Lena
LightNews LightNews
banner
puncake.bsky.social
Lena
@puncake.bsky.social
320 followers 53 following 3.6K posts
Trains trance trans | 25y/o Dutch trans woman (she/her) who likes transportation and music | Other interests: LEGO, STEM, politics, motorsports | No discrimination, homophobia, fascism and AI please | Motorsport: @puncakemotorsport.bsky.social
Posts Replies Media Videos
Reposted by Lena
yuusharo.com
Was it a “matter of freedom” when they pulled Devotion from sale at the behest of the CCP?

The game should be available on these storefronts, but they really need to shut their mouths when it comes to their inconsistent application of their own policies.
January 16, 2026 at 4:41 PM
Reposted by Lena
edzitron.com
I am terrified that the end of the AI bubble will be magnitudes worse for the global economy than the dot com bubble, and for society at large. So many people have been so confidently wrong, and I'm not sure what happens when that becomes impossible to ignore.
www.wheresyoured.at/dot-com-bubble/
Depending on how right I am, I worry that the following will happen: NVIDIA will lose anywhere from 40% to 90% of its revenue in a relatively short amount of time, which will annihilate the value of the S&P 500. Microsoft, Google, Amazon and Meta will face massive, multi-billion dollar impairments within the next few years. Oracle will face potential bankruptcy or at least a severe, horrendous series of layoffs and restructuring efforts to try and right the ship after agreeing to $248 billion in lease obligations primarily to get the revenue from its $300 billion, 5-year-long deal with OpenAI, a company that is going to run out of money in the next year. Venture capital is going to take such a severe haircut from the AI bubble that it will effectively kill any chances of raising a round after Series B for many, many years. Those $178.5 billion in data center debt deals will be predominantly underwater, leading to massive losses for private equity and banks (leading to a chill in effectively all debt markets), and massive job losses across the many, many construction projects that have started as a result of the AI bubble. If OpenAI or Anthropic can manage to go public, every investor will lose out as they collapse under the weight of their bullshit assumptions. Every partner, every company relying on them, every infrastructure provider will be burned. I also believe that we, as a society, will face a violent paradigm shift as a result of the AI bubble bursting, because it revealed how many people do not understand stuff. The amount of bosses, influencers, movie stars, politicians and other individuals of note that have fallen behind generative AI and claimed again and again that it can do things that it can’t and perpetuated myths that will lead to massive societal harms both today and in the future have revealed themselves as fucking frauds. Large Language Models are a global gaslighting experiment, where inefficient and questionably-functional software can convince you that it works because you managed to manipulate its handles with the right prompts — explaining away its failures as you go, and getting increasingly agitated with those who don’t agree with you that you’re living in the future. This piece forced me to live in the past for hours and hours and hours, and all it taught me is that people are desperate to compare this era to another as a means of explaining away the bad habits and specious ideas of a tech industry run by fucking influencers and management consultants. By comparison, the Dot Com Bubble was sensible, built on good ideas executed by greedy people rushing to do things based on ideas they didn’t understand. The AI Bubble is happening in a remarkable era of digital information and connectivity that allows us to process and analyze details at scale and with speed, and everybody who inflated this era should — and by my fucking sword will be made to — experience inextricable, permanent shame for the horrors that will follow as a result. And believe me, even if I’m half right, things will likely be so much worse than when the dot com bubble burst. I don’t see a scenario where anything survives the collapse of CoreWeave, let alone the collapse of OpenAI. I don’t see a scenario where any more than a few AI startups survive, and if that happens, there won’t be any customers for those GPUs — not that anywhere near enough existed to fill even a tiny percentile of the demand that currently exists, or for the capacity we’ve already built. On some level, I hope I’m wrong. I just don’t know how anymore. Anyway, see you next week.
January 16, 2026 at 5:43 PM
Reposted by Lena
edzitron.com
The AI bubble and Dot Com bubble were both bubbles — but that’s the only real point of comparison between them. The AI bubble has no "dark fiber" story, AI startups are inherently unprofitable thanks to LLMs' rotten economics. There's no way out I can see.

