AI Market Intelligence Daily

1. The OpenAI IPO question is being split into two by the market, and the valuation half is drifting higher. Sam Altman's public remark that a 2026 listing would be "ill-advised" has fully landed: "Will OpenAI not IPO by December 31, 2026?" now sits at 93%, the "IPO by December 31 2026" contract at 5%, and "OpenAI IPO before 2027?" at just 6%, so the market is near-certain a listing slips past 2026. But the closing-cap contracts on Polymarket tell a different story about size: the $1.25T-to-$1.5T band climbed from 8% to 12%, the $1T-to-$1.25T band from 3% to 5%, and the "less than $500B" contract finally printed a real 2%, while the "$1.5T or greater" leg eased from 80% to 74%. The practical read is not-2026, but when it happens, price it around a trillion, with the center of gravity of the ladder still migrating upward.

2. Google's Gemini release-market is the day's biggest genuine move, and it carries an internal inconsistency worth flagging. On Kalshi, "Will Google release next Gemini Pro model before Oct 16, 2026?" jumped from 35% to 73% in two days, a strong step-up that would normally read as traders hearing the next frontier model is imminent. Yet the same venue prices "Will Google release Gemini 4 before Nov 1, 2026?" at only 68%, and "before Oct 1, 2026?" at 9%. If the next Pro model is likely out by mid-October, the market should price the November contract for the same model much higher than it does, so either the two markets are defined on different model versions or one side of the book is mispriced. Meanwhile the Gemini app-download contracts are settling with the month closing out: "above 350" for September rose from 2% to 25% against a stale ten-day baseline while the "above 330" leg sits at 2%, which is consistent with downloads tracking near the bottom of that range.

3. The biggest percentage moves of the day are weekly-reset share and leaderboard contracts settling as the week of September 7-13 closes, and they carry less signal than their size suggests. On Kalshi, "Qwen market share this week?" jumped from 29% to 92% on one leg, "Google market share this week? (above 18.4%)" from 53% to 97%, and "Most popular AI this week? (OpenAI)" from 53% to 99% as the Tencent leg collapsed from 53% to 4%. These are end-of-week settlements, not new information. The durable signal underneath: Claude remains the consensus best model on the December 31 contract at 69%, and "Which AI company will have the best coding model on Dec 31, 2026?" saw one leg surge from 6% to 29%, keeping the end-of-year coding race genuinely open rather than a foregone conclusion.

4. Compute prices are repricing higher into the second half of the month, and it reads as fresh positioning rather than churn. The Kalshi contract asking whether B200 spot price would be above $7 at 4 PM ET on Sep 18 has now reached 69%, up from 56% at the noon snapshot, while the above-$6.50 leg sits at 84% and above-$6.00 at 97%, a stacked bet that hourly prices stay elevated over the next five days. That is directionally consistent with Jensen Huang's claim that Nvidia will grow 70% next year and with OpenAI pausing Pro subscriptions to free capacity for Astra demand, both of which point to sustained GPU demand. The one divergence to watch: "Will Anthropic have the highest IPO market cap 2026?" fell from 58% to 46% on Polymarket, a real de-escalation of Anthropic's relative valuation that runs against the bullish tone of the week's Anthropic headlines, even as "Anthropic IPO before 2027?" still trades at 85%.