AI hardware demand is surging but power shortages are stranding billions in idle Nvidia gear at Amazon warehouses. Today's briefing covers TSMC's record margins, HBM's 121% price forecast, Samsung overtaking SK Hynix, CoreWeave bond stress, and AMD and Microsoft's challenge to GPU dominance.
Audio is available on Spreaker — see link below.
Amazon is receiving five billion dollars worth of Nvidia hardware every month and leaving it sitting idle in warehouses. That's the single clearest signal of where the AI infrastructure buildout stands right now.
The financing side of this story is tightening at the same time. CoreWeave's one-point-two-five-billion-dollar bonds due twenty thirty-two peaked at a fourteen-point-five percent yield this cycle.
While deployment slows at the edges, the foundry layer is holding firm. TSMC posted forty-one-point-two percent year-on-year revenue growth in Q3, reaching forty-six-point-seven-two billion dollars.
Within memory, the competitive picture shifted. Samsung generated an estimated eight-point-seven billion dollars in HBM revenue in Q3, surpassing SK Hynix's seven-point-six billion.
AMD's move this cycle cuts in a different direction. The Ryzen AI Max and PRO four hundred series brings up to sixteen Zen five cores, forty GPU units, a fifty-five teraflop NPU, and one hundred and ninety-two gigabytes of unified memory into a single enterprise processor.
Microsoft is pushing from the other direction. The Maia three hundred claims forty percent more tokens per megawatt than leading GPU systems.
The through-line here is straightforward. Chip demand is real, but deployment infrastructure, financing costs, and custom silicon are all compressing the timeline and the margin on which the current buildout depends.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.