AI Market Intelligence Daily

1. Tencent and DeepSeek absorbed most of the AI share shift this week. The weekly model-usage share ladder settled with Tencent near 98% on its "above 6.1%" contract, while DeepSeek's "above 25.3%" line jumped from 11% to 60% in roughly a day. The rest of the ladder confirmed where the share went: Google's "above 18.8%" contract collapsed from 50% to 2% and OpenAI's "above 18%" line fell from 25% to 1%, leaving Anthropic's "above 2.65%" market already at 98% with no visible move. In aggregate the data reads as usage migrating from Western frontier models toward Chinese ones inside the tracked window, with the ladder's lower rungs now settling close to 99% as the week nears close. No top headline explains the share shift, which is itself a signal. The day's biggest AI stories — a Gemini model hacking other companies, an AI hallucination nearly triggering a US military operation, Trump floating a rebrand for "AI" and an "AI Force," and Anthropic running a biology lab — say nothing about weekly model usage. That gap between price moves and news flow is worth flagging: the market is pricing in a usage surge (consistent with a new cheap DeepSeek release drawing developer traffic) that the mainstream feed hasn't caught up to yet, so the moves are front-running the headlines rather than reacting to them. OpenAI's IPO cluster drifted lower across the board. "OpenAI not IPO by December 31, 2026" holds at 96% with the companion "IPO by December 31 2026" at 4%, and the pre-2027 markets sit at 3% — but the size-of-the-IPO ladder is the interesting part: "closing market cap above $800B" slipped from 96% to 93% and "above $1T" eased from 84% to 79%, implying the market still expects a listing eventually but is trimming the size of the valuation it attaches. That decoupling — timing nearly settled, valuation still compressing — suggests traders are pricing a late, possibly down-round-adjacent offering rather than a blowout debut. Anthropic's IPO timeline moved in the opposite direction, and the two largest Polymarket AI moves are in this cluster. "Anthropic IPO before 2027" fell from 78% to 72% in a day, the market that it will not IPO by end of 2027 rose from 3% to 4%, and the "$600B or greater at IPO" rung now sits at 96%. Meanwhile "Will OpenAI or Anthropic IPO first" (Kalshi) is at 94% for OpenAI-first despite OpenAI's own 2026 odds being 96% against — a small internal inconsistency that mostly reflects the two questions having different resolution dates. Read together: the market thinks Anthropic's window has slipped, but its eventual size is unchanged. Benchmark and capability markets show quiet repositioning around the next frontier score. On Text Arena, "Claude first to hit 1550" more than doubled from 9% to 19% and "ChatGPT first to 1550" rose from 6% to 15%, while the "any model 1520 before January" market sits at 32% and "any LLM 70% on Humanity's Last Exam" at 50% — a coin flip that traders clearly find genuinely uncertain. xAI's odds of a top-ranked model before 2027 doubled from 8% to 16% even with no Grok headline in the feed, the clearest divergence from news sentiment in today's report: something in private benchmark chatter or model rumors is moving that line ahead of any public story.