AI Compute Sold Out as Data-Center Spending Estimate Falls

By Grace Kim

AI data-center spending may reach $3 trillion to $4 trillion through 2030, down from a prior $7 trillion projection, Hightower’s Stephanie Link told CNBC on November 11, 2026. “Maybe it’s 3 to 4 trillion,” Link said. “That’s still a lot of money.”

The forecast covers spending, not confirmed revenue. Lower estimates can coexist with scarce compute because customers finance hardware differently. Companies renting AI computing power still report sold-out capacity.

Can AI compute be sold out while spending falls?

CoreWeave reported third-quarter revenue of $3.66 billion, up 168.3% year over year. The result exceeded the $3.54 billion consensus estimate. Its loss was $1.27 per share, smaller than the expected $1.55 loss.

Revenue backlog reached about $132.5 billion. CoreWeave raised full-year guidance to $13.0 billion to $13.3 billion. On the earnings call, CEO Michael Intrator said: “We are largely sold out through 2027.”

CoreWeave’s results lifted AI-cloud stocks and Nvidia that evening, easing bubble concerns as investors weighed strong contracted demand against the company’s continuing financing needs. On Mad Money, Jim Cramer said CoreWeave’s CEO told him demand was 10 times oversubscribed for 2027. Added 2028 demand also reached multiples of available supply.

Cramer said the CEO would begin earning more next year from Nvidia chips purchased four years ago than during any of the prior four years. Prices for older Ampere and Hopper chips rose 30% to 50% quarter over quarter.

Financing remains the pressure point. CoreWeave plans to raise at least $4 billion through a high-grade loan. Its shares fell 23.4% year to date, despite a 30.8% gain over the previous week.

Is Oracle pricing a seller’s market?

Oracle’s fiscal first quarter produced cloud infrastructure revenue of $7.39 billion, up 121% year over year. Remaining performance obligations reached $664 billion.

Oracle booked more than $30 billion in new AI contracts during fiscal first quarter, ran GPUs at 97.9% utilization, and renewed expiring capacity at a 20% premium.

During the September 10 call, co-CEO Clayton Magouyrk said prices normally rise when demand exceeds supply. Oracle expects fiscal 2027 revenue of at least $90 billion.

Oracle shares were down 34.4% year to date.

Why does usable capacity take months?

Elon Musk said on November 11 that his operation needs about two months of round-the-clock work to make a large compute cluster reliable. He guessed other operators need closer to six months.

This gap matters because spending on buildings and spending on working compute are separate measures. Power availability, permits, and months of tuning restrict how quickly capital becomes usable capacity.

Vertiv sells the power and cooling equipment that helps close that gap. Its third-quarter results showed the deployment constraint: revenue rose 61.2% to $4.31 billion, while organic orders grew more than 100% year over year.

Full-year guidance calls for net sales of $14.45 billion to $14.85 billion and adjusted EPS of $7.10 to $7.20. CEO Giordano Albertazzi said time to first token matters a lot to customers.

Power availability, permitting, and skilled labor remain constraints. Vertiv shares rose 26.62% this year. Former Cisco CEO John Chambers told CNBC on November 11 that AI is here to stay and still has substantial room to develop.

Which figures will settle the debate?

Financing costs can pull spending forecasts lower while customers buy more hardware themselves. Oracle says the vast majority of its new contracts use prepayment or bring-your-own-hardware terms.

Power, permits, and months of cluster tuning still determine how much spending becomes usable compute. The decisive test is not the spending forecast alone, but whether power, financing, and customer purchasing structures convert into reliable, revenue-producing clusters on schedule.

Three indicators will show which narrative gains support:

• CoreWeave: Progress toward more than 1.85 GW of active power by year-end, plus pricing on its new high-grade loan.
• Oracle: Conversion of about half its $664 billion backlog into revenue within 36 months, as the company expects.
• Vertiv: Actual fourth-quarter sales against $4.31 billion to $4.51 billion guidance. Rising GPU renewal prices would still signal sold-out capacity.

If CoreWeave reaches its power target and Oracle converts backlog on schedule, usable capacity should grow faster; otherwise, longer deployment times and firm renewal prices will keep AI compute scarce.