Compute glut or not: unresolved for now, with a window set to check — resolves 2027-01-31
Compute glut or not: unresolved for now, with a window set to check — resolves 2027-01-31
Scheduled check: 2027-01-31
Background
The debate: much of the money flowing into AI data centers goes in a circle (chipmakers invest in customers, who use the money to buy chips back), and the argument is over whether that hides a real glut of computing capacity.
Why it matters: if a glut is being hidden, today's build-out means putting up data centers that nobody will use later.
Who is on each side: venture capitalist Bill Gurley says the financing structure obscures the demand signal; NVIDIA CEO Jensen Huang says there is no overcapacity at all.
The call
Our call: neither side has decisive evidence yet. SecondSource says plainly that the question cannot be settled now and sets a window for checking it, rather than presenting an answer that does not exist yet. The contradiction stays on record, with both claims left standing.
Sources behind the call
The receipt
Settled by whichever of two events comes first: ① neocloud operators (a newer generation of data-center companies that rent out GPUs for AI work) publish how fully their facilities are used; ② an unannounced glut of computing capacity occurs.
What would overturn it
Either side can be overturned: if utilization figures are disclosed separately from backstop financing (financing that promises to cover losses), Gurley's claim is void; if an unannounced glut actually happens, Jensen Huang's claim is void.