TLDRocket
Sign in

Treasury's AI bubble warning sharpened today's finance-risk story

The Neuron

Treasury reportedly warned AI risk could ripple through banks, chipmakers and utilities. Same day, a startup raised cash just to make GPU computing power tradable.

Based on reporting by The Neuron — read the original for the full story.

Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error

Monday didn't feel like a normal AI news day. It felt like the AI boom walked into the finance department and started asking about hedging instruments. A startup called Ornn raised $33 million to build pricing, trading, and hedging infrastructure for AI compute capacity — the clearest sign yet that GPU access is turning into an asset class, something you buy, sell, and bet against rather than just rent by the hour.

That framing got heavier when NOTUS reported that Treasury analysts, in a draft report, flagged AI bubble risk as something that could spread well beyond Silicon Valley. Their worry list reads like a map of the entire AI supply chain: data-center financing, cloud providers, chipmakers, utilities, private credit, public stock markets, and institutional investors. In other words, if the compute boom wobbles, the shock doesn't stay contained to one balance sheet.

And the deals getting signed right now show why that worry isn't abstract. TeraWulf reportedly locked in a 20-year lease with Anthropic worth about $19 billion, tied to a Hawesville, Kentucky project that Business Insider described as a 400-megawatt capacity deal. TeraWulf's shares rose on the news, and its Justified campus is expected to start delivering initial capacity in the second half of 2027, ramping toward 401 megawatts the year after. Meanwhile SK Hynix launched a $28 billion U.S. share sale as investors chase exposure to the memory-chip side of the AI trade — a scramble that lines up with memory benchmark prices that Axios says have jumped roughly 660% over the past year.

But the hardware isn't cooperating with the timeline everyone wants. Nvidia's next rack-scale system, Kyber NVL144, reportedly got pushed to 2028 after PCB midplane manufacturing issues, and there's chatter about backing away from a more complex quad-die design for Rubin Ultra. So you've got long-dated leases, multibillion-dollar listings, and freshly minted trading infrastructure all stacking up around chips and racks that haven't fully solved their own manufacturing problems yet.

Put those pieces side by side and Monday looks less like a string of unrelated headlines and more like one story: money is racing to price and finance a boom whose physical foundations — the racks, the memory, the megawatts — are still being built and occasionally delayed. Treasury analysts noticing that is not a footnote. It's the whole plot.

My take — AI-written commentary, not fact-checked reporting

Nobody should be shocked that regulators are nervous about AI risk spreading through utilities and private credit when the same week produced a startup whose entire pitch is making GPU capacity tradable like a commodity. Financializing something before its supply chain is even stable is not innovation, it's just moving the risk somewhere less visible. The $19 billion lease and the $28 billion listing look great in a press release, but leases and share sales don't fix a delayed rack design. Someone is going to be holding the bag when the megawatts show up late and the hedge doesn't.

Read more about this at: The Neuron

Related stories

The daily briefing

Every AI story that matters, in your inbox by 8am.

TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the day in two minutes. Follow companies and topics for alerts, or get the briefing in Slack. Free, no spam, unsubscribe anytime.