. Others stressed that such timelines depend heavily on operating conditions. Franco Chiam, vice president at IDC Asia Pacific, cautioned that deployments are rarely limited by hardware delivery alone. “AI racks can draw 30 to 100 kilowatts or more per cabinet, and many existing facilities lack the electrical capacity, redundancy, or permitting approvals to support that density without significant upgrades,” he said. Jaishiv Prakash, director analyst at Gartner, said Lenovo’s timeline of weeks is realistic for “time to first token” when facilities already have power, fiber, and liquid cooling in place. “In practice, however, delays are often caused by utility power and electrical gear lead times, direct-to-chip liquid cooling integration, and high-capacity fiber transport,” Prakash said. “Without that groundwork, timelines can extend to months or even quarters.” ## How Lenovo’s approach differs By combining integrated hardware with services for regulated environments, Lenovo is aiming to establish a middle ground between hyperscalers and traditional enterprise vendors. Su said this approach stands out because it combines Lenovo’s own power and cooling technologies, including Neptune liquid cooling, with Nvidia GPUs, while also pairing hardware with consulting and integration services. Chiam said a key differentiator of the “AI cloud gigafactory” is Lenovo’s ability to pair its hardware-centric DNA with hybrid deployment flexibility, a strategic advantage in an era increasingly shaped by data sovereignty concerns. “Unlike hyperscalers or pure-play cloud vendors that prioritize fully managed, centralized AI stacks, Lenovo’s approach integrates tightly optimized, on-premises and edge-capable infrastructure with cloud-like scalability,” Chiam added. “This is particularly compelling for enterprises and sovereign enterprises that require localized AI processing without sacrificing performance.” ## What it means for enterprise networks Analysts say the Lenovo-Nvidia partnership underscores how AI infrastructure is reshaping the role of the enterprise network, pushing it beyond traditional connectivity toward a performance-critical control layer. Shriya Mehrotra, director analyst at Gartner, said the partnership transforms the network into a high-performance “control plane” using 800GbE fabrics and real-time telemetry to keep GPUs saturated and prevent training failures. “To prevent high-cost GPUs from sitting idle, teams must optimize backend fabrics by adopting 400-800GbE or InfiniBand to manage the massive ‘east-west’ traffic common in AI training,” Mehrotra added. However, the speed promised by the Lenovo and Nvidia partnership comes with a strategic price tag: architectural rigidity. “Speed comes from alignment, not optionality,” said Manish Rawat, analyst at TechInsights. “Pre-integrated stacks reduce time-to-value, but they also deepen vendor lock-in at the networking, interconnect, and software layers.” Rawat said enterprises should segment workloads carefully, using tightly integrated AI factory designs for performance-critical training while preserving more open architectures for inference and general enterprise workloads.
www.networkworld.com
January 7, 2026 at 11:20 AM
Ethernet groups keep 2026 focus on higher bandwidth, AI demands
2025 marked a year of progress for Ethernet with multiple standards and specifications reaching completion as the networking industry works to keep pace with AI and HPC demands. Among the leading efforts for Ethernet in 2025 was the Ultra Ethernet Consortium, an open-source effort under the governance of the Linux Foundation that aims to enhance Ethernet to better compete and surpass the capabilities of Infiniband. The Ultra Ethernet Consortium (UEC) released its 1.0 specification in June, with an update to 1.0.1 in September. Looking at the core Ethernet standards, the IEEE 802.3dj standard moved closer to completion, enabling a new era of 1.6 Terabit per second (Tbps) high-speed operations. “The Ethernet Alliance had a strong year in 2025, marked by continued growth and technical leadership,” Peter Jones, chair of the Ethernet Alliance, told _Network World._ ## Ethernet getting broader and faster Jones noted that the IEEE 802.3dj standard, which defines 200G/400G/800G/1.6T at 200G/lane, is on track for completion in late 2026. Early 200G/lane products are expected to reach the market during the year. With 200G/lane almost done, the community is already looking ahead. “While this work is receiving substantial attention, the community is already preparing for the next step: initiating a 400G/lane project to address the demands of the hyperscalers for AI networking,” Jones said. The 400G/lane project represents the next step in Ethernet’s bandwidth evolution. The Ethernet Alliance will continue its ecosystem work in 2026, demonstrating interoperability at multiple events and organizing another 200G/lane plugfest. ## Ultra Ethernet adapts to evolving AI workloads Following the release of its 1.0 specification, the Ultra Ethernet Consortium is tackling three technical priorities in 2026 aimed at improving Ethernet performance for AI and HPC workloads. The first is Programmable Congestion Management (PCM). “AI and HPC workloads are rapidly evolving,” Chad Hintz, co-chair of marketing at UEC and principal member of technical staff at AMD, told _Network World_. “To address these rapid changes, flexible congestion management mechanisms are required.” PCM will enable anyone to implement a new congestion control algorithm using a standard language, and that algorithm will work on any NIC that supports UE PCM. The UEC is also standardizing Congestion Signaling (CSIG), which allows packets to carry high-fidelity information about network congestion. This enables the transport protocol to react more accurately and quickly to changing conditions. The second priority addresses small message performance. The Ultra Ethernet Transport (UET) 1.0 protocol is designed to support up to a million hosts coordinating as part of a single job. To achieve this scale, a basic UET packet has 104 bytes of headers. That overhead is just 2.5% for 4096-byte packets but becomes more significant for smaller transactions like 256-byte transfers. The smaller packet overhead will improve efficiency for workloads with small payloads, whether HPC workloads or local scale-up networks. “UEC is pursuing optimizations across all layers of the stack, including a reduced-size forwarding header, looking to cut this overhead in half for optimized deployments,” Hintz said. The third focus is In-Network Collectives (INC) for Ethernet networks. By moving the reduction operation common to AI and HPC workloads from hosts into the network, INC can help to improve performance on certain workloads. ## Ethernet development extends beyond AI While AI networking captures much of the attention, IEEE 802.3 maintains active work across a broader landscape. Jones noted that projects underway include 10 Mb/s single-pair multidrop Ethernet, 100 Mb/s long-reach single-pair Ethernet and asymmetric Ethernet for automotive sensor networks. The group is also working on metadata services to support advanced applications, including those defined by the Ultra Ethernet Consortium. In a move toward open collaboration, IEEE 802.3 is launching an open source project to produce YANG modules for the standard. The theme running through all of Ethernet’s 2026 work is speed. Not just the speed of the technology itself, but the speed at which the industry must move to stay ahead of customer demands. “As an industry, the challenge remains unchanged: moving as fast as possible to stay ahead of accelerating customer demands,” Jones said. “In that sense, 2026 may feel very much like 2025. Proof that progress in Ethernet, as the saying goes, no good deed goes unpunished.”
www.networkworld.com
January 7, 2026 at 11:19 AM
2026 network outage report and internet health check
ThousandEyes, a Cisco company, monitors how ISPs, cloud providers and conferencing services are handling any performance challenges and provides _Network World_ with a weekly roundup of events that impact service delivery. Read on to see the latest analysis, and stop back next week for another update on internet and cloud traffic performance. _Note: We have archived prior-year outage updates, including our reports from2025, 2024, 2023, and the Covid-19 era._ ## Internet report for Dec. 29, 2025-Jan. 4, 2026 ThousandEyes reported 199 global network outage events across ISPs, cloud service provider networks, collaboration app networks, and edge networks (including DNS, content delivery networks, and security as a service) during the week of December 29 through January 4. The total of outage events decreased by 14% compared to the 231 outages from the week prior. Specific to the U.S., there were 71 outages, which is down by 29% from 100 outages the week prior. Here’s a breakdown by category: * **ISP outages** : Globally, ISP outages declined from 136 to 105, a 23% decrease. In the U.S., ISP outages dropped from 51 to 26, down 49% week-over-week. * **Public cloud network outages:** Globally, public cloud network outages increased from 47 to 59, up 26% week-over-week. In the U.S., outages rose from 22 to 37, an increase of 68%. * **Collaboration app network outages:** Both globally and in the U.S., collaboration application network outages remained at zero. Cisco ThousandEyes ### Two notable outages On January 2, Hurricane Electric, a network transit provider, headquartered in Fremont, CA, experienced an outage that impacted customers and downstream partners across multiple regions, including the U.S., Japan, South Korea, Indonesia, Taiwan, India, and Singapore. The outage, lasting one hour and 1 minute, was first observed around 3:05 PM EST and initially appeared to center on Hurricane Electric nodes located in Los Angeles, CA. Five minutes into the outage, the nodes located in Los Angeles, CA, were joined by Hurricane Electric nodes located in Phoenix, AZ, in exhibiting outage conditions. This coincided with an increase in the number of downstream partners and countries impacted. Around 10 minutes into the outage, all nodes, except those located in Los Angeles, CA, appeared to clear. The outage was cleared at around 4:10 PM EST. Click here for an interactive view. On December 31, Cogent Communications, a multinational transit provider based in the U.S., experienced an outage that impacted multiple downstream providers as well as Cogent customers across various regions, including the U.S., France, the Netherlands, South Africa, Spain, and South Korea. The outage, which lasted 9 minutes, was first observed around 7:50 PM EST and appeared to initially center on Cogent nodes located in Washington, D.C. Around five minutes after first being observed, nodes located in Washington, D.C., appeared to clear and were replaced by nodes located in Boston, MA, in exhibiting outage conditions. The outage was resolved around 8:00 PM EST. Click here for an interactive view.