www.wheresyoured.at/dot-com-bubble/
While website-based dot com startups died because they tried to grow too fast and go into too many verticals, dot com infrastructure startups died because they took on a ton of debt and were burdened with massive amounts of opex, capex and debt, on top of the massive headcounts necessary to keep growing. AI startups combine these into a unique Arnold Palmer of pain. Running AI models is incredibly expensive, with variable costs that are almost entirely impossible to reign in. Sites like Pets.com, WebVan, Boo and TheGlobe collapsed due to the unsustainable costs of expansion — companies like Perplexity and Cursor will die because of the unsustainable costs of running their services. Things are materially worse for AI labs, who have to both keep paying the unprofitable cost of inference (running the models) and the continual cost of training new models to keep up with each other. The reason that there are so few AI labs is that it costs hundreds of millions of dollars to keep models competitive, let alone adding new functionality. The cost of inference may, somehow, come down one day (though not before these companies explode), but the cost of training will only ever expand, and it’ll expand at whatever rate the largest, best-funded entity chooses to. I cannot find evidence of a single AI company that’s actually profitable. Not “gross margins,” not “profitable on inference” (fuck off!), not “profitable if you remove certain costs,” but actually making more money than it costs to run the services and pay for the business’ cost. Image Link: https://drive.google.com/file/d/1QnWpawZIW1Wt3UjwPbH1PDVbfBejQONr/view?usp=drive_link Image Alt Text: The dot com bubble was relatively distributed, and created useful infrastructure. The ideas of some of the most prominent dot com startups were valid, but based on stupid hypergrowth business models that didn’t make sense and were only sustained through mass hysteria and ignorance. By comparison, AI startups are fundamentally i…
January 16, 2026 at 5:43 PM
Reposted by Lena
edzitron.com
It’s no longer a matter of “if” but rather “when.” The Dot Com Bubble failed because startups spent lavishly and grew too fast.

The AI bubble is different. More capital intensive. More opex intensive. Every part of the chain is unprofitable other than NVIDIA.
www.wheresyoured.at/dot-com-bubble/
I’m really fucking sick of hearing people say “this is just like the dot com bubble,” especially when it comes to startups. In the late 90’s and very early 2000s, the incorrect assumption was that because anybody could start a website, any internet business would be huge. This assumption gave anybody who mentioned “internet” a stock bump and credulous coverage. My favourite example was about an investment made by Fyffes, a global producer of fruit with over 2 billion euros in revenue at the time, in a website called “World of Fruit” from a January 2000 piece in the Irish Times: It is only when a workable website turns a business model on its head that people really begin to understand how the Internet can change traditional business. The latest example to shake up an industry and have a direct impact on the company share price is Irish fruit company, Fyffes. Last week, it officially launched worldoffruit.com, the first Internet fruit market.… Launched with a $10 million (€9.7 million) investment from Fyffes, discussions have already begun for venture-capital funding which is expected to amount to more than $20 million, according to industry sources. In October 2000, WOF announced it had “chosen Coface, one of the largest credit insurance agencies in the world, as a partner to provide online credit ratings and commercial insurance to worldoffruit.com users.” In January 2001, Fyffes “axed plans to develop its internet venture worldoffruit” and laid off half of its staff. The Irish Independent added one remarkable detail: Just a year ago, some analysts said worldoffruit could be worth between euro 600m and euro 1bn if floated on the stock exchange. Other players have been slow to come on board and the only real business generated came from Fyffes itself, one analyst noted. World of Fruit’s only real customer was, unsurprisingly, Fyffes itself. So, for the sake of our sanity, let’s separate the two kinds of “dot com startups” into two comparables: Dot Com Website Startups Dot Com Infrastructure Startups And then immediately say that they do not remotely compare to the AI bubble. There are really only a few kinds of AI startups: AI infrastructure companies like CoreWeave that exist to buy GPUs and rent them to people. AI software startups built on top of models made by other people. AI labs that make their own models.
January 16, 2026 at 5:43 PM
Reposted by Lena
edzitron.com
The AI bubble is, as a result, a lot simpler — and thus, more dangerous. The magnificent 7 - especially NVIDIA - is now a load-bearing beam of the global markets, both in their valuations (which aren't based on actual AI revenue) and GPU spend.

