🦔Nvidia-backed Firmus pulled its $5 billion IPO on Friday, citing market volatility and bad offering terms. The Australian AI data center builder wanted a $30.6 billion valuation, roughly triple what investors paid in August. My Take Firmus brought in $50.8 million in revenue last fiscal year and lost $68.2 million. It runs two data centers and needs about $37.7 billion to build five more. Asking public investors to pay $30.6 billion for that was a huge swing, and Australian fund managers said so in the press all week before the deal fell apart. Firmus was also one of the first companies to sign up for Nvidia's offer to rent back unused chips if customers don't show up. Even with the biggest name in AI standing behind it, buyers still said no at that price. I think the IPO window is closing for AI infrastructure companies without serious revenue, and I expect the next neocloud listings, like Nscale and Lambda, to face the same hard questions. Hedgie🤗

🦔PNC Bank's credit card data shows about 2% of US households pay for an AI subscription. Roughly 3 million out of 131 million. Most spend $20 a month and cancel within seven months. For comparison, 25% pay for streaming. 5% pay for sports betting. My Take Tech companies spend on AI like it's already essential. PNC's data says it's nowhere close. The five biggest plan to drop $725 billion on infrastructure this year. OpenAI lost $21 billion on $13 billion in revenue. All that money and 98 out of 100 households won't pay $20 a month for any of it. Look at the seven-month cancellation number. People aren't ignoring AI. They try it, use it for a while, and quit. Netflix took off because you could cancel cable and immediately save money. AI doesn't have that moment yet for regular people. It's useful in a vague way but nobody can point to what it replaced. Millions use the free versions. People like AI fine, they just won't pay for it. I think that 2% climbs to maybe five or six percent over the next couple years as the products improve, and Wall Street will call it a win. But $725 billion in spending was priced for something a lot closer to 30% and Silicon Valley has built for a future that regular people haven't bought into yet. Hedgie🤗

🦔Nvidia wants banks and private credit to treat its GPUs like aircraft, a long-lived asset you can lend against for a decade. Wall Street isn't buying it. Lenders underwrite GPUs on a 3-4 year schedule and want stronger guarantees than Nvidia offered. Morningstar flagged parallels to dot-com era vendor financing. My Take Nvidia is on every side of every AI transaction at this point. They sell the chips, finance the purchases, invest in the buyers, and guarantee the debt when the buyers can't get credit on their own. Now they want to turn those same chips into a $500 billion asset class. One company touching every link in the chain and asking to create the collateral framework too. Jensen Huang says his GPUs last a decade. The banks who would lend against them say 3-4 years. That disagreement decides whether the whole plan works or whether lenders eat massive losses when the hardware goes obsolete halfway through the loan. Every deal that has closed so far confirms the market sides with the banks. CoreWeave's loan got its rating because Meta stood behind it, not the GPUs. Broadcom backstopped 80% of Anthropic's deal. Nvidia itself had to guarantee SB Energy's project. Morningstar compared this to dot-com vendor financing. I think they were generous. Cisco financed its own customers the same way 25 years ago and that didn't end the way Cisco expected either. Hedgie🤗

🦔AMD is acquiring World Labs, the AI lab founded by Stanford professor Fei-Fei Li, for $8.2 billion in an all-stock deal. World Labs develops "world models" that simulate 3D environments for robotics and autonomous vehicles. It has no commercial product. Li becomes AMD's chief scientist and EVP. AMD's second largest acquisition ever, behind the $50 billion Xilinx deal in 2022. This follows Nvidia's $33 billion on Groq and Hugging Face, OpenAI's $6.4 billion for Jony Ive's io, and Meta's $14 billion minority stake in Scale AI. My Take AMD paid $8.2 billion for a lab with no revenue and no commercial product. All stock, so AMD shareholders absorb the dilution if world models turn out to be a dead end. Lisa Su has been chasing Nvidia for years and this is her way of saying AMD won't win that race on chips alone. Fei-Fei Li is one of the most respected researchers in the field and getting her as chief scientist is probably worth more to AMD than anything World Labs has built so far. But look at the prices in this market. Nvidia paid $33 billion for Groq and Hugging Face. OpenAI paid $6.4 billion for Jony Ive's startup. Meta paid $14 billion for a minority stake in Scale AI. Now AMD pays $8.2 billion for a research lab. These are talent acquisitions with company-sized price tags. The acquirers are all paying with stock or cash they raised from investors who assume AI revenue will eventually justify the spend. If that revenue doesn't show up, every one of these deals becomes a writedown. And none of these acquired companies had significant revenue at the time of purchase. The whole AI M&A market is built on the assumption that the technology will eventually make money. Hedgie🤗

🦔Internal documents from Mother Jones' lawsuit against OpenAI and Microsoft show the companies knew exactly what they did. A Microsoft director called it "an astonishing theft of unprecedented proportions." OpenAI's president responded "ah nice" to a New York Times paywall hack. Microsoft's research found AI search cut clicks to news articles by 90%. After the lawsuits, OpenAI built a filter that blocked output only from publishers who sued, not to prevent infringement but to limit evidence. My Take The Buckmaster case hit hard because it was one mathematician and one model. This is the same thing at industrial scale, and now we have the companies' own executives confirming it in internal documents. They knew. They did it anyway. And when the lawsuits came, OpenAI's response wasn't to stop using copyrighted content. It was to stop the companies that sued from gathering evidence. I think these documents change the conversation around AI valuations. OpenAI wants $1.5 trillion. Microsoft trades at $3 trillion. Both companies built their AI products on content they took without permission, and their own people called it theft internally. If these lawsuits result in licensing obligations or damages at scale, the cost structure for every major AI model changes overnight. The models were trained on stolen labor and the executives knew it while they were doing it. That's not my opinion. That's what their own documents say. I think any investor looking at AI valuations right now needs to factor in the liability that comes with how these models were built, because the companies clearly weren't expecting anyone to read their internal communications. Hedgie🤗 motherjones.com/politics/202…

🦔Watch this police drone in Istanbul with a spotlight and speaker order people out of a park. If you think this stays overseas, Flock Safety bought a drone company for $300 million last year and is selling autonomous police drones to US cities right now. They fly to 911 calls on their own, read license plates from 2,000 feet, and stream live thermal video to officers. Flock is targeting 100 US cities by end of 2026 and the FAA approved more autonomous drone waivers between April 2025 and February 2026 than in the previous seven years combined. Every time someone says the Blade Runner comparisons are overdone, a video like this shows up. Pflugerville couldn't keep track of who was searching 28 stationary cameras and now the same company is selling things that fly. I honestly don't know if this makes cities safer or just makes surveillance cheaper. Probably both. What do you all think? Hedgie🤗