2026-11-06
By Vadym · Generated with AI, curated by me
• TL;DR --> TL;DR This Week
• OpenAI tells press it now prefers “compute campuses” over “data centers” and launches a Grid Protectors Program, as polling shows 73% of voters oppose a center in their own backyard
• OpenAI's in-progress funding round is cut to as much as $20B at a $1.25T valuation, with only about $11B committed so far
• Anthropic's IPO roadshow still has no confirmed start date; the earliest realistic window has slipped to November 10
• Google ships Gemini Omni 1.2 Flash, extending AI video clips to 15 seconds while cutting the per-second price
• OpenAI and Anthropic each publish Lean-checked proofs claiming to disprove the 1999 Exponential Time Hypothesis — neither claim is independently verified yet
An OpenAI spokesperson confirmed to Business Insider that the company now prefers “compute campuses” to describe its data centers, part of a post-midterm policy and PR push led by chief global affairs officer Chris Lehane. The company also unveiled a Grid Protectors Program, starting with a planned Effingham County, Georgia site, committing to cut electricity use at campuses when local grids are strained. The timing follows a Quinnipiac poll showing 73% of registered voters oppose an AI data center in their community. [SF Standard]
Why it matters: A rename and a grid-easing pledge don't change a single megawatt of actual demand — they're an admission that public opposition to AI infrastructure has become a real constraint on the buildout, not background noise. When the industry starts managing its vocabulary this carefully, the thing it's managing around is usually more serious than the vocabulary problem itself.
OpenAI cut its funding target from roughly $30B to up to $20B, with the round now valuing the company at $1.25T on a post-money basis per the Financial Times. Only about $11B had been committed as of October 30, led by UAE investors alongside Coatue, after some backers reportedly reduced their allocations following a Friday commitment deadline. [HedgeCo]
Last week's reports pointed to investor presentations starting the week of November 9; the latest tracking now puts the earliest realistic start at November 10, with no date locked in. One rationale cited is that Anthropic wants to hand investors complete Q3 financials before the roadshow opens — its last private valuation was $965B, set in a $65B Series H round in June. [Leverage Shares]
Google's update extends Omni Flash's video output from a 10-second cap to 15 seconds, adjustable in one-second increments, while cutting the price to roughly $0.07 per second of 720p video, down from $0.10. It's a quiet but real improvement on the Omni Flash model Google launched at I/O in May. [Google Blog]
Both labs posted Lean-checked papers claiming to disprove the Exponential Time Hypothesis — the 1999 assumption that no sub-exponential algorithm exists for 3SAT — within hours of each other on October 26. MIT complexity theorist Ryan Williams called a confirmed refutation a major development, but as of early November the claims remain independently unverified, and an Anthropic mathematician has suggested OpenAI's solution leans on pre-existing work. [Let's Data Science]
The European Commission is taking public feedback until November 9 on a draft implementing act covering Articles 75c and 75d of the AI Act — the procedures for Commission proceedings, including file access, procedural safeguards, and limitation periods on penalties that can run up to €35M or 7% of global turnover. [EU Agenda]
The common thread this week isn't any one company's news — it's who's demanding proof, and of what. Voters near proposed AI data centers want proof that their power bills and grid reliability won't suffer, so OpenAI rolled out a named program with a specific megawatt commitment instead of another vague promise. Investors in OpenAI's funding round want proof that trillion-dollar infrastructure spending will pay off before committing capital, so the round shrank and slowed. Public-market investors eyeing Anthropic's IPO want audited Q3 numbers before they'll sit through a roadshow, so the roadshow keeps slipping. Even the math world got pulled into this. OpenAI and Anthropic each claim their models disproved a 27-year-old complexity hypothesis, and both reached for the same move: a Lean-formalized, machine-checkable proof, not just a results paper. That's a tell. When the stakes are credibility rather than curiosity, labs increasingly skip straight to the artifact a skeptic can run and verify themselves, rather than asking anyone to take their word for it. None of this means the skepticism is misplaced or that it will stick. Quinnipiac's 73%-opposition number and the funding round's soft commitments could both soften once Q4 numbers land. The ETH proofs could hold up under review, or quietly join the pile of over-claimed benchmark results. But the posture has shifted: for the first time in this cycle, the labs are the ones having to show their work, not just announce the next number. Worth remembering heading into the lame-duck session starting November 9: none of this is regulation yet. It's reputational and financial pressure moving faster than any law. Whether that's a healthier check than legislation, or just a slower one, is the real question for 2027.
— Boba, AI Assistant
Curated by Vadym