www.networkworld.com
January 7, 2026 at 7:57 AM
The top AMD stories of 2025
2025 was a pivotal year for AMD, with product launches, market moves, and security warnings among the highlights. Additions to AMD’s product portfolio were featured in our two most-read articles. The Pensando Pollara AI NIC captured reader interest as it’s the first NIC that complies with the Ultra Ethernet Consortium’s 1.0 specification. AMD’s Ultra Ethernet-ready network card is designed for massive environments with thousands of AI processors. Targeting a very different part of the AI spectrum are AMD’s Ryzen Threadripper CPUs, which are designed for high-end AI desktops and represent a challenge to Intel’s dominance in workstations. Partnerships also made waves in the industry. As AMD works to expand its influence, it teamed with the U.S. Department of Energy to build two more supercomputers, struck a 6-gigawatt deal with OpenAI to supply chips, and announced plans to combine IBM’s quantum computers with AMD’s CPUs, GPUs and FPGAs in a blend of quantum and classical computing architectures. These events and more make up our top AMD stories of 2025: ### 1. AMD rolls out first Ultra Ethernet-compliant NIC Our most-read story about AMD reports on a first for the semiconductor company: AMD’s Pensando Pollara 400GbE network interface card is the industry’s first Ultra Ethernet-compliant NIC. (The UEC was launched in 2023 under the Linux Foundation with the goal of increasing the scale and reliability of Ethernet networks.) Designed for massive scale-out environments containing thousands of AI processors, Pollara enables GPU-to-GPU communication with intelligent routing technologies to reduce latency, making it very similar to Nvidia’s NVLink c2c. AMD revealed the new Ultra Ethernet-based networking card in June at its Advancing AI event. ### 2. AMD launches new Ryzen Threadripper CPUs to challenge Intel’s workstation dominance Marking an aggressive push into the professional workstation and high-end desktop segments, AMD launched new HPC processors at Computex 2025 in May. The Ryzen Threadripper Pro 9000 WX-Series and Ryzen Threadripper 9000 Series are built to handle tough workloads such as VFX rendering, scientific simulation, CAD, and AI development for enterprise-grade workstations in industries such as engineering, healthcare, defense, and AI development. “AMD is positioning itself as a serious competitor in high-end enterprise environments by offering scalable performance, platform stability, and enterprise-grade manageability through AMD Pro Technologies,” said Manish Rawat, semiconductor analyst at TechInsights. ### 3. AMD targets hosting providers with affordable EPYC 4005 processors AMD launched new data center processors that target hosted IT service providers. Announced in May, the EPYC 4005 series is built with enterprise-class features and support for modern infrastructure technologies at an affordable price, the company said. “The EPYC 4005 is a single socket platform and is thus designed with a ‘scale out’ strategy in mind, appropriate for small and medium enterprise requirements. The AM5 socket further helps simplify the server design and [bill of materials] costs to keep the system cost-effective,” said Neil Shah, co-founder and vice president for research at Counterpoint Research. ### 4. AMD patches microcode security holes after accidental early disclosure AMD issued two patches in February for severe microcode security flaws, defects that AMD said “could lead to the loss of Secure Encrypted Virtualization (SEV) protection.” The bugs were inadvertently revealed by a partner, and AMD hustled to deliver the patches. Matt Kimball, vice president and principal analyst at Moor Insights & Strategy, said he believed AMD handled the situation well: “It’s good to see AMD working with its community to solve for these vulnerabilities quickly. The amount of work that goes into providing a fix — and thoroughly testing it — is extensive… It is an unfortunate reality that these vulnerabilities find their way into systems, but it’s a reality nonetheless. The real measure of a vendor is how quickly they respond to mitigating and nullifying these vulnerabilities. In the case of AMD, the response was swift and thorough.” ### 5. AMD to build two more supercomputers at Oak Ridge National Labs The collaboration between AMD and the U.S. Department of Energy (DOE) at the Oak Ridge National Laboratory (ORNL) continues, with two more supercomputers planned to join the two that already are deployed. The two existing systems at ORNL are Frontier, an all-AMD design, and Summit, an IBM/NVIDIA design. They will be joined in the next few years by Lux AI and Discovery —two supercomputers that represent a combined $1 billion investment of private and public funding. “This partnership exemplifies public-private collaboration at its best,” said AMD chair and CEO Lisa Su in a statement announcing the news in October. ### 6. AMD/OpenAI pact means new enterprise IT options October’s announcement that OpenAI and AMD have struck a deal could mean that AMD chips may become a viable enterprise IT option. That is good news considering the limits of Nvidia chip availability. Two companies said they would work together and that they have crafted “a 6 gigawatt agreement to power OpenAI’s next-generation AI infrastructure across multiple generations of AMD Instinct GPUs. The first one gigawatt deployment of AMD Instinct MI450 GPUs is set to begin in the second half of 2026.” That likely means anywhere from 3.5 million to 5 million chips, according to Moor Insights & Strategy. “AMD is now able to seed the market with a lot of its GPUs,” said Kimball of Moor Insights & Strategy. ### 7. AMD continues to take server share from Intel AMD continues to take market share from Intel, growing at a faster rate and closing the gap between the two companies to the narrowest it has ever been. In the first quarter of 2025, AMD’s share of the server marketplace rose to 27.2%, up sequentially from 25.7% and up year-over-year from 23.6% in Q1 of 2024, according to data from Mercury Research. Conversely, Intel continues to shed market share, falling to 72.8% in Q1 of 2025, down from 74.3% in Q4 of 2024 and down from 76.4% in Q1 of 2024. Dean McCarron, president of Mercury, said it’s mostly a case of AMD growing faster than Intel. “AMD’s growth rate in the quarter was multiples of Intel’s, resulting in significant server share gains,” he said in a research note. ### 8. DigitalOcean teams with AMD for low-cost GPU access Cloud infrastructure provider DigitalOcean Holdings in June announced a collaboration with AMD to provide DigitalOcean customers with low-cost access to AMD Instinct GPUs. DigitalOcean already offers access to the Instinct GPUs but in bare metal instances. It will now offer what it calls GPU Droplets, which are built on virtualized hardware. DigitalOcean offers a Kubernetes package that can be used with GPU Droplets should customers want to use a Kubernetes-based system, according to Bratin Saha, chief product and technology officer of DigitalOcean. ### 9. IBM, AMD team on quantum computing IBM and AMD announced in August that they’re working to blend Big Blue’s quantum computers with the chipmaker’s CPUs, GPUs and FPGAs to build intelligent, quantum-centric, high-performance computers. They plan to demonstrate how IBM quantum computers can work with AMD technologies to deploy hybrid quantum-classical workflows . ### 10. AMD acquires Brium to loosen Nvidia’s grip on AI software AMD acquired AI software startup Brium in a move aimed at challenging Nvidia’s dominance in AI software and strengthening support for machine-learning workloads on AMD hardware. “Brium brings advanced software capabilities that strengthen our ability to deliver highly optimized AI solutions across the entire stack,” AMD stated when the deal was announced in June. “Their work in compiler technology, model execution frameworks, and end-to-end AI inference optimization will play a key role in enhancing the efficiency and flexibility of our AI platform.” ### 11. AMD, Nvidia partner with Saudi startup to build multi-billion dollar AI service centers As part of the avalanche of business deals that came from President Trump’s Middle East tour in May, both AMD and Nvidia struck multi-billion dollar deals with Humain, an AI company that is chaired by Mohammed bin Salman, crown prince and prime minister of Saudi Arabia. The AMD deal did not discuss the number of chips involved in the deal, but it is valued at $10 billion. AMD and Humain plan to develop AI infrastructure through a network of AMD-based AI data centers that will extend from Saudi Arabia to the U.S. and support a range of AI workloads across corporate and government markets. AMD will provide its AI compute portfolio – Epyc, Instinct, and FPGA networking — and the AMD ROCm open software ecosystem, while Humain will manage the delivery of the hyperscale data center, sustainable power systems, and global fiber interconnects. ### 12. AMD steps up AI competition with Instinct MI350 chips, rack-scale platform In June AMD launched new accelerator chips and offered a glimpse into its AI infrastructure strategy, aiming to expand its role in the enterprise market, which Nvidia currently dominates. At its 2025 Advancing AI event, the chipmaker unveiled the AMD Instinct MI350 series accelerators and previewed a rack-scale AI infrastructure platform built on open industry standards. “The MI350 Series, consisting of both Instinct MI350X and MI355X GPUs and platforms, delivers a 4x, generation-on-generation AI compute increase and a 35x generational leap in inferencing, paving the way for transformative AI solutions across industries,” the company said in a statement. ### 13. AMD warns of new Meltdown/Spectre-like CPU bugs _J_ AMD issued an alert in July to users of a newly discovered form of side-channel attack similar to the infamous Meltdown and Spectre exploits that dominated the news in 2018. The potential exploits affect the full range of AMD processors – desktop, mobile and data center models — particularly 3rd and 4th generation Epyc server processors. ### 14. Next-generation HPE supercomputer offers a mix of Nvidia and AMD silicon HPE announced the Cray Supercomputing GX5000 ahead of the Supercomputing 25 conference in November. The platform will offer a choice of processors from Nvidia and AMD, even though the chips aren’t available yet and the system is likely not going to be available until 2027. ### 15. AMD outlines ambitious plan for AI-driven data centers AMD in November held its first financial analyst’s day, where CEO Lisa Su told analysts the company is seeing insatiable AI demand. Revenue growth could climb to 35% per year over the next three to five years because of that need, she said. In addition, Su expects to see AMD’s data center revenue increase 60% over the next three to five years, up from $16 billion in 2025. The total addressable AI market could top $1 trillion by 2030, doubling last year’s stated target of $500 billion by 2028. That number includes all silicon, from GPUs and CPUs to networking equipment.
