A Chinese Pink Floyd fan is giving Claude and Chat their own DeepSeek moment — an AI model just as good and half the price

The market’s first reflex will be to treat this as a binary “China caught up” headline, but the more durable effect is pricing pressure at the model layer, not an immediate collapse in the AI capex trade. If frontier capability is increasingly available in open source at half the price, the value capture shifts away from raw model access and toward distribution, proprietary data, workflow lock-in, and inference efficiency. That is a headwind for premium API monetization, but it is also a tailwind for application builders that can swap in cheaper models and expand usage without blowing up COGS.

For public names, NVDA is less vulnerable than the headline suggests because cheaper models tend to increase token consumption and inference load; the bigger risk is regional mix, not global demand. The real loser is any software business whose AI feature set is indistinguishable from a commodity model wrapper. AAPL is a more subtle beneficiary if on-device and edge inference become more viable, since that strengthens the case for AI features that do not depend on expensive cloud calls, but that is a 6-18 month story rather than an immediate earnings driver.

Contrarian view: the consensus is over-indexing on model parity and underweighting trust, deployment, and enterprise integration. Open-source Chinese models can compress headline valuations, but they do not automatically displace U.S. incumbents in regulated enterprise workflows. The falsifier for the bearish AI-software thesis is simple: if usage and retention at AI-native apps keep rising even as token prices fall, the pie is expanding faster than margins are compressing.