www.wheresyoured.at/dot-com-bubble/
While the dot com bubble was a mixture of dodgy venture capital deals in websites that could never turn a profit combined with a global mania around the interconnectivity of high-speed internet companies, the AI bubble is one company selling GPUs, a bunch of companies building data centers to put them in, and a bunch of companies building shit that runs on GPUs. Economically-speaking, this meant that telecoms companies, server hardware companies, internet service providers, construction firms, optical cable providers, wireless technology companies and basically anybody related to the business of providing internet access in any way was directly and materially benefiting from the dot com boom and suffering as a result of the dot com bubble bursting. One could get a startup funded if you had a website, or even take it public, one could raise debt to build a nascent ISP or a fiber network. As far as the dot com part of the dot com bubble — the unsustainable websites — the problem wasn’t so much the industry but the businesses themselves, which were then hyped and dumped onto the public markets with little regard for their long-term health. The problem here was relatively simple: these were bad companies that people ignored the issues with because “the power of the internet” would somehow save them, which it did not. Had they been kept private and died in the dark, I don’t think these companies would have had the same reputation. I also don’t really think they were an accurate comparison to anything happening today. While valuations were ridiculous — TheGlobe’s market cap was at one point over $840 million — the scale of destruction caused by dot com startups was significantly smaller, even in today’s money. The economics were bad, but not anywhere near as bad. For the first nine months of 1998, TheGlobe made $2.7 million in revenue and lost $11.5 million, largely due to trying to move into multiple different business lines (like a Voice over IP phone service, and … While the dot com bubble rested on the back of companies like Lucent, Cisco, Nortel, WorldCom, Enron, and others, the AI bubble rests, fundamentally, on one company: NVIDIA, and to a lesser extent the valuations of the remainder of the magnificent 7 — Microsoft, Amazon, Google, Meta, Apple and Tesla. Four of those companies (Amazon, Microsoft, Google and Meta), through intermediaries I’ll get to in a minute, have spent hundreds of billions of dollars on GPUs and their associated infrastructure for reasons that none of them can seem to explain. NVIDIA’s revenue is predominantly (88%) in its data center segment, and its customers are those who can afford, at the very least, 50 to 100 GPUs retailing at $400,000 or so for a pod of 8, requiring tens of thousands of dollars of networking gear to go with them. The customers of those renting those GPUs are either AI labs training or running inference for models, and their customers are AI startups. The problem isn’t so much that nobody can afford a GPU, but that you can’t get very far with just one — you have to buy so many of them, and you need to build a data center around them, and you need to get power to that data center. This naturally means that there are really only two customers who can afford these chips at scale: The magnificent 7, who have begun to take on debt after previously financing their GPU purchases with cashflow. Companies that raise debt, with “companies” meaning “anybody who wants to build a data center” and Oracle, which had negative $13 billion in cash flow last quarter and is steeped in debt to the point that bondholders are suing. We also have no idea if the economics of renting GPUs actually makes sense, and based on everything I’ve found, I’m not sure anybody renting them can ever make a profit, due to either or both the up-front cost (and debt necessary to pay it) and the power-intensive nature of providing AI compute. It is fundamentally insane that we don’t know for sure. What we do kno… The demand for NVIDIA chips is fueled by hype, and that hype has caused this company — and to a lesser extent the magnificent 7 — to become load-bearing parts of the American stock market. An analysis from portfolio manager Danke Wang from January 2025 found that the Magnificent 7 stocks accounted for 47.87% of the Russell 1000 Index's returns in 2024 (an index fund of the 1000 highest-ranked stocks on FTSE Russell’s index). In really simple terms, without the mostly vibes-driven nature of the magnificent 7’s growth (as none of this is based on anyone’s actual AI revenues), the US stock market would be in incredibly rough shape. Except unlike the dot com bubble, most of these companies have taken on incredibly large amounts of GPUs, debt, finance and operating leases, and data centers full of GPUs that can’t be used for anything else. And because GPUs are guaranteed to depreciate, each and every one of them will, without fail, have to write down the value of upwards of a hundred billion dollars of investments. This is going to happen whether or not the bubble bursts, as NVIDIA is on a yearly cycle of GPU launches. Every single year every single GPU investment loses value. Even if the AI bubble doesn’t burst, the US stock market has an unhealthy relationship with NVIDIA, which, by this time next year, will have to make over $90 billion a quarter to keep up with its 50%+ year-over-year growth. And by 2028, NVIDIA will — to keep its ridiculous valuation — have to be making more than Apple ($416 billion) a year in revenue.