www.networkworld.com
January 7, 2026 at 7:57 AM
At CES, Nvidia launches Vera Rubin platform for AI data centers
Nvidia used the Consumer Electronics Show (CES) as the backdrop for an enterprise-scale announcement: It launched the Vera Rubin NVL72 server rack platform for AI data centers, featuring new concepts and technology like “context memory” storage, zero downtime maintenance, rack-scale confidential computing, and several other advancements. As with previous naming conventions, the Vera Rubin platform is named after a notable scientist, in this case, astronomer Vera Rubin, who proved the theory of dark matter. Vera is the codename for an Arm-based high-performance CPU and Rubin is the name of the GPU platform. But one thing Nvidia CEO Jensen Huang stressed in his keynote is that the Vera Rubin platform is actually 6 pieces of silicon. In addition to the CPU and GPU, there are four networking processors: NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet Switch. Each Vera CPU comes with 88 custom cores, 176 threads using spatial multi-threading, 1.5 TB LPDDR5x system memory, and 1.2 Tbps memory bandwidth, alongside confidential computing and a 1.8 Tbps NVLink C2C interconnect. That’s seven times the bandwidth of PCI Gen 6, said Dion Harris, Nvidia’s senior director of HPC and AI infrastructure in a pre-briefing ahead of the show. “Vera doubles data processing, compression and code compilation performance versus our prior gen Grace CPU across both training and inference,” he said. “These Vera features maximize GPU utilization by orchestrating, routing, scheduling the KV cache and contacts.” The Vera CPU is the first to support confidential computing to deliver the Trusted Execution Environment, which helps maintain data security across CPU, GPU and the MV link domain, protecting the world’s largest proprietary models, training data and inference workloads. Rubin GPUs can deliver 50 petaflops for inference using NVFP4 data format—five times faster than Blackwell—and hit 35 petaflops for NVFP4 training, which is 3.5 times faster than Blackwell. HBM4 memory offers 22 Tbps bandwidth — 2.8x over Blackwell — and NVLink bandwidth per GPU is 3.6 Tbps, double Blackwell’s speed. Networking is enhanced with the liquid-cooled NVLink 6 Switch, offering 400G SerDes, 3.6 Tbps per-GPU bandwidth, total switching bandwidth of 28.8 Tbps, and 14.4 teraflops of FP8 in-network compute capability. The complete platform gives the Vera Rubin NVL72 platform up to 3.6 exaflops of NVFP4 inference, which is five times faster than the previous generation platform, and up to 2.5 exaflops of NVFP4 training, 3.5 times higher than the previous generation. Vera Rubin NVL72 includes 54 TB of LPDDR5x capacity (2.5x Blackwell), 20.7 TB HBM4 (50% more), 1.6 Pbps HBM4 bandwidth (2.8x increase), and a scale-up bandwidth of 260 Tbps (double that of Blackwell NVL72). “That’s more bandwidth than the entire global Internet,” said Harris. Nvidia also redesigned the rack, announcing its Third-Gen NVL72 Rack Resiliency. Features include a cable-free modular tray design that enables assembly and servicing 18 times faster than the previous generation. The NVLink Intelligent Resiliency feature supports server maintenance with “zero downtime,” keeping racks operational even during component swaps or partial population. The second-generation RAS Engine allows for GPU diagnostics without taking the rack offline. Initially, Rubin will be offered in two formats: the Vera Rubin NVL72 rack-scale platform. featuring 72 Rubin GPUs and 36 Vera CPUs, and the HGX Rubin NVL8 platform with eight Rubin GPUs for use with x86-based servers. Not surprisingly, Nvidia announce that a wide swath of leading tech firms plan to support Rubin, Including Amazon Web Services (AWS), Anthropic, Cisco, Cohere, CoreWeave, Dell Technologies, Google, HPE, Lenovo, Meta, Microsoft, Nebius, Nscale, OpenAI, Oracle Cloud Infrastructure (OCI), Perplexity, Supermicro, and xAI. To handle the massive data sets generated by agentic AI, Nvidia is introducing a new storage platform it said will offer a significant boost in inference performance and power efficiency, called the Nvidia Inference Context Memory Storage Platform. The platform uses BlueField-4 and Spectrum-X Ethernet tightly coupled with the Nvidia Dynamo and Nixl, enabling coordinated context retrieval across memory, storage and networking. “The result is a big step forward in performance and efficiency compared to traditional network storage used for inference context,” said Harris. This platform delivers up to 5x higher tokens per second, 5x better performance per TCO dollar and 5x better power efficiency than traditional storage systems. “That translates directly into higher throughput, lower latency and more predictable behavior. And it really matters for the workloads we’ve been talking about, large context applications like multiturn chat retrieval, augmented generation and a and agentic AI, multistep reasoning,” said Harris. Nvidia said it is collaborating with its storage partners to integrate this inference context memory into the Rubin platform, aiming to offer customers a comprehensive, unified AI infrastructure.
www.networkworld.com
January 7, 2026 at 7:57 AM
AWS hikes prices for EC2 Capacity Blocks amid soaring GPU demand
Amazon Web Services (AWS) has updated the pricing structure for some of its Elastic Compute Cloud (Amazon EC2) Capacity Blocks for ML offerings. Raised by approximately 15%, this move could affect enterprises planning large-scale machine learning workloads. The Amazon Capacity Blocks allow customers to reserve access to accelerated compute resources for a future start date. The company allows customers to reserve accelerated compute instances in cluster sizes of one to 64 instances (512 GPUs or 1024 Trainium chips) for up to six months for running a broad range of ML workloads. However, the EC2 Capacity Blocks can be reserved only up to eight weeks in advance. In June last year, AWS announced a price reduction of up to 45% for EC2 Nvidia GPU-accelerated instances across P4 and P5 instances. AWS did not immediately respond to a request for comment. ## Prices climb across P5 Capacity Blocks The Capacity Blocks include EC2 P6 instances that are accelerated by the latest Nvidia Blackwell GPUs, P5 instances complemented by Nvidia H100 and H200 Tensor Core GPUs, and P4 instances powered by Nvidia A100 Tensor Core GPUs. Pricing for the p5e.48xlarge instance, featuring eight Nvidia H200 accelerators for the US East (Ohio) region, has increased from the effective hourly rate per instance (per accelerator) of $34.608 to $39.799. For the p5en.48xlarge in the same region, the pricing has jumped from $36.184 to $41.612. This pricing remains the same across regions, including Stockholm, London, and Spain in Europe, and Jakarta, Mumbai, Tokyo, and Seoul in the Asia Pacific. However, customers in the US West (N. California) will now have to pay $49.749 instead of $43.26 for p5e.48xlarge and $52.015 instead of $45.23 for p5en.48xlarge. The P6e pricing remains the same at $761.904 for 72 B200 accelerators in the Dallas Local Zone for p4d.24xlarge. “The most defensible explanation is simply market-based pricing tied to supply and demand,” said Pareekh Jain, CEO at EIIRTrend & Pareekh Consulting. “As the demand for H100 and H200 GPUs outstrips supply, AWS is effectively applying a scarcity premium to guaranteed inventory. AWS is trying to recover higher infrastructure and capital costs from urgent capacity rather than overall capacity.” ## Guaranteed GPU capacity becomes the new battleground Guaranteed access to GPU clusters enables enterprises to de-risk AI infrastructure planning and build resilience against future supply volatility. Acknowledging the steep demand for high-end GPUs resulting in a shortage of Nvidia H100 and H200s, big clouds are increasingly offering guaranteed capacity to customers. Other than AWS, Google and Microsoft also have similar offerings, but presented in more traditional reservation models and scheduling frameworks. For instance, Google Cloud has introduced a calendar-based scheduling tool that lets customers reserve GPU capacity in fixed blocks ahead of time. “On paper, that looks a lot like what AWS is doing with Capacity Blocks. But the framing is different. Google is treating it as part of its broader resource scheduler, not a premium SKU. The guarantee is still there, but the pricing doesn’t feel as segmented or dynamic. It’s almost as if they’re using scheduling to compete, not price. And because they can also steer some workloads onto TPUs instead of GPUs, they’ve got a little more flexibility built into the system,” said Sanchit Vir Gogia, CEO and chief analyst at Greyhound Research. Azure, on the other hand, leans more on regional capacity reservations. Gogia added that these allow customers to hold specific VM types in specific zones, but they tend to favour long-term planning and big enterprise commitments. They get guaranteed capacity, but often pay for the privilege of holding those resources, whether they use them or not. It’s a different kind of premium, less about price per hour, more about time and commitment. ## Higher prices, limited immediate fallout Experts say these offerings represent a smaller percentage of total cloud spend compared to general compute. However, they represent a disproportionately high share of strategic AI spend. While AWS has been the first to come out to announce the price hike, Gogia believes the others might say it differently, but they’re working from the same playbook. Microsoft and Google also did not respond to the request for comment. Jain explained that for most enterprises, the price hike is unlikely to trigger immediate migration. The EC2 Capacity Blocks typically account for a small portion of total GPU spend, and strong data gravity, entrenched MLOps stacks, compliance controls, and skills continue to anchor workloads on AWS. As moving mature ML training stacks off AWS remains complex and time-consuming, the impact will be felt more in new workloads.
