AI Market Intelligence Daily

1. Kalshi weekly share markets repriced toward Google and OpenAI, away from DeepSeek and Anthropic. The "Google market share this week" ladder climbed almost across the board, with the above-20.8% strike jumping from 1% to 7% and the above-19.5% strike rising from 50% to 92%, so traders are now pricing Google landing near 20.8% of weekly API share. OpenAI moved in the same direction, with the above-18.1% strike jumping from 34% to 85%. On the other side, every DeepSeek strike fell (the above-24.1% strike from 54% to 37%) and the above-4.5% Anthropic strike slipped from 42% to 52%, a modest uptick that still leaves Anthropic firmly last in the pack. The cleanest read: one week's share data is being settled with Google clearly ahead and the gap to the rest of the field widening.

2. Gemini has taken over the leaderboard, and the markets say so in both the weekly and the month-long books. The "Best AI this week" market is priced at 99% for Gemini, while the "What will be the top AI model on Oct 19" market sits at 96% on gemini-4-argon-high, near-certain pricing for a market that resolves in two days. The longer-dated "Best AI on Dec 31, 2026" book is more interesting: Gemini is at 40% and Claude at 45%, so the market sees a genuine two-way contest over the next two months, while Grok and ChatGPT are effectively written off at 2% and 10%. The Oct 2026 monthly book already prices Gemini at 76%, consistent with the weekly data, and the open-weights share of token volume at 60% for Oct 18 suggests the open model camp is holding roughly 60 cents of every dollar of routed traffic.

3. The OpenRouter token-usage ladder collapsed as the week of Oct 12-18 nears its final data points. Strikes above 168T fell from 91% to 5%, above 176T from 61% to 2%, and every strike at 180T and above is now priced at 1-2%, so the market is converging on actual weekly volume sitting below about 170T tokens. That is a meaningful data point for AI demand: the aggregate routing volume through OpenRouter is running below the level traders expected a week ago, with no headline in today's news explaining the drop. The closest corroboration is that Grok October app downloads are only 9% likely to clear 70M while Gemini (95%) and Claude (98%) clear their download bars comfortably, pointing to a traffic mix that is shifting toward the Google and Anthropic stacks.

4. Anthropic IPO pricing is the most internally consistent corner of the board, and the news is quietly feeding it. "Anthropic IPO before 2027" sits at 72% on Polymarket, matching the 72% on the Kalshi "announced before Jan 1, 2027" market, and "Will OpenAI or Anthropic IPO first" is at 91% for Anthropic on Kalshi. Meanwhile the OpenAI book is flat: IPO before 2027 at 3%, IPO by Dec 31 2026 at 4%, and the market-cap strike ladder centered on a $1.5T-or-greater close at 44%. TechCrunch's report that Replit is no longer working with Anthropic is the one headline that could bear on Anthropic's IPO story, yet the price did not move against it, which suggests traders are treating the Replit split as a distribution or product-friction issue rather than a revenue one. The divergence to flag: "Will Anthropic not IPO by December 31, 2027" rose from 3% to 4% while the before-2027 market held at 72% - the two are not mathematically contradictory, but the market is pricing in a thin sliver of Anthropic stalling out of 2026 entirely.

5. The macro risk markets barely moved, which is itself the signal. "AI bubble burst in 2026" crept from 7% to 7% (essentially flat) despite a news cycle full of model launches, and "NVIDIA the largest company in the world by market cap on Dec 31" held at 70%. The "U.S. enacts AI safety bill before 2027" market ticked from 6% to 6% - the +18% relative change looks large but it is off a 1-point base, so it is noise, not conviction. Tesla Optimus on sale before 2027 slipped from 10% to 6%, a small but notable de-rating even though the robotics news cycle (the Nvidia-VC pieces on physical AI) ran warm. The one genuine divergence between price and news sentiment: India's report of rogue AI-agent activity was the most regulatory-flavored headline of the day, yet the US AI-safety-bill market did not react, and no dedicated "rogue agent incident" market exists to absorb that news. The market is treating agent-incident news as a feature, not a risk.