AI Market Intelligence Daily

1. Anthropic's market share is being repriced sharply higher while the headlines tell a quieter story. The week's most dramatic single move was the "Anthropic market share this week?" contract on Kalshi, which surged from 3% to 21% — a 600% jump — and the related "above 2.85%" bucket rose from 40% to 53%. No top AI story explains that swing, but the most plausible read is that this week's weekly usage window closed while Anthropic absorbed a genuinely busy product week: merging Claude chat and Cowork into a single interface, and embedding independent safety evaluators. The market is effectively rewarding Anthropic's execution pace over the trailing week rather than any single announced milestone.

2. GPU compute prices are trading up across the board, and that is the clearest macro signal in today's data. The "H100 above $2.5 at 4 PM ET on Sep 25" contract jumped from 35% to 90% in roughly a day, the H100 "above $4 on Oct 02" more than tripled from 5% to 15%, the RTX 5090 monthly-average-above-$0.75 contract climbed from 9% to 30%, and even the low-probability B200-above-$9 contracts ticked higher. The direction is consistent across every GPU tier — traders expect spot and monthly compute pricing to stay elevated into late September and early October, which is exactly the kind of sustained demand pressure that the "Google, Nvidia and Anthropic want Emerald AI to find space on the grid" story describes. That story is the one headline that genuinely lines up with the price action: big labs are still hunting for more power and more capacity, and the compute markets are pricing in a shortage that has not yet broken.

3. The "best model" rankings quietly rotated toward Anthropic while DeepSeek cooled. Claude is the dominant answer on both "best AI this week" (98%) and "best AI on Dec 31, 2026" (70%), and "which company has the best coding model on Dec 31" sits at 67% for the same lab. At the same time DeepSeek's weekly-share buckets slid across the curve — the "above 25.3%" bucket fell from 80% to 56% and "above 26.3%" from 32% to 14% — even though DeepSeek is still the clear top Chinese name. The rotation is modest in absolute terms but directionally coherent: the market sees Claude as the current quality leader and DeepSeek as strong-but-slightly-fading on the weekly usage metric.

4. IPO timing markets are telling a consistent, slightly more bullish story about Anthropic than about OpenAI. "Anthropic IPO before 2027" is 86% (up from 84%), "Anthropic has the highest IPO market cap of 2026" is 55%, and the "who IPOs first" contract reads 91% in Anthropic's favor. The OpenAI side is near-closed: "OpenAI IPO by Dec 31 2026" is 5%, "OpenAI not IPO by Dec 31 2026" is 96%, and "OpenAI IPO before 2027" is 4%. The interesting nuance is the valuation ladder — "OpenAI market cap $1.5T+ at IPO" is 70% (down from 76%), while the $1.25T–$1.5T bucket is 10%. Traders are not questioning whether OpenAI will eventually be a mega-cap; they are trimming the top of the range a bit. The practical read is that Anthropic is now the market's preferred near-term IPO story.

5. A few markets diverge from what the news would suggest, and they are worth flagging. "AI wins an IMO gold medal in 2026" collapsed from 14% to 4% overnight even though no top story reports a model actually winning the medal — that looks like a correction of overpricing rather than fresh negative news. "Will OpenAI be acquired before 2027" ticked up from 2% to 2% (a rounding-level rise) and "Will OpenAI have the highest IPO market cap of 2026" is essentially zero at 1%, which together imply the market has largely de-risked an OpenAI buyout scenario. And "U.S. enacts an AI safety bill before 2027" slipped from 16% to 12% — a mild bearish nudge on near-term AI regulation that sits in mild tension with the safety-evaluator headlines, which read more like voluntary lab governance than imminent federal legislation.