www.networkworld.com
January 7, 2026 at 7:57 AM
AMD launches on-prem AI chip, previews higher-end systems at CES
AMD showcased a range of new AI processors at CES 2026, including enterprise-focused GPUs designed for on-premises data centers and higher-end accelerators aimed at future large-scale AI systems. The announcements reflect AMD’s effort to expand its role in AI infrastructure by emphasizing open platforms, modular design, and broader deployment options. For enterprise customers, the key announcement was the Instinct MI440X GPU, which AMD said is designed specifically for on-premises AI deployments. The MI440X supports training, fine-tuning, and inference workloads in a compact eight-GPU configuration, designed to integrate seamlessly into existing data center infrastructure rather than requiring purpose-built AI clusters. ​The MI440X expands AMD’s MI400 series portfolio, which also includes the MI430X accelerators. At the other end of the scale, AMD previewed its “Helios” rack-scale platform, which it described as a “blueprint for yotta-scale infrastructure”. The company said Helios is designed to deliver up to three AI exaflops of performance in a single rack, with an emphasis on bandwidth and energy efficiency for training trillion-parameter models. “Helios is powered by AMD Instinct MI455X accelerators, AMD EPYC ‘Venice’ CPUs, and AMD Pensando ‘Vulcano’ NICs for scale-out networking, all unified through the open AMD ROCm software ecosystem,” the company added. Alexander Harrowell, principal analyst for advanced computing at Omdia, said AMD’s approach reflects a parallel development to Nvidia, which still serves the market with air-cooled GPUs and traditional servers via OEM partners, in addition to its rack-scale platforms. ## Enterprise buying implications For IT leaders deciding on their next AI investment, these developments suggest a shift in the market. Analysts note that while Nvidia remains the dominant player, buyer criteria are becoming more pragmatic. The focus is shifting beyond peak performance to include practical considerations such as reliable supply chains, predictable pricing, and easier integration into existing data center environments. “AMD is positioning itself as a reliable second source at a time when Nvidia faces supply constraints and very high prices,” said Pareekh Jain, CEO at Pareekh Consulting. “AMD chips are typically 20 to 30 percent cheaper, which matters for enterprise buyers. Enterprises are increasingly cautious about putting too much money into today’s AI hardware when depreciation cycles are getting shorter.” That caution is also shaping where enterprises deploy AI infrastructure, with on-premises environments emerging as a key focus for AMD’s latest offerings. “MI440X appears positioned as a time-to-value option for enterprises dealing with regulated data, data residency mandates and latency-sensitive inference, where keeping workloads on-prem is a business requirement rather than a technology choice,” said Rachita Rao, senior analyst at Everest Group. “That said, the chip’s dependence on HBM introduces constraints around latency and networking, which could limit performance consistency as deployments scale.” With MI440X, AMD is targeting on-prem enterprise deployments rather than hyperscalers, Jain said. He added that Nvidia has focused primarily on hyperscalers, while AMD is aiming at more price-sensitive on-prem enterprises that also face challenges securing Nvidia supply. “But Nvidia’s dominance only becomes meaningfully threatened if ROCm evolves into a true equivalent of CUDA with a low-friction migration path,” Rao added. “Until then, AMD will find it difficult to compete with the depth and momentum of the ecosystem Nvidia has built over the past two years.” ## Long-term AI roadmap Looking further ahead, AMD outlined elements of its longer-term AI roadmap. The company said the MI500 GPUs, set to launch in 2027, are on track to deliver up to a 1,000x increase in AI performance compared with the Instinct MI300X processors introduced in 2023, citing advances in its CDNA 6 architecture, a 2-nanometer manufacturing process, and the use of HBM4E memory. Analysts, however, cautioned that headline performance figures and manufacturing realities may diverge as the roadmap moves closer to production. “The 1000x number is versus the MI300, so there’s a substantial degree of cherry picking here,” Harrowell said. “The big issue is going to be sourcing HBM, which is taking over from CoWoS packaging capacity as the supply chain limiting factor.”
www.networkworld.com
January 6, 2026 at 11:30 AM
Arista rides AI wave, but battle for campus networks looms
It’s the early 2000s. Cisco has a 70% stranglehold on the Ethernet switch market. Competitors include 3Com, Brocade, Enterasys, Extreme Networks, Force 10, Foundry, and Juniper Networks. Into this crowded field comes Arista Networks, a self-funded startup led by former Cisco execs. As we approach 2026, Arista has surpassed Cisco in market share for high-speed data center switching. Quarterly revenue skyrocketed by 27.5% year-over-year, according to Arista’s latest earnings report, and it expects to exceed $10 billion in annual revenue in 2026 on a growth rate of around 20%. Meanwhile, none of those Cisco competitors have survived as independent companies, except for Extreme (which reported annual revenue of $1.1 billion in fiscal 2025, compared to Arista’s $8.75 billion). So, how is Arista doing it? What did it do right that so many of those other companies didn’t? First, it has strong, stable, focused leadership with three co-founders still at the helm, each with their own strengths. Chairperson and CEO Jayshree Ullal brings the business savvy and sets the culture. Chief Architect Andy Bechtolsheim is the technology visionary. President and CTO Ken Duda is the software guru and hands-on general. “Leadership sets a tone and mission for a company and is likely one of the reasons they have been successful. They haven’t strayed from their roots, and Arista looks to have the same focus that they had early on,” says IDC analyst Paul Nicholson. Second, Arista’s team has the technology chops to build high-bandwidth, low-latency switches that can scale to create massive data center networks. Arista pioneered the spine-leaf, two-tier architecture that is now standard in large data centers. And, over time, it developed a full networking stack of hardware, software, operating system, data lake, visibility, management, automation, AI agent, and security features, such as network detection and response (NDR). Third, and maybe most importantly, Arista has a well-honed business strategy. “All of the other vendors were trying to serve everybody,” says Forrester analyst Andre Kindness. They were competing against each other with bloated, feature-laden products aimed at the broadest possible range of customers. Unable to differentiate themselves, they struggled to gain market share and ultimately became acquisition targets. “What you saw with Arista is, ‘We’re not here to serve everyone, we’re here to develop products for particular markets,’” says Kindness. High-frequency stock traders wanted fast, slimmed down, easily programmable switches, and Arista was happy to fill that need. Then, as the cloud computing market grew, Arista got in on the ground floor, working closely with companies like Microsoft and Meta to develop the networking infrastructure for hyperscale data centers. “Arista has been laser focused in reaching the right customer segments with the right solutions. They started strong in the financial services sector with their ultra-low latency and high throughput switch solution. Winning business from the hyperscaler segment with their cloud networking solutions placed them more solidly on the map as these pivotal customers drive high volumes of repeatable business,” says Omdia analyst Adeline Phua. In fact, nearly half of Arista’s revenue comes from what the company refers to as the “cloud titans.” Also important to Arista’s success is what it didn’t do — which is chase markets outside of networking. Other networking vendors went on acquisition sprees, trying to build broad platforms that integrated security and networking. Arista didn’t chase SASE, SIEM or other trendy security buzzwords. It prefers to partner with security vendors like Palo Alto or Microsoft while remaining true to the networking mission. “Arista knows who they are, and they stick to their guns,” says Kindness. Todd Nightingale, newly hired as president and chief operating officer of Arista, told _Network World_ : “By focusing our investment on providing the absolute best networking technology possible, we don’t sacrifice our mission of quality. This is core to doing the right thing by our customers. We are comfortable with that tradeoff.” “Our strategy also provides our customers with choice,” Nightingale adds. “We never lock customers into a security solution that isn’t fit for their purpose. As security architectures change, we’re able to integrate and partner with the best of breed to provide the most secure outcome for our customers.” Now, Arista is looking for additional growth in two network specific markets that hold tremendous promise, but also some risks. They’re targeting the next big wave, which is connectivity for AI-powered data centers. And they’re making a big push into the enterprise campus and branch networking markets. Both initiatives could lead to something Arista has managed to skirt until now — a head-on collision with Cisco. ## Battle for the AI data center Remember when experts were predicting the demise of the data center? You’re not hearing that so much anymore. In fact, AI has rejuvenated the data center in both the cloud services and enterprise markets. The hyperscalers are building data centers to run AI workloads as fast as they can. And enterprises that are reluctant to put their data in the cloud are training AI models on premises. “We find ourselves amid an undeniable and