November 3, 2026
Nvidia and SK Hynix hit fresh US-China export roadblocks, halting expansion and cutting off AI compute workarounds.
What happened
Nvidia is telling data center partners to limit chip sales bound for China and to cut off Chinese access to AI computing through overseas data centers, closing a loophole in US export controls. Separately, SK Hynix said its US expansion plans are stalled while it waits for US licenses tied to its Chinese factories. The combined news pressure moved roughly 500 billion dollars of semiconductor market value.
Why it matters
This double barreled tightening physically starves China of advanced AI chips and memory just as enterprises run more agents on core data, where every working agent needs paid inference. Nvidia loses a workaround revenue stream while a major memory supplier pauses its American buildout, which directly narrows the supply of high-bandwidth memory that our evidence shows will be short through 2030.
The case against
Defence stocks are already pricing in a permanently elevated geopolitical premium that may not persist if a diplomatic off-ramp appears. Nvidia could reroute some Chinese demand through licensed partners, and SK Hynix might eventually secure its permits, making today's delays temporary.
Our read
Our grounded evidence holds that demand for accelerators and high-bandwidth memory exceeds supply through 2030. Cutting off Chinese compute workarounds tightens that supply further for everyone else, while a construction halt at a top memory maker deepens the bottleneck.
What settles it
Whether the US Commerce Department issues SK Hynix's license or signals a broader waiver process, because that single decision lets expansion resume or keeps a critical memory supplier sidelined.