Quick state of play. Since the last briefing, the big labs didn't ship a flagship. What they did was file paperwork, raise debt, and lobby Congress. So that's where we go.
This is YegerPod. I'm Yeger. It's Wednesday, June tenth. Today in AI: the week AI stopped being about models and started being about money.
Quick state of play. Since the last briefing, the big labs didn't ship a flagship. What they did was file paperwork, raise debt, and lobby Congress. So that's where we go.
Start with Anthropic. On June first, they filed a confidential draft S-1 with the SEC. That's the first step toward going public. No share count yet, no price — they're giving the regulator time to look before the prospectus goes out.
Here's the context that matters. The filing landed days after they closed Series H — sixty-five billion dollars raised, at a nine hundred and sixty-five billion dollar valuation. Run-rate around forty-seven billion. Reports say OpenAI is preparing its own IPO paperwork too. Two of the largest listings of the year, chasing the same pool of investors.
Analysts are now treating a trillion-dollar debut as the base case. Which means that private valuation is heading straight for a public-market test — where the growth rate sets the multiple, not the story.
Now the part underneath the IPO. The compute bill.
According to reports, Apollo and Blackstone are structuring about thirty-six billion dollars of debt — to buy Google TPU chips, which then get leased to Anthropic. Data centers in New York, Texas, Louisiana, Indiana. Broadcom, who co-developed those TPUs, backstops part of the largest tranches.
Sit with the structure for a second. This isn't equity. It's debt, raised to buy silicon, to rent back to an AI lab. For scale — Anthropic's SpaceX compute contract alone runs one and a quarter billion dollars a month.
Then Washington. On June second, President Trump signed an executive order — "Promoting Advanced AI Innovation and Security." Instead of mandatory pre-release testing — the draft the administration pulled in May — it leans voluntary. A Treasury-run cybersecurity clearinghouse. And a framework where developers may give the government access to a frontier model for up to thirty days before release. May. Not must.
The order explicitly bars any mandatory licensing to publish a model. The posture is innovation first, security optional.
Two days later, June fourth, Congress went the other direction on scope. Representatives Obernolte and Trahan unveiled the Great American AI Act. Two hundred and sixty-nine pages. The headline provision — a three-year freeze on state AI laws. It would override state efforts already in motion, including Colorado's.
Companies above five hundred million in revenue would publish governance frameworks, report serious safety incidents, admit auditors. There's a hundred-million-a-year federal standards center in it. Reactions split hard — the AFL-CIO called it a giveaway to the industry; tech groups praised it. It's a discussion draft. Not law.
So in one week: one branch says security is voluntary, another says states aren't allowed to make their own rules. That's the regulatory picture right now — pulling in two directions at once.
Now the market read, because this is the number you can't get elsewhere. Polymarket, on which company has the best AI model at the end of June. Twelve point nine million dollars of volume. Anthropic — ninety percent. Google, six. OpenAI, two. xAI and the rest, rounding to zero.
Two things there. Anthropic at ninety is up from the mid-eighties in the last briefing — the S-1 week didn't dent the market's confidence in the model. And OpenAI at two percent, despite being the name everyone knows. The traders are pricing the model, not the brand.
One more, for the local side — Bob's lane. The open-weight shelf keeps filling. Gemma 4's twelve-B that runs in sixteen gigs of RAM, Kimi K2 point six, Qwen three point six, GLM five point one, NVIDIA's Nemotron Cascade. The gap between what you can run on one machine and what you rent through an API keeps narrowing. None of these is a flagship moment. Together they're the actual story of the year — frontier-adjacent capability, downloadable, free.
And one signal to watch. Reports say the Pentagon is testing OpenAI and Google models to potentially replace Anthropic's Claude in classified systems. Testing. Not switched. But worth tracking — the same week Anthropic files to go public, its government footprint is up for review.