AI Market Intelligence Daily

1. Compute markets are telling a mixed-demand story while the headline news screams scarcity. The single most telling cluster is on Kalshi, where odds that H200 prices rise next week jumped from 3% to 12% and H100-above-$3 contracts doubled from 1% to 2%, while at the same time B200 contracts above $7 and $7.50 collapsed by 86-88% toward 1%. The market is not pricing a uniform shortage: it sees the older H100/H200 fleets getting tighter and the newer B200 fleet staying comfortably priced near $5.50-6.50, exactly what you would expect if Amazon tripling its Nvidia chip order and Anthropic's fresh $45B Nscale compute deal are being absorbed by existing capacity rather than spilling over into the newest SKUs.

2. AGI and frontier-model markets moved in ways that do not line up cleanly with the day's news. On Polymarket, "OpenAI announces it has achieved AGI before 2027?" tripled from 8% to 20% even though nothing in today's feed points to an OpenAI AGI announcement; on Kalshi, by contrast, "OpenAI announces AGI before 2030" fell from 55% to 48% and the before-2028 leg slipped from 39% to 37%. The most plausible read is that the Polymarket move is being driven by the executive-exodus coverage — a leadership shakeup at OpenAI can look like pre-announcement repositioning — rather than any genuine capability signal, and the cross-venue disagreement is the kind of divergence traders should watch before the next snapshot.

3. The "best AI on December 31, 2026" ladder on Kalshi quietly consolidated around Anthropic. Claude sits at 69% and has not moved much, while ChatGPT slipped to 14%, Muse Spark to 2%, and Grok edged up to 11% — a modest re-rating for Grok that tracks with the xAI coverage cycle, but the real story is that the market now assigns Anthropic roughly a two-to-one lead over every rival combined for year-end. That is reinforced by the Mythos release calendar, where "next Mythos-Class before November 1" sits at 93% and "before September 1" rose from 37% to 42%, i.e., traders are front-loading Anthropic's next model into September.

4. OpenAI's model-release markets are pricing a compressed Q4 window. GPT-6 before September 1 sits at just 2% (up from 1%), before October 1 is 18%, and before November 1 is 57%, which means the market's median expectation lands GPT-6 in late October — while Polymarket's broader "GPT-6 by December 31" contract actually fell from 70% to 62%. The two venues are telling different stories about the same release: Kalshi's calendar markets see a fast October drop, Polymarket's binary sees the year-end bar getting harder, and the divergence is worth one more day of data before anyone treats either price as settled.

5. A final divergence worth flagging is usage share versus model quality. OpenAI's weekly share contracts jumped hard (above 16.5% now 93%, above 19.5% up from 13% to 21%) and the "top AI model this month" contract that led the day's 400% relative move, yet ChatGPT's odds of being the best model on December 31 fell to 14% and DeepSeek's "most popular this week" leg dropped from 15% to 11%. Traders are pricing OpenAI as the default usage choice while simultaneously discounting its frontier position — a split that mirrors the news flow, where the company is shipping ads into free ChatGPT tiers in India (monetization, not capability) amid the executive exodus.