January 16, 2026 at 5:43 PM
Reposted by Lena
edzitron.com
Whereas the Dot Com Bubble’s losses (either through death or contraction) were distributed, the AI bubble is concentrated in a few companies, all of which now account for a larger share of the S&P 500, and AI startups are centralized around only a few providers.
www.wheresyoured.at/dot-com-bubble/
In July 2000, Corning (which made far more than just optical cables) became the largest supplier of fiber-optic wires, and ended 2000 with $7.13 billion in revenue on net income of $422 million. By April 2001, Corning had revised its earnings estimates down three times, estimating revenue for the year to be between $7.8 billion and $8 billion. Corning would eventually reveal net sales of $6.27 billion with a loss of $5.498 billion for the year. Fiber optic cable firm JDS Uniphase, a conglomerate of different optical companies merged in 1999, would go on an acquisition spree to dramatically expand its fiber optic offerings, eventually crowning itself the “no.1 supplier of fiberoptic components” and revealing net earnings of $208 million in January 2001. In July 2001, it would announce a 35% decrease in sales quarter-over-quarter, and a $44.8 billion decrease in the value of companies it had acquired, on top of a net loss of $477 million in the quarter. The stock would lose 99% of its value before an eventual rebrand and collapse. Okay, so now you know all that crap, let’s get even more specific. NVIDIA currently represents 7% of the S&P 500, with a market capitalization of $4.4 trillion. In September 2000, JD Uniphase made up .70% and Corning .692% of the S&P 500, when it totalled (and it is surprisingly difficult to find the exact number!) around $11.7 trillion, with the highest-weighted tech stocks being Cisco at 3.1%, Microsoft at 2.5%, Intel at 2.2%, Oracle at 1.75% and EMC at 1.7%. In simpler terms, the dot com bubble didn’t burst because of the market’s obsession with one company, but an evenly-distributed and incredibly vague sense that The Future Was Going To Be Online And We Wanted To Be Involved. The world didn’t wait with baited breath for every single earnings announcement from Corning or JD Uniphase, nor did the health of the market depend on their continued ability to beat-and-raise revenues that were echelons higher than any other company in the market. Now, the problem with trying to find an exact comparison is that fiber optic cabling and GPUs are fundamentally different things, requiring (amongst other things) massive optical network terminals to actually turn big looms of cables into actual internet connections. The massive costs associated with the fiber buildout could make Nortel or Lucent — who each boasted “record revenues” in 2000 and got the shit kicked out of them in 2001 — the NVIDIAs of the dot com bubble, if you really want, though I must insist that I think that what we’re going through is its own unique thing. In any case, they’re the only real revenue comparison. NVIDIA’s last quarterly revenue was $57 billion, with around 88% of that coming from its data center vertical (IE: where it sells GPUs and the associated networking gear). In Lucent’s case, the Service Provider Network segment (where it bundled all of its telecoms infrastructure like optical networking and switching, per its 2000 annual report) accounted for 78% of its $33.8 billion in revenue — a slight improvement from 1999 (when it was 81%) and 1998 (when it was 83%). For Nortel, the comp was its service provider and carrier segment, which made up 82% of its $30.25 billion in revenue in 2000 (per its annual report).
January 16, 2026 at 5:43 PM
Reposted by Lena
edzitron.com
The Dot Com Bubble comparison collapses immediately when you realize that even at the time, laying fiber optic cable and building websites made sense - they just rushed to grow at rates that didn't match the growth of the internet.

AI is completely different.

www.wheresyoured.at/dot-com-bubble/
The bubble burst when the bullshit dot-com stocks died on their ass and the world realized that the magic of the internet was not a panacea that would fix every business model, and there was no magic moment where a company like WebVan or Pets.com would turn a horribly-unprofitable business into a real one. Similarly, companies like Lucent Technologies stopped being rewarded for doing dodgy, circular deals with companies like Winstar, leading to the collapse of the telecommunications bubble that led to millions of miles of dark fiber being sold dirt cheap in 2002. The oversupply of dark fiber was eventually seen as a positive, leading to an eventual surge in demand as billions of people came online toward the end of the 2000s. Now, I know what you’re thinking. Ed, isn’t that exactly what’s happening here? We’ve got overvalued startups, we’ve got multiple unprofitable, unsustainable AI companies promising to IPO, we’ve got overvalued tech stocks, and we’ve got one of the largest infrastructural buildouts of all time. Tech companies are trading at ridiculous multiples of their earnings-per-share, but the multiples aren’t as high. That’s good, right? No. No it isn’t. AI boosters and well-wishers are obsessed with making this comparison because saying “things worked out after the dot com bubble” allows them to rationalize doing stupid, destructive and reckless things. Even if this was just like the dot com bubble, things would be absolutely fucking catastrophic — the NASDAQ dropped 78% from its peak in March 2000 — but due to the incredible ignorance of both the private and public power brokers of the tech industry, I expect consequences that range from calamitous to catastrophic, dependent almost entirely on how long the bubble takes to burst, and how willing the SEC is to greenlight an IPO. The AI bubble bursting will be worse, because the investments are larger, the contagion is wider, and the underlying asset — GPUs — are entirely different in their costs, uti…
January 16, 2026 at 5:43 PM
Reposted by Lena
edzitron.com
Last week, I was at CES, where I saw the “future” according to a tech industry with no ideas — robots that’ll never be released that do things I can already do, albeit much slower, and AI assistants that were just rehashed ChatGPT.
