(00:00:00) DeepSeek's Chip Move, H200 Rationing & TSMC's Slowing Growth
(00:01:05) China's H200 Rationing Logic
(00:01:58) ASML EUV Smuggling Allegation
(00:02:41) China's Domestic Chip Architecture Push
(00:03:23) Samsung Gaia and TSMC's Slowdown
(00:04:14) Watchpoints and Closing Signal

DeepSeek has confirmed it is building its own AI inference chip — a strategic shift from software efficiency to semiconductor independence that could reshape how China's leading AI lab competes long-term. The move raises immediate questions about foundry access, manufacturing node constraints, and the five-to-seven-year timeline separating a hiring announcement from a competitive product.

Running alongside that long-term bet is a short-term policy concession: Beijing is expected to approve limited H200 chip imports for Alibaba, ByteDance, and DeepSeek, capped below two hundred thousand units — roughly half of what was requested. Chips will be restricted to training workloads only, a deliberate phased strategy that manages compute dependency rather than eliminating it.

On the enforcement front, US Commerce Secretary Lutnick pressed ASML executives directly on allegations that a complete EUV lithography machine reached China through underground channels. ASML denied it with tracking data. The US declined to share its evidence publicly. That standoff reached cabinet level — a sign this is not an isolated concern.

Three Chinese startups — Suanmiao, Transtreams, and Dongfang Suanxin — completed tape-outs on three-dimensional stacked AI chips using domestic processes, raising hundreds of millions of yuan to pursue architectures that work around US export controls. Tape-out is not production, and thermal engineering at scale remains unsolved.

Samsung's System LSI division is developing Gaia, a four-nanometre AI PC chip targeting mass production in 2027. Meanwhile, TSMC's April-to-May revenue grew just twenty-four percent year-over-year — well short of the thirty-five percent analysts expected. CEO Wei's second-half guidance will be the decisive signal on whether the AI infrastructure capex cycle is peaking or recalibrating.

This episode includes AI-generated content.