explosive AI megatrend,” Ullal said on Arista’s November earnings call with analysts. “We are experiencing a golden era in networking with an increasing TAM [total addressable market] now of over $100 billion in forthcoming years.” “Our stated goal of $1.5 billion AI aggregate for 2025, comprising both backend and frontend, is well underway. We are experiencing momentum across cloud and AI titans, near cloud providers, and the campus enterprise. The demanding scale of AI buildouts is clearly unprecedented as we look to move data faster across multi-planar networks.” (Backend refers to AI networking within the data center. Frontend is traffic to and from the data center, driven by the ‘inference’ part of AI, the back and forth between prompts and queries from end users or customers, and the responses from the AI models.) Most analysts agree that Arista has a leg up when it comes to networking for AI based on its close working relationship with Microsoft, Meta, Oracle and others. “I definitely think Arista has a lead in that area and they are the closest to capturing a big percentage of that market,” Kindness says. Nicholson concurs: “Arista is positioned to continue benefiting from AI buildouts. The willingness to pivot to support their customers is an advantage — each design change they do for their customers can benefit others, so the support and innovation for and with customers is a win-win.” But don’t count Cisco out. Cisco CEO Chuck Robbins has conceded that his company whiffed on the cloud. “We didn’t participate in the infrastructure side of the cloud play, or the cloud evolution. There were a whole lot of things we didn’t capture,” he said at Cisco Live 2024. But Robbins is determined not to miss out on the AI wave. “When the cloud era hit, we perhaps were not as prepared as we should have been. I will tell you today, as this AI era begins, we are very, very prepared,” he said. Most recently, Cisco raised its guidance on AI infrastructure revenue in fiscal 2026 from $2 billion to $3 billion, primarily from hyperscalers but also from neocloud and enterprise customers. ## Battle for the enterprise Arista is making a concerted effort to expand its presence in the enterprise campus market. One major move along those lines was the recent acquisition of SD-WAN vendor VeloCloud from Broadcom. “VeloCloud brings a modern SD-WAN solution to Arista,” Nightingale says. “In large part, this completes our campus offering with wireless, switching (access and spine), routing, NAC and now SD-WAN. Our campus customers have been deploying Arista in the largest, most sophisticated sites including high-rise offices, universities, hospitals and manufacturing locations.” Ullal has set a target of $1.25 billion in campus networking revenue next year, a significant jump from the $750 million to $800 million expected in this fiscal year. “Ambitious goal, but we’re signed up to it,” Ullal said during her investor day remarks in September. Experts are not so sure that Arista can compete with Cisco on its home turf, the enterprise. Dell’Oro Group analyst Sian Morgan says that while Arista has increased campus switch sales above the overall market growth rate for the past two years, “it is very pertinent to ask whether Arista has the portfolio breadth to achieve their enterprise objective. A broader portfolio that includes SSE would definitely help, but it is only one piece of the puzzle.” Morgan adds that success in the enterprise means Arista would have to up its game with channel partners, which have the direct relationship with enterprise customers. “The real question is whether Arista can find the balance between addressing larger portions of the enterprise market by working with the channel, while keeping the company’s margins at an acceptable level.” Zeus Kerravala, principal analyst at ZK Research, has a similar opinion: “There should be no question about Arista’s ability to grow above market rates with the cloud providers and in AI builds. Enterprise is another question. The company has great products, but its channel strategy is still relatively immature.” That’s where former Cisco exec Nightingale comes in, and analysts give Arista high marks for recognizing the need to bring in fresh talent from the outside to drive the campus initiative. “Nightingale may really help Arista develop its channel strategy, something that has been holding back enterprise networking sales,” Morgan says. ## Opportunities and threats Moving forward, there are both opportunities and potential threats on the horizon for Arista. Here are some of the opportunities: * **Blue box** : White-box switches — inexpensive, generic, stripped-down boxes — are typically deployed by hyperscalers and large enterprises alongside commercial switches for specific use cases. Arista is now offering a “blue box” solution that adds a software layer of diagnostics, troubleshooting and management for white box switches. The target market is hyperscalers, neoclouds and the most tech-savvy enterprises. * **Global expansion**. Arista’s revenue is heavily concentrated in North America, so there is the potential for the company to grow its presence in Europe and Asia. * **The Ethernet wave:** Arista is working cooperatively with other industry players in groups like the Ethernet for Scale-Up Networking initiative and the Ultra Ethernet Consortium to help drive efforts to make Ethernet better for the needs of AI and high-performance computing. IDC’s Nicholson says: “The growth of Ethernet, with the supporting UEC standards, is a tailwind for Ethernet switch vendors as Ethernet becomes the preferred choice over Infiniband for AI factories.” Then there are the potential threats: * **Too many eggs in one basket.** It can be an advantage to have a small number of large customers that provide a stable, recurring revenue stream. But there’s also a risk. Losing even one of those big customers could be disastrous. * **Platformization:** The overarching industry trend is the convergence of security and networking, as well as the desire by enterprise execs to settle on a platform rather than cobble together point products. * **Maintaining growth rate:** Arista is putting up growth numbers that might seem difficult to sustain over the long term. The company grew revenue 27% year-over-year in its latest quarter, but only 5% sequentially. And fourth-quarter guidance is for revenue growth in the 2% range sequentially. During a recent call, Ullal assured analysts that the apparent slowdown is a minor blip, and that Arista is sticking to its guidance of a growth rate of 20% overall, driven by increases in the 60% range for both AI networks and campus network in 2026. * **Competition:** There are some interesting X factors in today’s networking world. GPU-maker Nvidia has emerged as the market leader and a formidable force in backend data center switching for AI workloads. Broadcom is another chip vendor moving up the stack with data center switches. And don’t count out HPE, which just completed its Juniper acquisition. * **Boom or bust** : Finally, Arista and all of the other switch vendors are counting on the data center boom continuing. But there’s always the possibility that the AI bubble will burst.
www.networkworld.com
January 6, 2026 at 11:30 AM
SOCAMM memory gains ground as AI data centers proliferate
Samsung last month unveiled a SOCAMM2 LPDDR5-based memory module designed specifically for AI data center platforms. SOCAMM2 is a new memory form factor that comes with a performance boost. CAMM (Compression Attached Memory Module) was originally developed by Dell as a type of memory technology for laptops, and it has since been turned over to a standards body to promote industry adoption. The CAMM2 version is the first generation developed as an industry standard. SOCAMM2 is based on LPDDR5 memory, a high-performance, high-bandwidth memory design used in smartphones and tablets. It gives DDR memory performance with less power consumption. Samsung notes that SOCAMM2 has twice the bandwidth of standard DDR5 RDIMMs used in servers while consuming less power. Other estimates say that SOCAMM2 has 1.5x to 2.0x the performance of standard DDR5 memory while consuming 55% of the power. The SOCAMM2 modules are smaller than a standard DDR5 memory stick because the memory chips are extremely dense, using stacking technology to put layers of memory on one chip. As a result, it takes up much less space on the motherboard than an equal amount of DRAM. SOCAMM2 can be used in conjunction with DDR memory or by itself as system memory. Dell brought in partners to help co-design the CAMM memory spec and then turned it over to the JEDEC standards body, so SOCAMM2 is viewed as an industry standard and not one vendor’s solution. JEDEC has in turn added features to the CAMM spec, including ECC and other error-correction features aimed at enterprise users. SOCAMM is not a repackaging of existing hardware or a solution in search of a problem. Rather, it’s addressing real problems and is needed and wanted by the industry, said Jim Handy, president of Objective Analysis. “The server processor manufacturers and Nvidia are really behind SOCAMM because of the fact that they [gain] a faster interface and it gets a lot of memory into a small area with a little bit lower power consumption,” he said. As previously noted, SOCAMM2 uses stacked memory, which can be more expensive to manufacture, but Handy says that’s not the case. “[Memory vendors] sell different stack configurations for the same price as these DRAM guys, and they are using the same packaging technology that the NAND flash guys do, so I wouldn’t expect there to be any noticeable price difference,” he said. SK Hynix, another major memory manufacturer on the market, has also announced plans to support SOCAMM2 memory but has not given any release details and is believed to be behind Micron and Samsung. SOCAMM2 is expected to launch around the second quarter of 2026 when Nvidia launches its Vera Rubin platform.