www.wheresyoured.at/dot-com-bubble/
LG, a company known for making home appliances and televisions, demonstrated a robot (named “CLOiD” for some reason) that could “fold laundry” (extremely slowly, in limited circumstances, and even then it sometimes failed) or cook (by which I mean put things in an oven that opened automatically) or find your keys (in a video demo), one that it has no intention of releasing. The media generally gave it an easy go, with one reporter suggesting that a barely-functioning tech demo somehow “marked a turning point” because LG was now “entering the robotics space” with a product it had no intention of selling. So, why did LG demo the robot? To con the media and investors, of course! Hundreds of other companies demoed other robots you couldn’t buy, and despite what reports might say, we were not shown “the future of robotics” in any meaningful sense. We got to see what happens when companies run out of ideas and can only copy each other. CES 2026 was the “year of robotics” in the same way that somebody is a sailor because they wore a captain’s hat while sitting in a cardboard box. “AI is evolving and getting new capabilities, sensing our three-dimensional world, understanding how things move and connect,” said Lenovo CEO Yang Yuanqing, leading into a demo of Lenovo Qira, before claiming it “redefines what it means to have technology built around you.” One would think the demo that follows would be an incredible demonstration of futuristic technology. Instead, a spokesperson walked up, asked Qira to show what it could see (IE: multimodal capabilities available for years in many models), received a summary of notifications (available in effectively any LLM integration, and incredibly prone to hallucinations), and asked “what to get her kids when she had some free time,” at which point Qira told her, and I quote, that “the Las Vegas Fashion Mall has some Labubus that children will go crazy for,” referring to the kind of tool-based web search that’s been available since 2024. The presenter noted that Qira also can add reminders — something that has been available for years on most iOS or Android devices — and search for documents, then showed a proof-of-concept wearable that can record and transcribe meetings, a product that I saw no less than seven times during my time at CES. Lenovo rented out the entirety of the Las Vegas Sphere to do a demonstration of a fucking chatbot powered by OpenAI’s models on Microsoft Azure, and everybody acted like it was something new. No, Qira is not a “big bet” on AI — it’s a fucking chatbot forced on anybody buying a Lenovo PC, full of features like “summarize this” or “transcribe this” or “tell me what’s on my calendar,” features peddled by business idiots that have little experience with any real-world applications of just about anything, marketed with the knowledge that the media will do the hard work of explaining why anybody should give a shit.
January 16, 2026 at 5:43 PM
Reposted by Lena
hleehurley.com
Interestingly, he was manager of Sutton Coldfield Town's women's team. Sutton United's women's team have a trans woman as a manager and there was a campaign to get her removed. There was no campaign that I am aware of from the papers to get Ryan Hamilton removed.
January 17, 2026 at 3:17 PM
Reposted by Lena
dov.bsky.social
While Denmark might be the land's security custodian, a hired bodyguard if you wish, fundamentally, Greenland belongs /to Greenlanders./ What we are defending against the US is not Denmark's right to own it but Greenland's right to belong to itself.
January 17, 2026 at 2:04 PM
Reposted by Lena
tobbinatorscw.bsky.social
I am living as a very privileged immigrant here and my partner's family didn't even know I'm simply not eligible for a tonne of regular services, let alone the conditions applied to any migrant worker from southeast Asia
January 17, 2026 at 7:58 PM
Reposted by Lena
tobbinatorscw.bsky.social
Lots of people in America and across the EU talk about ETA/ESTA as if they're real visa impediments! Nobody really knows what a migrant actually goes through, let alone one from any poorer country than their own
January 17, 2026 at 7:55 PM