Nvidia's most advanced export-controlled chips reached China via a shell company — and both Commerce and Nvidia missed it. Plus: TSMC's three-dimensional DRAM report, Intel's substrate bottleneck, and a 49-gigawatt U.S. data centre power shortfall.
Audio is available on Spreaker — see link below.
TSMC's stock jumped six percent this week on a single unconfirmed report: the world's dominant logic foundry is developing three-dimensional DRAM chips, apparently at Nvidia's request. No timeline.
The signal here isn't that TSMC has decided to become a memory company. It's that Nvidia is desperate enough to push them toward it.
Intel is running a parallel race with its own supply chain problems. The company is aggressively shifting wafer production toward Xeon six Granite Rapids server chips to meet surging AI infrastructure demand, backed by twenty billion dollars in annual capital expenditure targeting new fabs in Arizona and Ohio.
Then there's the story that carries the most direct policy weight. An investigation found that Nvidia's H200 accelerators, its most advanced export-controlled chips, reached China through a company called New Matrix.
One more constraint is rising to equal footing with chip availability. Morgan Stanley puts the U.S. data center power shortfall at forty-nine gigawatts by twenty twenty-eight.
What to watch from here: whether TSMC names a production timeline or capex commitment for three-dimensional DRAM, because right now it remains a market-moving rumour with no operational detail behind it. And whether U.S. regulators respond to the New Matrix case with any concrete enforcement reform, or whether the silence continues.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.