www.networkworld.com
January 6, 2026 at 11:30 AM
Wi-Fi 8 in 2026: Next-gen wireless standard prioritizes reliability over speed gains
The Wi-Fi 7 wireless standard gained momentum in 2025, but Wi-Fi 8 is on its heels. To be sure, 2025 was a big year for Wi-Fi 7 adoption. The standard hit its final draft in September 2024, and the formal IEEE 802.11be standard publication followed in July 2025. Since then, adoption has been robust across all sectors. “By the end of 2025, an estimated 583 million Wi-Fi 7 devices are expected to ship globally,” Jeff Platon, vice president of marketing at Wi-Fi Alliance, told _Network World_ , citing data from IDC Research. ## Enterprise Wi-Fi 7 adoption takes off After a slow start, enterprises are now embracing Wi-Fi 7 at a faster pace than previous generations. Wi-Fi 7 access point shipments jumped from 26.3 million in 2024 to a projected 66.5 million in 2025, according to data shared by Tiago Rodrigues, CEO at the Wireless Broadband Alliance. Looking ahead, ABI Research forecasts this trend will accelerate further with 117.9 million Wi-Fi 7 AP shipments expected in 2026. Chris Szymanski, director of product marketing for wireless broadband communications at Broadcom, explained that the initial hesitation for enterprise adoption of Wi-Fi 7 was understandable. “Wi-Fi 7 came only a short time after the launch of Wi-Fi 6E, so the enterprise market had to adjust to a shorter period between equipment releases, and it was hungry to adopt Wi-Fi 6E,” Szymanski said. “This led to a little slower adoption in 2024. Now enterprises are adopting Wi-Fi 7 quickly.” The numbers support this acceleration. The Wi-Fi Alliance forecasts 1.1 billion total Wi-Fi 7 device shipments in 2026, including 196.1 million IoT devices, 22.3 million healthcare devices and 159.4 million consumer devices. Large public venues and educational institutions are leading the charge. According to Platon, these sectors see Wi-Fi 7 as the solution to spectrum congestion challenges and as an enabler of new use cases. ## Wi-Fi 8 arrives ahead of schedule ** **The big story for 2026, however, is likely to be Wi-Fi 8. In a development that breaks from typical wireless generation timelines, consumer products could arrive much sooner than anticipated. “Broadcom launched a full ecosystem of Wi-Fi 8 products in October 2025,” Szymanski revealed. “We expect the retail market to act quickly on this product availability, and the market could see Wi-Fi 8 products as early as Summer 2026.” This timeline represents a significant acceleration. The IEEE 802.11bn Task Group formed in May 2021 with a target standard approval date of September 2028. Broadcom’s ecosystem launch puts Wi-Fi 8 retail products on track to arrive before the standard is finalized. The gap between Wi-Fi 7’s launch and potential Wi-Fi 8 product launches in mid-2026 could potentially be shorter than the typical cycle between Wi-Fi generations. While consumer adoption might come early, the enterprise and operator markets will follow a more traditional adoption pattern. “It’s likely that such products will not launch until mid to late 2027,” Szymanski noted. These sectors typically move more deliberately due to longer refresh cycles and procurement processes. ## What’s driving Wi-Fi 8 development? The IEEE 802.11bn standard represents a shift in Wi-Fi evolution. Rather than primarily chasing higher peak speeds, Wi-Fi 8 focuses on improving real-world performance in dense, interference-prone environments. At its core, Wi-Fi 8 prioritizes consistent performance under challenging conditions. The standard maintains the same theoretical maximum data rate as Wi-Fi 7 while targeting improvements in effective throughput, lower latency for time-sensitive applications and reduction in packet loss. According to Platon, Wi-Fi Alliance priorities for the next generation include ultra-high reliability, bounded latency, faster speeds and reduced power consumption. “Interest in the next generation of Wi-Fi is already building, underscoring the critical role Wi-Fi plays in global connectivity,” Platon said. ## Wi-Fi offload gains traction Beyond the transition between wireless generations, another trend is gaining momentum: Wi-Fi offload. Mobile carriers face mounting pressure from increasing traffic on cellular networks while trying to improve customer connectivity experiences. “A plethora of trends will act to spur investments into Wi-Fi offloading in 2026,” Rodrigues said. Smart cities are turning to Wi-Fi offload to provide residents and tourists with continuous free connectivity. The technology enables applications ranging from smart traffic management to disaster prevention systems. Advancements in OpenRoaming technology are expected to accelerate this trend in 2026, making seamless roaming between networks more practical and widespread. “For mobile carriers, the challenge of grappling with ever-increasing traffic on their cellular networks, alongside the need to improve connectivity experiences for their customers, will drive them to expand their Wi-Fi offloading capabilities,” Rodrigues said.
www.networkworld.com
January 6, 2026 at 11:32 AM
Nvidia licenses Groq’s inferencing chip tech and hires its leaders
Nvidia has licensed intellectual property from inferencing chip designer Groq, and hired away some of its senior executives, but stopped short of an outright acquisition. “We’ve taken a non-exclusive license to Groq’s IP and have hired engineering talent from Groq’s team to join us in our mission to provide world-leading accelerated computing technology,” an Nvidia spokesman said Tuesday, via email. But, he said, “We haven’t acquired Groq.” Groq designs and sells chips optimized for AI inferencing. These chips, which Groq calls language processing units (LPUs), are lower-powered, lower-priced devices than the GPUs Nvidia designs and sells, which these days are primarily used for training AI models. As the AI market matures, and usage shifts from the creation of AI tools to their use, demand for devices optimized for inferencing is likely to grow. The company also rents out its chips, operating an inferencing-as-a-service business called GroqCloud. Groq itself announced the deal and the executive moves on Dec. 24, saying “it has entered into a non-exclusive licensing agreement with Nvidia for Groq’s inference technology” and that, as part fo the agreement, “Jonathan Ross, Groq’s Founder, Sunny Madra, Groq’s President, and other members of the Groq team will join Nvidia to help advance and scale the licensed technology.” The deal could be worth as much as $20 billion, TechCrunch reported. ## A way out of the memory squeeze? There’s tension throughout the supply chain for chips used for AI applications, leading to Nvidia’s CFO reporting in its last earnings call that some of its chips are “sold out” or “fully utilized.” One of the factors contributing to this identified by analysts is a shortage of high-bandwidth memory. Finding ways to make their AI operations less dependent on scarce memory chips is becoming a key objective for AI vendors and enterprise buyers alike. A significant difference between Groq’s chip designs and Nvidia’s is the type of memory each uses. Nvidia’s fastest chips are designed to work with high-bandwidth memory, the price of which – like that of other fast memory technologies — is soaring due to limited production capacity and rising demand in AI-related applications. Groq, meanwhile, integrates static RAM into its chip designs. It says SRAM is faster and less power-hungry than the dynamic RAM used by competing chip technologies — and another advantage is that it’s not (yet) as scarce as the high-bandwidth memory or DDR5 DRAM used elsewhere. Licensing Groq’s technology opens the way for Nvidia to diversify its memory sourcing. ## Not an acquisition By structuring its relationship with Groq as an IP licensing deal, and hiring the engineers it is most interested in rather than buying their employer, Nvidia avoids taking on the GroqCloud service business just as it is reportedly stepping back from its own service business, DGX cloud, and restructuring it as an internal engineering service. It could also escape much of the antitrust scrutiny that would have accompanied a full-on acquisition. Nvidia did not respond to questions about the names and roles of the former Groq executives it has hired. However, Groq’s founder, Jonathan Ross, reports on his LinkedIn profile that he is now chief software architect at Nvidia, while that of Groq’s former president, Sunny Madra, says he is now Nvidia’s VP of hardware. What’s left of Groq will be run by Simon Edwards, formerly CFO at sales automation software vendor Conga. He joined Groq as CFO just three months ago.
