Roughly half of all announced AI data centers may never get built on time. That is the lead fact from this panel discussion at the Big Technology AI Summit, featuring Anissa Gardizy of The Information, Max Cherney of Reuters, and Lauren Goode of Wired, three reporters with direct sourcing inside the infrastructure buildout. The conversation covers Nvidia's dominance, the shift from training to inference, and whether the greatest capital expenditure boom in tech history can deliver returns before the money runs out.
The structural risks here are specific and underreported. Taiwan's stability is called the industry's most underappreciated single point of failure. OpenAI is actively working to reduce its dependence on Nvidia. Jensen Huang is lobbying to resume chip sales into China. The debate over in-house silicon versus merchant chips is live and unresolved. These are not hypotheticals. They are decisions being made now by the companies spending hundreds of billions of dollars on this infrastructure.
The reason to read or watch the full conversation is not the conclusions. It is the reporters' sourcing on where the cracks are forming: overbuilding risk, a potential chip-industry bust, and the unresolved question of whether more compute is actually producing better models. The lightning round on that last point alone is worth your time.
[WATCH ON YOUTUBE →]