AI Spending May Hit $3T–$4T While Compute Stays Sold Out
Hightower now sees $3 trillion to $4 trillion in data-center spending through 2030, while CoreWeave remains largely sold out through 2027 and Oracle reports $664 billion in obligations.

Why it matters
- Hightower's Stephanie Link said on November 11, 2026 that data-center spending through 2030 may be $3 trillion to $4 trillion, not $7 trillion.
- CoreWeave reported third-quarter revenue of $3.66 billion, up 168.3%, $132.5 billion in backlog and a largely sold-out position through 2027.
- Oracle reported 121% cloud-infrastructure revenue growth to $7.39 billion, $664 billion in obligations and more than $30 billion in new AI contracts.
- Vertiv reported third-quarter revenue of $4.31 billion, up 61.2%, and full-year sales guidance of $14.45 billion to $14.85 billion.
- Checkpoints include CoreWeave's 1.85 GW year-end target, Oracle's 36-month backlog conversion and Vertiv's $4.31 billion-to-$4.51 billion fourth-quarter guidance.
By Marcus Bennett
AI data-center spending may total $3 trillion to $4 trillion between November 2026 and 2030, rather than the previously projected $7 trillion, Hightower's Stephanie Link told CNBC on November 11, 2026.
“Maybe we're not going to spend $7 trillion on data centers between now and 2030. That was what we were projecting. Maybe it's 3 to 4 trillion. That's still a lot of money,” Link said.
Even the lower range leaves trillions of dollars of expected data-center investment, according to Link. Recent results from CoreWeave, Oracle and Vertiv showed that scarce usable capacity can persist alongside a reduced spending forecast. Both trends can be true.
24/7 Wall St. framed the market stakes around conversion rather than appetite: financing costs, customer-funded hardware, power availability, permits and months of cluster commissioning determine how quickly spending becomes working compute.
What does CoreWeave's backlog show?
CoreWeave's third-quarter results set off a rally in AI cloud stocks and lifted Nvidia, according to 24/7 Wall St.; the publication linked the numbers to easing bubble fears.
That evening on Mad Money, Jim Cramer said CoreWeave's CEO told him: “They are 10 times oversubscribed for 2027 and the added demand for 2028 is also multiples of what he can provide.”
Cramer relayed a second detail: “starting next year he'll begin earning more on Nvidia chips that he bought four years ago than he has been for any of the past four years.”
CoreWeave's third-quarter release supplied the operating counterpart to Cramer's account:
- Third-quarter revenue rose 168.3% year over year to $3.66 billion, above the $3.54 billion consensus estimate, according to CoreWeave.
- The company lost $1.27 per share, compared with the expected loss of $1.55, according to the release.
- Revenue backlog reached about $132.5 billion, and CoreWeave raised full-year revenue guidance to $13.0 billion to $13.3 billion.
- Prices for older Ampere and Hopper chips rose 30% to 50% quarter over quarter, according to CoreWeave's results.
“We are largely sold out through 2027,” CoreWeave CEO Michael Intrator said on the earnings call.
Funding remains the pressure point. CoreWeave plans to raise at least $4 billion through a high-grade loan, or investment-grade borrowing, according to 24/7 Wall St.
The report showed CoreWeave shares down 23.4% year to date despite a 30.8% gain over the past week. Those figures leave two market questions visible at once: contracted demand exceeds available supply, while funding remains an investor concern.
Why does Oracle still have pricing power?
Oracle's fiscal first quarter showed the same combination of high demand and constrained supply. The utilization rate matters.
- Oracle's cloud infrastructure revenue rose 121% to $7.39 billion.
- Oracle's remaining performance obligations, a measure of contracted revenue not yet recognized, reached $664 billion.
- The company booked more than $30 billion in new AI contracts.
- Oracle ran its GPUs (graphics processing units) at 97.9% utilization.
- Oracle renewed expiring capacity at a 20% price premium.
On the September 10 call, co-CEO Clayton Magouyrk said: “Prices in a world where demand exceeds supply typically do not go down. They do go up.”
Oracle expects fiscal 2027 revenue of at least $90 billion, according to the company. The same 24/7 Wall St. report showed Oracle shares down 34.4% year to date.
Why does usable capacity take months?
Musk supplied the clock. On November 11, Elon Musk posted on X: “‘Actually usable’ is an important metric. It takes us ~2 months working 24/7 to get a large compute cluster working reliably. I'd guess closer to 6 months for others.”
24/7 Wall St. used Musk's estimate to separate spending on buildings from working computing power; that delay keeps delivered capacity tight.
Vertiv sells the power and cooling equipment that helps close that gap, according to 24/7 Wall St. Deployment remains slower than spending.
- Vertiv's third-quarter revenue rose 61.2% to $4.31 billion.
- Vertiv's organic orders, a measure excluding acquired operations, grew more than 100% year over year.
- The company's full-year guidance calls for net sales of $14.45 billion to $14.85 billion and adjusted EPS (earnings per share) of $7.10 to $7.20.
Giordano Albertazzi said: “Time to first token matters a lot to our customers.” A token is one unit of generated text. He listed power availability, permitting and skilled labor as constraints, according to the company.
24/7 Wall St. reported Vertiv shares up 26.62% this year. The publication classified Vertiv as an infrastructure supplier benefiting from the buildout rather than a chipmaker, and separately covered seven such companies across power, cooling and networking.
On November 11, former Cisco CEO John Chambers told CNBC: “AI is here to stay. We are in the second inning of it.” The 24/7 Wall St. analysis framed Chambers's view as an early-inning assessment of AI's market.
What can reconcile lower spending with sold-out compute?
Money does not equal capacity. 24/7 Wall St. said spending forecasts can come down as financing costs rise and customers pay for hardware themselves.
Oracle says the vast majority of its new contracts use prepayment or bring-your-own-hardware terms, in which the customer supplies the equipment. Power, permits and months of cluster tuning, meaning configuration and testing for reliable operation, limit how fast any dollar becomes working compute, according to 24/7 Wall St.
Which figures will settle the debate?
The analysis identified three operating checkpoints:
- CoreWeave: Whether the company reaches its target of more than 1.85 gigawatts (GW) of active power by year-end and how its new loan is priced.
- Oracle: Whether about half of its $664 billion backlog converts into revenue within 36 months, as the company expects.
- Vertiv: Whether actual results meet fourth-quarter sales guidance of $4.31 billion to $4.51 billion. Continued increases in GPU renewal prices would indicate that capacity remains sold out, according to 24/7 Wall St.
The next test will come from execution rather than another spending estimate. CoreWeave's year-end power target, Oracle's 36-month conversion rate and Vertiv's fourth-quarter sales will show whether financing and deployment delays still constrain usable compute.
A lower spending forecast does not guarantee cheaper AI capacity. If renewal prices remain elevated while power and backlog conversion lag, operators can remain sold out under the $3 trillion-to-$4 trillion case, according to 24/7 Wall St.
Source: 247wallst.com
More from Marcus Bennett
Show full bio
Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
539 articles
Related articles
- Enterprise AI Network Spending Surges Before Readiness Catches Up
- AMD's Q3 Beat and $17B Guide Calm the AI Spending Scare
- Goldman Sachs Sees $1.2 Trillion Big Tech AI Spending Wave by 2027
- Jim Cramer Warns Off the AI Data Center Trade as Nvidia and Broadcom Split
- Anthropic maps $518bn infrastructure spend as IPO backers eye $2tn valuation