DeepSeek has released a preview version of its DeepSeek-V4 flagship model, claiming "world-class inference performance"...

DeepSeek has released a preview version of its DeepSeek-V4 flagship model, claiming "world-class inference performance" and significantly improved agent capabilities. The model is now available on web, app, and API, as per the official website source 1. This launch signals DeepSeek's continued investment in large language model development, which may drive increased demand for AI training and inference hardware. Separately, a prediction market on Manifold source 2 suggests low probability (4.01%) that DeepSeek misrepresented GPU compute for V3 training, implying general trust in their disclosure. While this is not a direct hardware event, the V4 preview likely accelerates compute requirements, benefiting AI infrastructure supply chain. The window for market impact is estimated at 30 days as adoption grows.