Now Playing
TSMC Bottleneck Breaks Open: Intel, Nvidia & HBM4 Reshape the AI Supply Chain
Audio is available on Spreaker — see link below.
TSMC's CoWoS capacity is sold out indefinitely, and the industry is moving fast: Google commits 3M TPU chips to Intel's foundry, Nvidia tests Intel 18A, and SK Hynix locks in HBM4 across all four next-gen Nvidia platforms. From Broadcom's $35B infrastructure play to AMD's UK sovereign AI push, today's briefing maps the accelerating shift away from single-supplier dependency in AI hardware.
Audio is available on Spreaker — see link below.
Google has committed to over three million TPU chips from Intel's foundry through twenty-twenty-eight. That single order represents roughly half of Google's projected TPU output for that year, and it's the clearest signal yet that Intel's foundry ambitions have moved from aspiration to execution.
The Google order doesn't stand alone. Nvidia is actively running multi-project wafer tests on Intel's eighteen-A process node, evaluating it for a four-die GPU design tied to its next-generation Feynman architecture.
TSMC's first quarter results clarify the pressure driving all of this. Revenue came in at thirty-five-point-nine billion dollars, gross margins at sixty-six-point-two percent, and net income up fifty-eight percent.
SK Group's deepening partnership with Nvidia adds another layer to this story. SK Hynix is supplying HBM4 memory across all four of Nvidia's next-generation platforms, using an MR-MUF bonding process targeting sixteen-layer stacks in the second half of twenty-twenty-six.
AMD is moving on a different axis. The company announced a multi-billion pound initiative across the UK, partnering with Dell, Cambridge, Imperial College, and the UK Atomic Energy Authority.
Broadcom, Apollo, and Blackstone have launched a thirty-five billion dollar platform targeting over twenty gigawatts of global AI infrastructure deployment. The first tranche funds Anthropic's capacity expansion via Fluidstack, targeting over one gigawatt starting mid-twenty-twenty-six.
The through-line across today's developments is straightforward. TSMC's bottleneck is real and confirmed.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.