Markets may have just experienced their second DeepSeek shock, this time thanks to a Chinese AI lab named after a Pink Floyd album

The market is reacting to a margin narrative, not an immediate demand collapse: if frontier performance is being commoditized faster than expected, the first casualty is the premium multiple on closed-model AI vendors and the assumption that every incremental dollar of training spend will preserve pricing power. That is most bearish for firms whose equity story depends on durable model differentiation; it is less bearish for end-users that can arbitrage lower token costs into workflow automation and pricing leverage.

For semis, the selloff looks overstated on a 1-5 day horizon. Lower model prices can increase total inference volume, accelerate local/on-prem deployment, and pull forward more use cases, which is still GPU-intensive even if unit economics improve. The real falsifier is not benchmark bragging but 1-3 month hyperscaler capex commentary and order book tone; if those weaken, the trade shifts from sentiment to earnings damage.

The deeper second-order effect is competitive: Chinese open-weight models lower the barrier for domestic platforms and enterprise software to build AI features without paying U.S. API rents. That is a relative tailwind for China internet ecosystems with cloud distribution and balance-sheet flexibility, while U.S. software names that priced in AI exclusivity may see multiple compression. Consensus is likely underestimating how much cheap open models expand TAM, but overestimating how quickly Western enterprise customers will trust Chinese models for sensitive workloads; that compliance gap should slow, not stop, adoption.