TSMC weighs a dramatic move into 3D DRAM memory for Nvidia as AMD locks in a $5B Anthropic partnership and Chinese AI labs exploit a gaping export control loophole. Today's briefing covers the semiconductor stories reshaping the AI infrastructure race.
Audio is available on Spreaker — see link below.
TSMC is considering manufacturing its own 3D DRAM memory chips for Nvidia, and that single report sent the company's ADRs up nearly eight percent in a single session. That's a significant market reaction to what is, at its core, a strategic question: does the world's most important foundry expand horizontally into memory, or does it stay focused on what made it dominant?
The memory problem in AI infrastructure is structural, not cyclical. Training large models requires enormous, continuous data throughput between compute and memory.
Pull back to the broader market picture, and AMD had a strong quarter. Data center revenue hit eight point five billion dollars, up ninety-seven percent year-over-year.
Intel's situation is a sharper story. The stock fell twenty-nine percent in a single session after Q4 guidance of fifteen to sixteen billion dollars came in well below the seventeen billion the Street expected.
The China export control situation is evolving in a direction that U.S. policymakers should find uncomfortable. Chinese AI lab Moonshot has been pursuing leased Nvidia Blackwell GPU capacity in European data centers, targeting clusters of eight thousand or more GPUs for training and inference.
One number worth keeping in view: TSMC trades at eighteen times forward earnings. Nvidia trades at forty-one times.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.