AI Market Intelligence Daily

1. The single largest move of the cycle was in GPU spot pricing: a Kalshi contract asking whether the NVIDIA H200 compute price per hour would be above $6 at 4 PM ET on Sep 18 surged from 3 cents to 75 cents, a 2400. percent jump in roughly a day, with RTX 5090 contracts moving in the same direction (the above-$0.75 Sep expiry at 38%, the above-$0.70 contract from 25% to 50%). This is a sharp repricing of GPU scarcity: traders are betting that hourly compute prices spike over the next two weeks, consistent with the demand story behind Jensen Huang's explanation of 70 percent Nvidia growth and OpenAI halting Pro subscriptions to free capacity for Astra demand. If that repricing holds through Sep 18, it will be direct market evidence that frontier-lab demand is still outrunning GPU supply, not a bubble narrative.

2. OpenAI's IPO timeline got repriced decisively after Sam Altman said going public in 2026 would be 'ill-advised'. The contract 'Will OpenAI not IPO by December 31, 2026?' jumped from 92% to 97%, while the mirror 'Will OpenAI IPO by December 31 2026?' fell 52% from 8% to 4%, and 'OpenAI IPO before 2027?' dropped 35% to 6%. The move is coherent across the ladder: the $750B-$1T closing-cap contract nudged up to 2% while higher caps decayed to 0-2%, a pattern consistent with traders discounting a 2026 listing and repricing the eventual valuation downward. The 2027 window is where the action now sits, and the market still prices an OpenAI IPO before 2027 at a majority via the not-IPO-by-end-2026 complement.

3. Anthropic is the clear favorite to reach an IPO first: 'Anthropic IPO before 2027?' sits at 91%, and the contract asking which firm lists first prices Anthropic over OpenAI at roughly 94% to 6% on the Kalshi side. The news backdrop supports the split: Anthropic's CEO publicly outlined a plan to slow AI development and the company disclosed distillation campaigns by Alibaba, Moonshot, and DeepSeek, while OpenAI's leadership is actively ruling out a 2026 listing. The divergence between the two headlines - Anthropic planning a paced, controlled exit versus OpenAI deferring - is exactly what the market is pricing.

4. The model-leaderboard markets moved big but read more as position churn than conviction. 'Most popular AI this week?' for OpenAI exploded from 5% to 53% while Tencent fell from 93% to 53%,. Qwen's weekly market-share contract collapsed from 33% to 1%, and Google's above-18.4% share contract fell from 81% to 53%. The same week saw ChatGPT's 'Best AI this week' contract fall 50% to 2% while Claude held at 99%, and 'Top Coding AI this week' for Claude rose to 7%. These weekly contracts reset constantly, so the size of the swings should be discounted; the durable signal is that Claude is the market's pick for year-end leader at 69% (up from 63%) against Gemini at 6% and ChatGPT at 14%.

5. Usage and macro bets stayed calm. OpenRouter's weekly token-usage contract above 122T tokens jumped from 40% to 94%, a genuine confirmation that frontier-model traffic is still accelerating, not softening. Against that, 'AI bubble burst in 2026?' edged down to 11% and 'Will OpenAI release a top-ranked AI model before 2027?' slipped 20% to 33%, while the contract on NVIDIA becoming the largest company in the world by market cap next December held at 74%. The overall read: demand-side evidence (token volume, GPU spot prices) points up, while the market keeps trimming near-term AI optimism - a setup where the AI trade is being repriced as an infrastructure story rather than a speculative one.