Lightnews — Scholar-powered news

FuriosaAI @furiosa.ai · 1d

Read the full announcement here: www.bytebt.com/bytebridge-s...

FuriosaAI @furiosa.ai · 1d

"Furiosa is committed to delivering cutting-edge processors that power the next wave of AI workloads. By working with ByteBridge, we can ensure that these innovations are deployed at scale, supporting resilient digital infrastructure across the region." – Alex Liu, SVP of Product and Business

1

FuriosaAI @furiosa.ai · 1d

We’re pleased to announce a new partnership between FuriosaAI and ByteBridge to advance next-generation AI infrastructure across the APAC region.

1

FuriosaAI @furiosa.ai · 6d

🤝 Visit us at KLCC Hall 6, Booth 6037. We’d love to connect with you.

FuriosaAI @furiosa.ai · 6d

🛬 We’re headed to #SDEC2025 in Kuala Lumpur next week to explore the future of AI and semiconductors.

👀 Don’t miss our co-founder and CEO, June Paik, onstage October 10 as he shares how Furiosa is building next-gen AI chips for a more sustainable future.

1

FuriosaAI @furiosa.ai · 7d

Under the hood:
🔹 Dual CPUs + optimized PCIe topology tuned for RNGD
🔹 Enterprise-grade reliability with dual management paths + secure boot
🔹 Cloud-native from day one with pre-installed SDK and vLLM-compatible serving interface

👉 Learn more: furiosa.ai/blog/introdu...

FuriosaAI @furiosa.ai · 7d

The result: dramatically lower power costs, higher deployment density, and the ability to build AI Factories with faster token throughput.

Ready to serve LLMs from day one — no integration delays, no specialized power or cooling required.

1

FuriosaAI @furiosa.ai · 7d

Introducing Furiosa NXT RNGD Server 🚀

Engineered for efficient AI inference, the NXT RNGD Server hosts up to 8 RNGD accelerators, delivering:
⚡ 4 PFLOPS (FP8) compute
📦 384GB HBM + 12TB/s bandwidth
🔋 Just 3kW power draw (compared to 10.2kW for H100 SXM servers)

1

FuriosaAI @furiosa.ai · 21d

Last week, we demoed OpenAI’s open-weight gpt-oss 120B model running live on RNGD, our flagship AI accelerator.

Here’s a recording of the demo in action, showing that cutting-edge models can be deployed well within the existing power budgets of typical data centers:

FuriosaAI @furiosa.ai · 27d

This collaboration with OpenAI highlights the essential role that AI-native hardware like RNGD plays in building more economically and environmentally sustainable AI for enterprises around the world.

FuriosaAI @furiosa.ai · 27d

The combination of RNGD's performance and power efficiency with advanced new models like gpt-oss will play a critical role in making advanced AI more sustainable and accessible.

1

FuriosaAI @furiosa.ai · 27d

This setup demonstrates that cutting-edge models can be deployed within typical data centers' existing power budgets, removing the prohibitive energy costs and complex infrastructure requirements of GPUs and making advanced AI truly accessible for enterprise customers.

1

FuriosaAI @furiosa.ai · 27d

Today we partnered with OpenAI for the grand opening of its new Seoul office, where we showcased its new open-source gpt-oss 120B model running live on RNGD.

We demonstrated a real-time chatbot efficiently running the model on just two RNGD cards, using MXFP4 precision: furiosa.ai/blog/furiosa...

1

FuriosaAI @furiosa.ai · 28d

Thank you to everyone who has already stopped by to connect with us. If you haven’t yet, we’d love to meet and discuss our Tensor Contraction Processor (TCP) and our flagship AI accelerator, RNGD, at 📌 Booth #728!

FuriosaAI @furiosa.ai · 28d

Day 1 at #AIInfraSummit was energizing. The conversations at our booth make it clear: the future of AI depends not just on smarter models, but on more efficient compute.
That’s why we were excited to have Alex Liu onstage sharing how we are redefining performance and efficiency in AI infrastructure.

1

FuriosaAI @furiosa.ai · Sep 5

🛬 Next week, we’re at #AIInfraSummit.

👀 Watch our SVP of Product and Business, Alex Liu, present on the Enterprise AI stage at 4:10 PM on September 9.

🤝 Connect with us in the app, stop by Booth 728 from September 9 to September 11, and book a meeting in advance: lp.furiosa.ai/ai-infra-sum...

FuriosaAI @furiosa.ai · Sep 3

Our engineers tackle groundbreaking challenges head-on. This month, we feature Seung Ho Song, a key engineer on our compiler team, and ask him about his experience.

Read the full spotlight in our newsletter: www.linkedin.com/pulse/furios...

FuriosaAI @furiosa.ai · Aug 27

The annual Hot Chips conference this week reminded us to share our article from the recent special issue of IEEE Micro, which highlights our Hot Chips 2024 presentation as one of the best at the event.

👀 Read the full article here: dxttx52ei7rol.cloudfront.net/Micro_202503...

1

FuriosaAI @furiosa.ai · Aug 21

🔥 From our LG AI Research news to our $125 million funding raise, and from new executive hires to presenting six papers at ICML and ACL, it’s been a busy few months.

We also attended events in Singapore, Paris, and NYC.

Read our latest newsletter for details. 🗞️

www.linkedin.com/pulse/furios...

FuriosaAI news: LG AI Research, $125M in funding, new executives, and more

LG AI Research taps FuriosaAI to achieve 2.25x better LLM inference performance vs.

www.linkedin.com

FuriosaAI @furiosa.ai · Aug 19

🙏 Thank you to our team and all our collaborators, including Kangwook Lee, Yuchen Zeng, Thomas Zeng, Coleman Hooper, Amir Gholami, and Kurt Keutzer.

👀 Learn more: furiosa.ai/blog/furiosa...

Furiosa presents papers about improving AI performance at ICML 2025…

Last month, we presented four papers at ICML 2025 in Vancouver and two papers at ACL 2025 in Vienna, authored by our employees and our collaborators.

furiosa.ai

FuriosaAI @furiosa.ai · Aug 19

We’re grateful to have collaborated on these projects with the talented researchers at Korea University, Seoul National University, University of Wisconsin-Madison, Ajou University, UC Berkeley, UC San Francisco, ICSI, LBNL, Microsoft’s Gray Systems Lab, and University of Lisbon.

2

FuriosaAI @furiosa.ai · Aug 19

Last month, we presented four papers at ICML 2025 in Vancouver and two papers at ACL 2025 in Vienna.

These six papers dig into ways to make advanced AI systems more efficient, more capable, and more flexible.

1

FuriosaAI @furiosa.ai · Aug 15

Regardless of model structure, most AI computations today consist of tensor contraction operations, which is why we adopted tensor contraction as a primitive at the critical interface between hardware and software.

Learn more about our Tensor Contraction Processor (TCP): furiosa.ai/blog/tensor-...

Tensor Contraction Processor: The first future-proof AI chip…

To meet the world’s demand for AI compute, it’s not enough to build chips capable of more operations per second. You need the Tensor Contraction Processor.

furiosa.ai

FuriosaAI @furiosa.ai · Aug 15

Our Head of Product, Donggun Kim, explained how creating a flexible and efficient architecture can address the challenges AI accelerator companies face, given the rapid pace of AI model development and the long lead times that hardware requires.

1

FuriosaAI @furiosa.ai · Aug 15

At OCP APAC last week and at OCP Korea this week, we presented “From Silicon to AI Serving: Optimizing Inference and Engineering What’s Next”, which covered everything from designing and manufacturing silicon to applying it to real-world AI services.

1