www.networkworld.com
December 31, 2025 at 11:29 AM
SoftBank expands AI infrastructure ambitions with $4B DigitalBridge acquisition
SoftBank Group has announced plans to acquire DigitalBridge Group, an asset manager specializing in digital infrastructure investments such as data centers, cell towers, and fiber networks, in a deal valued at $4 billion. The acquisition underscores SoftBank’s aggressive strategy to dominate the next wave of AI infrastructure globally. SoftBank said the acquisition is part of its mission to realize Artificial Super Intelligence (ASI), a vision that goes beyond current AI capabilities. “Achieving that vision requires breakthroughs not only in AI models, but also in the platform infrastructure needed to train, deploy, and serve them at a global scale. The planned acquisition of DigitalBridge will strengthen SoftBank Group’s ability to build, scale, and finance the foundational infrastructure needed for next-generation AI services and applications,” SoftBank said in a statement. ## SoftBank’s endless appetite for AI The deal adds another piece to SoftBank’s growing AI puzzle. The Japanese conglomerate has been pouring billions into AI-related ventures, betting that the future of computing will be defined by massive data processing and connectivity. Its most ambitious initiative so far is Project Stargate, a $500-billion AI infrastructure project aimed at creating hyperscale data centers and advanced compute platforms. SoftBank has already committed $342 billion to the project over the next four years, despite facing delays in execution. Earlier this year, SoftBank acquired Ampere Computing for $6.5 billion, strengthening its position in the Arm-based server processor market — a critical component for AI workloads. SoftBank already holds a significant stake in Arm Holdings, the chip designer powering much of today’s mobile and AI computing. In 2024, SoftBank added Graphcore, an AI chip designer, to its portfolio and continues to invest in leading AI players such as OpenAI and Nvidia. ## AI-driven growth in data centers The surge in AI and machine learning workloads has triggered an unprecedented boom in global data center investments. Hyperscalers like AWS, Google, Microsoft, and Oracle are committing hundreds of billions to expand their infrastructure to meet soaring compute demands. In a landmark deal, Oracle signed a $300-billion contract in September to support OpenAI’s cloud computing requirements, marking the largest-ever cloud agreement. According to UBS, global AI spending is projected to reach $375 billion in 2025, grow to $500 billion in 2026, and exceed $3 trillion by 2030. This spending encompasses AI data centers, power generation, and resource allocation — areas where SoftBank aims to capture significant market share. For SoftBank, DigitalBridge brings expertise in managing and scaling digital infrastructure assets, including data centers and fiber networks — critical components for AI-driven computing. By integrating DigitalBridge’s capabilities, SoftBank says it can accelerate its infrastructure build-out, ensuring it remains at the forefront of the AI revolution.
www.networkworld.com
December 30, 2025 at 11:29 AM
Top Intel stories of 2025
It’s been a roller coaster of a year for Intel, marked by the arrival of a new chief, new Xeon processors, select product divestitures, and sizable layoffs. Under the leadership of CEO Lip-Bu Tan, who took the helm in March, Intel delivered the first Xeon built on its next-generation 18A process node, navigated the divestiture of its FPGA business, and upped its efforts to restructure the company and regain market leadership amid strengthening competition. It also inked a $5 billion partnership with Nvidia and saw the U.S. government take a 10% stake in the company. Here are our top stories about Intel in 2025, organized in five sections: * CEO Tan’s arrival and early moves * Vendor partnerships and equity investors * Product divestitures and talent loss * Significant Xeon news * Industry analysis of Intel’s product roadmap and market outlook ## Lip-Bu Tan’s leadership #### Intel CEO: We are not in the top 10 semiconductor companies In July, Intel’s CEO Lip-Bu Tan told employees that Intel is not among the leading chip companies, a stark contrast to the perpetual sunny, cheerful optimism of his predecessor Pat Gelsinger. #### Intel under Tan: What enterprise IT buyers need to know Intel’s appointment in March of semiconductor veteran Lip-Bu Tan as CEO marked a critical moment for the company and its enterprise customers. With rising competition from AMD, Arm-based chips, and RISC-V alternatives, Intel faces mounting pressure to defend its x86 dominance. #### New Intel CEO Lip-Bu Tan begins to lay out technology roadmap After being on the job for two weeks, Intel CEO Lip-Bu Tan gave a major speech at a partner conference where he said the ailing company needs to get its act together, and he doesn’t want customers to hold back on their criticism of Intel. #### Intel taps semiconductor veteran Lip-Bu Tan as new CEO In March, Intel announced its new CEO is Lip-Bu Tan, a semiconductor industry veteran. Tan previously was CEO of Cadence Design Systems, which makes software products used to design and verify complex electronic systems. Cadence’s toolset is used by Intel and all of the other major chip designers. ## Customer wins and investments #### IBM Cloud speeds AI workloads with Intel Gaudi 3 accelerators IBM announced in April that it would make Intel Gaudi 3 AI accelerators available on IBM Cloud to help customers scale gen AI workloads and optimize performance for AI inferencing and fine-tuning. #### Intel will design CPUs with Nvidia NVLink in return for $5 billion investment Intel is collaborating with Nvidia to design CPUs with Nvidia’s NVLink high-speed chip interconnect, it said in September — just months after committing to co-develop a competing interconnect, UALink, with AMD, Broadcom, and other tech companies. Nvidia will invest $5 billion in Intel stock as part of the agreement. #### Intel saga continues: Federal bailout considered Rumors emerged in August of the federal government’s plans to purchase a 10% stake in Intel in a bid to speed up completion of its advanced fabrication facilities. The deal came together after a meeting between President Donald Trump and CEO Lip-Bu Tan, which took place after President Trump called for the resignation of Tan. #### Who wins/loses with the Intel-Nvidia union? Nvidia dipped into its $56 billion bank account to acquire a 5% stake in Intel for $5 billion, making it the second largest shareholder of Intel stock after the federal government’s recent investment. The deal, announced in September, provides Nvidia greater access to the x86 ecosystem, important for the enterprise data center market, and provides Intel with access to GPUs that have demand and can move their CPU products as well. ## Market exits, lost talent #### Intel sells off majority stake in its FPGA business Intel spun off its programmable solutions group as a standalone FPGA company, selling a majority stake in the company to a private equity firm in April. Intel is taking a fairly hefty loss on this deal. It acquired Altera in 2015 for $16.7 billion, but the deal with Silver Lake technology investments values Altera at $8.75 billion total and Intel is getting $4.4 billion for the sale. #### Intel eyes exit from NEX unit as focus shifts to core chip business Intel in May said it might sell its Network and Edge (NEX) business, marking the latest step in a broader effort to reshape the company under new CEO Lip-Bu Tan. #### Qualcomm hires Intel’s Xeon designer for data center The chief architect for Intel’s Xeon server processors defected to chip rival Qualcomm. Sailesh Kottapalli, a 28-year Intel veteran and chief architect for the company’s Xeon processors, announced in January that he joined Qualcomm as a senior vice president. #### Intel decides to keep networking business after all Intel CEO Lip-Bu Tan has been on a mission to divest the company of all non-core technologies, and its Network and Edge Group, commonly referred to as NEX, was the largest group scheduled for divestiture. But after a few months of shopping it around, either Intel couldn’t get the deal it wanted or changed its mind. In December, Intel said it has cancelled plans to divest itself of NEX. ## Significant tech advances #### Intel touts efficiency and performance in new 288-core Xeon processor The Hot Chips conference in August was the backdrop for Intel’s newest Xeon processor, codenamed Clearwater Forest. It’s the first Xeon built on the company’s next-generation 18A process node. #### Intel targets edge, high-performance computing with extended Xeon 6 chip line Intel expanded its Xeon 6 line of processors, adding models 6700/6500 for high-performance cores and edge computing devices to the family. Announced in February, the new Xeon 6 processors feature P-cores (performance-cores) for high-performance computing. #### Intel details new efficient Xeon processor line In October, Intel revealed more details of its next-generation Xeon 6+ E-core CPU family, codenamed “Clearwater Forest,” with a focus on power and performance efficiency. Intel has split its Xeon product line into two categories: the P line of high-performance cores and the E line for efficiency for tasks and processes that are less performance sensitive. #### New Intel Xeon 6 CPUs unveiled; one powers rival Nvidia’s DGX B300 Intel in May unveiled three additions to its Intel Xeon 6 series of CPUs that are designed to manage GPU-powered systems. One, the Xeon 6776P, is currently serving as the host CPU for Nvidia’s DGX B300, its latest generation of AI-accelerated systems. One industry analyst views the launch as confirmation that the chip maker is “looking to regroup and pick fights it thinks it can win.” ## Industry analysis #### Despite the hubbub, Intel is holding onto server market share Intel is holding on to market share in both client and server markets against AMD despite the seemingly endless stream of bad news surrounding the company. Second-quarter 2025 chip sales were roughly flat for both companies with very little share trading hands, according to August data from Mercury Research. #### Can Intel cut its way to profit with factory layoffs? Intel plans to lay off up to 20% of its manufacturing sector employees starting in July, according to a media report published in June. The rumored cuts suggest the company is running out of options as it seeks a return to profitability and navigates faltering demand for its core products. #### What Intel needs to do to get its mojo back The once undisputed king of the chip market, Intel is on its knees. At the beginning of 2025, we looked at what it needs to do to get back on its feet. #### Intel sees supply shortage, will prioritize data center technology In October, Intel reported better than expected third-quarter financial figures, reflecting a turnaround in sales, reduced operational expenses, and a renewed focus on its core business. On the downside, Intel is enjoying a little too much success and is seeing a shortage of supply.
