Nvidia says hyperscalers will spend $1.3 trillion on AI infrastructure in 2027. That's the number to bid your data center backlog against
Nvidia's August 26 earnings beat came with a bigger number buried in the call: hyperscaler capex jumping from $800 billion this year to $1.3 trillion next. But the CEO also admitted Nvidia is chip-supply constrained, not demand constrained — which means GPU delivery, not your schedule, now paces data center fit-out.
Nvidia's CFO told analysts on the August 26 earnings call that the five biggest hyperscalers are on pace to spend $1.3 trillion on AI infrastructure in 2027, up from roughly $800 billion this year. For anyone bidding, staffing, or scheduling data center work, that 62% jump is the backlog number worth planning against — more than the moratorium and backlash headlines that have dominated data center coverage this month.
What did Nvidia actually report?
Nvidia posted fiscal Q2 2027 revenue of $96.2 billion, up 106% year-over-year, beating the roughly $92 billion analysts expected. Data center revenue — the segment that funds GPU clusters, power infrastructure, and cooling for the AI buildout — hit $89.0 billion, up 117% and now more than 90% of total sales. Non-GAAP earnings of $2.22 a share topped estimates near $2.08. For the current quarter, Nvidia guided revenue to $108 billion, plus or minus 2%, above the roughly $104 billion Wall Street expected and the first time the company has ever guided a single quarter past $100 billion. Gross margin held at 75% in the quarter but is guided down to about 74% next quarter as memory and wafer costs rise — the same input-cost pressure already showing up in electrical gear pricing. The guidance assumes zero data center compute revenue from China.
Why does the $1.3 trillion figure matter more than the model itself?
CFO Colette Kress's capex number is a demand forecast for physical infrastructure, not just chips. A meaningful share of that $1.3 trillion goes to the buildings, switchgear, generators, cooling plants, and site power upgrades that hyperscalers need before a single rack goes live — the work a GC, electrical sub, or mechanical contractor actually bids. Whatever the exact split, a 62% year-over-year jump in the spending pool behind data center construction is a stronger planning signal than a city council's moratorium vote or a single analyst's bubble call, because it's the client naming its own budget in front of investors.
What's the part most contractors will miss?
Buried in the same call is a more useful operating detail. Asked about next year's growth, Huang said Nvidia has "supply for 70% growth" but that "our demand is much higher than that." Translation: Nvidia's own chip output, not customer appetite, is the ceiling on how fast the buildout moves. That inverts the usual construction risk picture. Normally a GC worries about permitting, weather, and trade availability slowing down a schedule the owner has already committed to. On a data center job today, the building can be ready and the power energized, and the project still waits — because the GPUs the owner ordered haven't shipped yet.
| Traditional GC schedule risk | Chip-supply schedule risk |
|---|---|
| Weather, permitting, trade availability | GPU allocation date set by Nvidia's supply chain |
| GC controls the critical path to substantial completion | Owner's chip delivery window sets the fit-out and energization date |
| Retention tied to occupancy/certificate of occupancy | Final energization may lag occupancy by months, unrelated to construction quality |
| Schedule float negotiated with subs and vendors | Schedule float negotiated with the owner's chip supplier, outside the GC's contract |
What should an estimator or PM actually do with this?
- Ask the owner for the chip delivery window before you finalize a fit-out schedule. On mission-critical work, treat GPU allocation dates as a schedule input on par with switchgear lead time — because for the owner, it now is.
- Decouple shell/core substantial completion from IT-load energization in the contract. If the two milestones can slip independently, say so explicitly rather than letting a chip delay read as a construction delay.
- Size labor and specialty-commissioning staffing for a backlog that's growing 62% year-over-year, not flat. The demand signal from Nvidia's own numbers argues for locking trade labor and long-lead electrical gear now, not waiting for the next headline to decide.
- Watch the margin line, not just the revenue line. Nvidia flagged rising memory and wafer costs pressuring its own margins — the same commodity pressure that's already pushing up prices on the electrical and controls gear a data center job depends on.
The AI buildout's growth rate isn't in question after this earnings call — the constraint is. Knowing that the bottleneck sits in a chip fab rather than on your jobsite is the difference between blaming your own schedule for a slip that was never yours to control, and writing a contract that says so upfront.
For the physical build side of that same bottleneck — rack density and cooling load — see our look at Cerebras's new rack-scale chip and what it means for data center MEP.
Friday one chart. Every week, one piece of data that should change a decision on your project. Subscribe at constructionaibrief.com.
- What did Nvidia report for fiscal Q2 2027 on August 26, 2026?
- Revenue of $96.2 billion, up 106% year-over-year, with data center revenue of $89.0 billion, up 117%. Non-GAAP EPS came in at $2.22, ahead of estimates near $2.08. Nvidia guided next quarter's revenue to $108 billion, plus or minus 2% — the first time the company has guided a single quarter above $100 billion.
- How much will hyperscalers spend on AI infrastructure in 2027?
- Nvidia CFO Colette Kress told analysts on the August 26 earnings call that capital spending among the top five hyperscalers is on pace to hit about $1.3 trillion in 2027, up from roughly $800 billion in 2026 — a 62% jump in one year.
- Is Nvidia's AI chip business limited by demand or by supply?
- By supply, according to CEO Jensen Huang, who said Nvidia has "supply for 70% growth" in fiscal 2028 but that "our demand is much higher than that." Nvidia itself is the bottleneck on how fast the buildout can move, not customer appetite.
- Does this change how data center construction schedules should be planned?
- It should. If chip allocation, not building readiness, is the binding constraint, the fit-out and energization phase of a data center — racking, final electrical tie-in, commissioning — is increasingly paced to GPU delivery dates set by the chip supply chain rather than the GC's traditional substantial-completion milestone.
- Does Nvidia's Q3 guidance include revenue from China?
- No. Nvidia said its fiscal Q3 2027 guidance assumes zero data center compute revenue from China, reflecting ongoing U.S. export restrictions on advanced AI chips.