AI Stockpocalypse, Anthropic Nears RSI?, Prediction Markets vs. Pollsters

Data Center Backlash After Midterms Rattles the AI Trade

Funding and deals

• Anthropic. AI lab seeking an IPO the week of November 16 at $1.2tn to $1.5tn, with an $80bn October run rate leaked and investor skepticism.

News

• Voters curb data centers in 38 jurisdictions. Democrats took the House and Senate, and 38 jurisdictions banned or restricted data centers. The pro-data-center governor of Texas lost. Ranjan Roy called the backlash the least surprising result and said it adds unpredictability to Anthropic's IPO.
• AI stocks swing sharply after the midterms. Tech stocks rose after the election, fell 4% on Thursday and bounced on Friday. Ranjan Roy said the Nasdaq fell more than 6%. Alex Kantrowitz noted trillions in market value rest on two companies with $100bn in revenue.
• Data center build-out will miss valuation math. Ranjan Roy argued build-out will not happen at the scale valuations require, citing Ben Thompson's view that political capacity is the binding constraint and Anthropic's $518bn in compute commitments. Alex Kantrowitz called it the start of a negotiation.
• METR finds Anthropic research pace doubling. METR reported that the rate at which Anthropic's models improve roughly doubled as of October, with about 80% of the acceleration coming from internal AI use. Anthropic redacted much of the work, which Ranjan Roy criticised.
• Prediction markets trailed the polls on election night. CNN and ABC News showed prediction market odds, with prices at 97 to 99 cents before calls. Kalshi's Tarek Mansour said favorites won 479 of 487 races, while Nate Silver and Chris Hayes said markets followed the polls.
• OpenAI pledges limits on data center deals. OpenAI pledged to avoid NDAs in data center deals and to spare residents the cost of project electricity. It will also curtail data center use at peak hours. The New York Times found sites pitched under code names.
• Local permitting threatens hyperscalers more than antitrust. Ranjan Roy cited New Mexico's $25bn case against Meta and Lina Khan's FTC as past action that changed little, but argued local data center permitting is far more unpredictable and more likely to bite.
• Hosts split on whether AI capex is peaking. Alex Kantrowitz doubts AI capital expenditure is peaking and expects it drawn out over more years. Ranjan Roy would sacrifice frontier training compute first if capacity runs short, and noted efficiency gains came from constrained markets.