AI Market Intelligence Daily

1. GPU rental markets repriced sharply lower overnight, and the move was broad rather than isolated. Kalshi's H100 SXM monthly contract for above $3.06 on August 31 collapsed from 48% to 3%, the above-$3 spot contract for 4 PM ET on August 28 went from 23% to 1%, and the "H100 SXM price up in next week?" market dropped 58% to 21%. Lower strikes in the same ladder held firm, with the above-$2.36 monthly now at 95%, so the curve implies H100 rental prices near $2.40 per hour with little chance of a rally into month-end. The B200 ladder tells the same story: the above-$8.5 contract fell 80% to 1% while the above-$6 strike sits at 92%, pinning B200 pricing in the low $6s. Traders are pricing in a late-August glut of compute capacity, consistent with new data center capacity coming online even as OpenAI's Jalapeño inference chip benchmarks and a fresh stream of departures from its data center team suggest the frontier labs are still actively reshaping their capacity plans.

2. xAI had the most notable upside move in frontier-model markets. "Will xAI have a top-ranked AI model before 2027?" jumped 112% from 8% to 17% over the last three days, the largest sustained model-relevance move of the week. That re-rating landed in the same window as Grok 4.7 release markets sitting in the low 30s for a September 4 launch, and it coincides with the "Best AI on Dec 31, 2026" leaderboard shifting toward ChatGPT, which rose 29% to 15% while the Grok line slipped 27% to 9%. The net read is that traders see xAI's pipeline as credible for a top-tier release in the second half but still do not expect Grok to overtake Claude (69%), the year-end favorite, before 2027.

3. Anthropic release-timing markets moved in a way that suggests a September Mythos-class launch is now the market's working assumption. The "Mythos-Class model before September 1" contract sits at 37%, "before October 1" at 78% (up 30% from 60% in under two days), "before November 1" at 89%, and "before December 1" at 14% for the plain Mythos name. The ladder implies roughly a 70% probability the next model-class drop lands in September, a meaningful revision given no headline in today's news flow announced it. The only direct news touchpoint is Claude Cowork gaining persistent chat memory, which is a product improvement rather than a model-class launch, so this move looks like traders positioning on pipeline expectations rather than a reported release.

4. IPO and capital-markets pricing stayed firmly anchored to the two-story setup: Anthropic goes public first, and at a big valuation. "Anthropic IPO before 2027" held at 91%, while the mirror market on "Will OpenAI or Anthropic IPO first?" shows OpenAI at just 10% for the front-runner spot, with OpenAI IPO-by-end-2026 at 16% and a sub-$500B IPO-day market cap priced at a flat 0%. The only wobble was the OpenAI pre-2027 line nudging up 6% to 16% in a day where the news flow carried no IPO headlines at all, so it reads as noise rather than information. With TechCrunch reporting another high-profile OpenAI data center exec departure as the exodus continues, the market is effectively saying OpenAI's near-term capital-markets path is blocked by execution and staffing questions, while Anthropic's is clear.

5. The clearest divergence between prices and today's news flow is the "Ox Alpha" mystery-model complex, which settled decisively. The market on whether a source agency will confirm Z.ai as the developer jumped 9% to 99%, while the no-company, Google, Tencent, and Moonshot alternatives all crashed 86-88% to 1%. "Most popular AI this week" now prices Ox Alpha at 94%, and "Will a Chinese AI model be #1 this year" sits at 11%. None of today's 15 top stories reports a Z.ai confirmation, so the move appears to have been driven by an off-headline attribution that traders have effectively priced as final. The second divergence is Tesla Optimus, which fell 17% to 3% for a December 31, 2026 release even as robotics news was positive, with Generalist raising at a $3B valuation; the market is separating Optimus-specific skepticism from sector momentum and still sees Tesla's robot timeline as slipping.