SpaceX reportedly seeks $40B, led by Apollo, to buy Nvidia chips
The compute boom found its lender, safety regulation found its loudest advocate, and the open-source world got a model that only picks answers.
SpaceX is reportedly lining up $40 billion, led by Apollo, to buy Nvidia chips — and almost all of it is debt. The Financial Times broke the plan, citing people familiar with it; Reuters and Bloomberg have since relayed the report, and SpaceX, Apollo, Nvidia and Pimco have all declined to comment, so this is reported rather than confirmed. The structure is what stands out: roughly $10 billion in bank loans plus $30 billion of investment-grade debt, with Pimco among a small group of lenders and closing expected in 2027 — a package Reuters describes as among the biggest debt financings ever raised for the AI buildout. That is a different animal than the cash flow and equity that paid for rockets: now the bill for Musk's chip habit comes due whether or not the data centers earn their keep, against a backdrop where Morgan Stanley estimates AI infrastructure needs $1.5 trillion of external financing by 2028. Apollo is already inside the machine — it is part of the Nvidia-led partnership targeting more than $500 billion of AI infrastructure financing — so the same shop is arranging SpaceX's debt and underwriting the industry's funding round at large.
Geoffrey Hinton wants an FDA for AI models: nothing ships until a regulator is convinced it is safe. The proposal came in a podcast episode released Tuesday, where the "godfather of AI" spelled out the drug-approval analogy: "You're not allowed to just make a new drug and release it on the market… you have to convince the FDA. And to do that, you have to do a lot of work, about a billion dollars worth of work. That seems like the very least we should have for AI." The urgency behind the soundbite is the more interesting part — Hinton argued that recursive self-improvement has already begun, and gave it "a year or two before everything gets much worse than it is now." One honest caveat: Business Insider is currently the only outlet reporting the remarks, and no other figure has endorsed the idea, so for now it is a serious man's one-line policy platform rather than a movement.
Amazon open-sourced Strands Decider 2B, an agent "decision model" that never writes a word. Out of Strands Labs — the experimental arm of AWS's agent framework — it takes a Qwen3.5-2B torso, strips out the text-generation head, and replaces it with a scorer that returns one of your options plus a calibrated confidence in about 115 milliseconds on a desktop GPU (153 on an M3 MacBook). The pitch: agents burn constant cheap calls on routing, tool selection and guardrail checks, and a small model that always answers from a fixed list, fast, fits that path where a frontier LLM call never could. It's Apache-2.0 with weights and training data published, and it ranks third on the 2B-class JevBench board — though a rival still beats it outright, so the win here is openness, not the leaderboard. We tracked the decision-model wave it joins — Open Source Radar — September 21: decision models go local.
What to watch: whether Apollo can place the full $30 billion, and at what spread — the price of this deal is the price of the AI buildout.
If debt is what it takes to keep the chips coming, who actually owns the AI boom in five years? Tell us in the comments.
