Now Playing
Data Center Capacity Crisis: $750B Meets Concrete, Copper & HBM Shortages
Audio is available on Spreaker — see link below.
Half of planned US AI data center capacity won't arrive on time — and Nvidia's Rubin Ultra is already being resized due to HBM shortages. Today's briefing maps the physical and supply-chain bottlenecks reshaping the AI infrastructure race.
Audio is available on Spreaker — see link below.
Half the AI computing capacity the industry has scheduled for twenty twenty-eight isn't going to arrive on time. That's the clearest signal from the latest analysis of US data center construction, and it reframes everything you've heard about the hundred-and-fifty-billion-dollar annual capex commitments from big tech.
Here's what's driving the slippage. Transformer wait times have tripled.
The financial picture gets more complicated. Alphabet, Microsoft, Amazon, Meta, and Oracle collectively carry one point zero nine trillion dollars in future data center lease obligations that don't appear on their balance sheets as reported debt.
While the buildout stalls, the chips that would fill those data centers are hitting their own constraint. Nvidia has been testing lower-memory configurations of its Rubin Ultra accelerator over recent weeks.
Samsung's response to the HBM crunch points in a different direction. The company unveiled zHBM, a three-dimensional stacking architecture that places high-bandwidth memory directly on top of the GPU die rather than alongside it.
The geopolitical layer adds another complication. Both Samsung and SK Hynix have been evaluating etching tools from Chinese manufacturer AMEC for their China-based factories, a process that began roughly two years ago.
The near-term signals worth tracking are straightforward. Watch whether transformer and materials lead times start compressing, because that's the fastest indicator of whether twenty twenty-eight capacity projections get revised upward or downward again.
Chapter summary auto-generated from the verified script. Listen to the full episode for the complete content.