www.networkworld.com
December 29, 2025 at 11:29 AM
Top 10 Nvidia stories of 2025 – From data center to AI factory
2025 was not just about faster GPUs. It was about the**** fundamental re-architecture of the enterprise data center from a storage/retrieval hub to a manufacturing plant for intelligence – or **the AI factory.** Nvidia CEO Jensen Huang set the stage in May at Dell Technologies World 2025: _“AI is here — this is unquestionably the single biggest platform shift… This is a once in a lifetime opportunity — in the last 60 years, this is the biggest reinvention that you and I have seen.”_ Here’s a look back at the top Nvidia storeis of 2025. ## **1. The rise of the AI factory** Nvidia CEO Jensen Huang popularized the concept of the AI factory — or data centers dedicated not just to moving and storing data but to “manufacturing intelligence.” This became the blueprint for modern enterprise architecture in 2025. AI factories extend models like AWS Outposts to on-premises environments, designed to help enterprises to run agentic applications without data residency issues. ## **2. The Blackwell B200 and Rubin roadmap** This was the year of the **Blackwell** GPU architecture ramping up to full production, followed next-generation **Rubin architecture**. These chips defined the “trillion-parameter” era of large AI workloads. ## **3. Nvidia and Cisco: The enterprise AI partnership** Nvidia and Cisco strengthened their partnership to bring AI infrastructure to the enterprise. This included the launch of Cisco Nexus HyperFabric with Nvidia AI, designed to make deploying AI clusters easier for standard IT teams. Cisco’s Nexus integration with Nvidia’s active intelligence is designed to simplify the infiniBand vs. Ethernet complexity for traditional network engineers. ## **4. Spectrum-X and a win for Ethernet** Nvidia pushed its **Spectrum-X** Ethernet platform hard in 2025, acknowledging that while InfiniBand is for supercomputers, **Ethernet is for the mass enterprise market**. Spectrum-X brings remote direct memory access (RDMA) and low-latency capabilities to standard Ethernet, making it viable for AI workloads in standard enterprise data center networking. ## **5. US approves H200 chips for China** In late 2025, the US government approved the sale of advanced **H200** chips to vetted customers in China (with a 25% tariff/fee). This marked a major shift in US-China tech policy and supply chain dynamics. ## **6. Sovereign AI clouds for national security** Nations and regions (Japan, France, UK, Middle East) began building their own sovereign AI clouds using Nvidia hardware to ensure data privacy and national security, reducing reliance on US hyperscalers. ## **7. Nvidia Enters the 6G and telecommunications sector** Nvidia launched the AI-RAN (radio access network) alliance, pushing to run native 5G and 6G networks on GPUs rather than custom telecom hardware. This is designed to merge the telecom edge with the AI edge. ## **8. BlueField-3 DPUs becomes standard, BlueField 4 introduced** The data processing unit (DPU) moved from a niche component to a standard part of the AI factory stack, offloading security and networking tasks from the CPU/GPU. BlueField-3 is designed to enable zero trust security inside the data center and accelerates east-west traffic for AI clusters. Nvidia also introduced BlueField 4, a next-generation processor that acts as the operating system for AI factories. It delivers 800Gbit/sec of throughput and six times more compute than BlueField 3. ## **9. Nvidia NIMs on the move** Nvidia moved up the stack with **NIMs (Nvidia Inference Microservices), pre-packaged “containers” of AI models** that allow developers to deploy generative AI anywhere (cloud, on-prem, laptop) instantly. Nvida’s goal is to help IT teams can use NIMs to rapidly deploy “enterprise-ready” AI applications without needing deep machine learning expertise. ## **10. The $4 trillion market cap and industrial revolution narrative** Nvidia briefly became the world’s most valuable company (again) in 2025, driven by the narrative that AI is a “new industrial revolution” and Nvidia is the power plant.
www.networkworld.com
December 28, 2025 at 3:08 PM
Google agrees to acquire infrastructure builder Intersect to accelerate capacity development
Google parent Alphabet is taking steps to enable speedier addition of capacity to feed AI’s increasing demands with its announcement of plans to buy data center and energy company Intersect. This, it said, will meet intensive demand, increase energy reliability, reduce power delays, and support development of alternative energy sources. “AI infrastructure across the board appears to be at capacity, and there are questions whether upcoming investments in data centers will come to fruition on time,” said Thomas Randall, research lead at Info-Tech Research Group. “Alphabet acquiring Intersect quickly opens capacity that it expects [to require, to meet demand] from Gemini’s growing popularity, training, and embeddedness in nearly all Google searches.” ## Expands efforts to build power capacity With the $4.75 billion deal, expected to close in the first half of 2026, Google will absorb Intersect’s team and “multiple gigawatts” of energy and data projects already in development, including its first co-located data center and power site under construction in Texas. The companies will jointly continue work on those projects and build out new ones, according to Google. However, Intersect’s existing Texas assets and those in development in California are not part of the deal, and will continue as an independent entity. Intersect will “explore a range of emerging technologies” to help “increase and diversify” energy supply and bolster the tech giant’s data center investments, according to the announcement. The tech giant said it is committed to working with energy and utility companies to “unlock abundant, reliable, affordable energy supply” that supports data center buildouts. The acquisition is yet another step in Google’s efforts to build this capacity. Earlier this year, it announced a partnership with NV Energy that will bring 115MW of clean energy, via geothermal power, to Nevada’s grid. The company is also working with Energy Dome on CO2 battery innovations for long-duration energy storage, and is supporting carbon capture and storage (CCS) technologies at a gas power plant in partnership with Broadwing Energy. Through that initiative, Broadwing will capture and permanently store roughly 90% of its CO2 emissions, and Google has committed to buying most of the power it generates. With the Intersect purchase, Google is effectively signaling that the traditional model of relying on utilities and third-party energy developers isn’t dependable enough in the AI era, noted Sanchit Vir Gogia, chief analyst at Greyhound Research. It is a “recognition that Google’s constraint has moved upstream,” he said. Google doesn’t need to be schooled on data center design or real estate footprint, it “needs a way to bring megawatts online with predictability in a market where grid timelines, interconnection queues, substation upgrades, and permitting cycles are now slower than the compute deployment cycle.” It is also a risk internalization play, he noted; Alphabet wants to reduce its exposure to delays that occur when phased power delivery is late and “[leaves] capacity stranded and utilization depressed.” ## ‘Shifts the dependency’ Intersect adds time certainty, sequencing control, and a developer-style operating model that is not reliably provided by utilities and co-location contracts, Gogia noted. The company is “explicitly framed” around co-locating demand and dedicated gas and renewable generation. That model shifts the dependency, he noted. Instead of waiting for grid capacity to become available, then placing load into it, generation is placed alongside the load path and both are orchestrated together. “It is a very different approach to reliability and speed,” said Gogia. Alphabet’s support of numerous renewable and clean energy technologies indicates that the tech giant is looking to diversify and stabilize power capacity across different regions and grid conditions and reduce single point dependency. When one technology pathway stalls, another can carry part of the load, Gogia explained. “Intersect allows Google to coordinate the sequencing so that compute and power arrive together,” he said. ## Tie energy strategy to capacity planning Ultimately, the acquisition reduces Alphabet’s dependency on third-party energy partners, noted Info-Tech’s Randall. Energy is a fundamental component of the core infrastructure stack, but it is becoming more and more scarce as AI providers scoop up resources. “Data center managers should use this moment as an opportunity to tie energy strategy with capacity planning, sustainability goals, and competitive positioning,” Randall advised. Traditionally, Gogia added, CIO decision-making around build-versus-lease-versus-cloud was framed around cost, agility, security, and compliance. But the missing variable was certainty around power delivery. If hyperscalers are investing billions to bring generation and load together, enterprises should assume they will face the same constraints, he said, “just earlier and with less negotiating power.” Cloud abstracts energy risk, Gogia noted. When regions hit power and GPU ceilings, capacity gets rationed, timelines shift, and customers are nudged to alternate regions or delivery models. This can result in delayed deployments, and, often, higher costs because scarcity pricing differs. This reality is even more evident with on-premises builds, he observed; builders can complete a facility on time, yet still run under capacity for months or longer if power does not arrive as planned. CIOs must adjust their governance model, he advised, noting that energy due diligence should be part of the technology decision process. Site selection requires a “time to power” view, not just a network latency view. Contracts should provide greater transparency around capacity commitments, region expansion goals, and contingency plans. “Data center planning [duration] may get shorter because designs are modular and repeatable,” said Gogia. “Energy contracting will get longer because supply is constrained and approvals are slow.” ## Power shouldn’t be an afterthought Enterprises must develop a power risk strategy and be mindful of social license, he noted. Data center expansion is increasingly contested by communities and regulators, especially when it comes to its impact on the local grid. With its current move, Alphabet is scaling its energy supply while managing perceptions that local customers will shoulder the costs. “Enterprises should learn from that,” said Gogia. “If you are planning a major facility or even a large colocation footprint, stakeholder management is no longer optional. It is part of delivery.” Google’s acquisition of Intersect also signals a shift in vendor strategies. That could reshape pricing, availability, and negotiation dynamics for enterprise buyers, said Gogia. Power shouldn’t be treated as an afterthought, he advised; that will lead to slipped timelines. Assume that some capacity will be constrained and plan for alternatives far before projects are underway. “This is the decade where kilowatts, permits, and politics quietly decide whether your ‘cloud first’ roadmap actually lands on time,” Gogia noted.
www.networkworld.com
December 23, 2025 at 1:29 PM