Compare

NVIDIA's Two Data Center Legs: Hyperscale at $60.1 Billion Versus AI Cloud and Enterprise at $47.5 Billion

Inside NVIDIA's $107.6 billion data center quarter, hyperscale compute platforms contributed $60.1 billion, up 71% year over year, while AI cloud, industrial and enterprise contributed $47.5 billion, up 195%. The growth rates differ by nearly a factor of three. This comparison breaks down the two legs across five dimensions: who is buying, what the equipment is for, how durable the growth is, and why the gap should widen again next quarter.

Key takeaway

Split NVIDIA's data center business in two: hyperscale at $60.1 billion, up 71%, and AI cloud plus industrial and enterprise at $47.5 billion, up 195%. The first leg is incumbents expanding; the second is new buyers arriving. The new buyers grow faster, but they depend far more on financing conditions and inference demand.

Why split it at all ​

NVIDIA's data center revenue of $107.6 billion, up 110% year over year, obscures something more important: the buyer mix has changed.

Per the company's CFO commentary, data center splits into two categories:

  • Hyperscale compute platforms: $60.1 billion, up 71% year over year, about 51.9% of total revenue.
  • AI cloud, industrial and enterprise: $47.5 billion, up 195%, about 41.0% of total revenue.

The two add to $107.6 billion, but the growth rates differ by nearly a factor of three. That is not a statistical footnote; it is the direction of this compute cycle.

Comparison table ​

DimensionHyperscale compute platformsAI cloud, industrial and enterprise
Revenue this quarter$60.1B$47.5B
Year-over-year growth+71%+195%
Share of total revenue~51.9%~41.0%
Prior-year quarter (back-solved)~$35.1B~$16.1B
Typical buyersHyperscale cloud providers with owned data centersNeoclouds, AI-native companies, industrial firms, sovereign AI projects
Purchase motivationCapacity expansion for existing business, inference loadNew capacity build-out, outsourced training and inference, private deployments
Funding structureMostly own cash flowHeavier use of financing, leases and customer prepayments
Bargaining powerStrong, multi-year framework agreementsWeaker, order sizes volatile
RiskCapex cycle peakingFinancing conditions, project delivery, poor utilization

Difference 1: one is expansion, the other is entry ​

The $60.1 billion hyperscale leg is essentially the established cloud providers adding to existing capex plans. Their purchases typically serve two purposes: replacing older GPU generations and expanding capacity for inference traffic.

The $47.5 billion AI cloud and enterprise leg is different. Buyers include neoclouds such as CoreWeave, Nebius and Nscale, AI-native companies buying compute through clouds, industrial firms and sovereign AI projects. Most are not paying from operating cash flow but from financing, prepayments and leases.

In one sentence: incumbents with cash keep buying, and new entrants borrowing money have started buying. That determines both the growth gap and the risk gap.

Difference 2: the 195% contains one-off components ​

A 195% growth rate cannot persist. Three reasons:

A low base. The prior-year quarter was roughly $16.1 billion, less than half the size of the hyperscale leg. A low base magnifies growth.

Concentrated project deliveries. Many neocloud and sovereign AI projects came online during 2026, releasing purchases within a few quarters.

Inference demand is only starting. Training-side purchases are relatively stable; inference is just beginning to scale. Once inference becomes the main driver, purchasing shifts from project-based to usage-based and growth naturally moderates.

So 195% is better read as the speed at which new buyers are entering, not as a long-run growth rate.

Difference 3: next quarter's margin guidance touches both legs ​

Fourth-quarter gross margin guidance falls to 71%-72% from 74.8% this quarter. The company attributes this to the early Vera Rubin ramp.

Structure matters too. New buyers — neoclouds and AI-native companies — are considerably more price-sensitive than hyperscalers. They favor full-rack, highly integrated configurations and are willing to commit to longer contracts in exchange for discounts. That further compresses unit margin.

In other words, the margin decline is not only about a new platform costing more; it is also about the customer mix moving toward the price-sensitive end.

Difference 4: the commitment book determines durability ​

NVIDIA disclosed three sets of figures: $390 billion in future supply and capacity purchase commitments (up from $279 billion), $102 billion in cloud service agreements and not-yet-commenced data center leases (up from $74 billion), and $118 billion in maximum guarantee exposure.

Most of these correspond to the AI cloud and enterprise leg, because hyperscalers typically carry their own data center and power costs and do not need NVIDIA credit support. The $105 billion guarantee tied to the SB Energy Ohio project, with OpenAI's 20-year lease, is a textbook sample from that leg.

Conclusion: the faster-growing leg is also the leg that requires NVIDIA to provide credit and lease support. That is both proof of demand and a concentration of risk.

Difference 5: sustainable growth rates differ ​

DimensionHyperscaleAI cloud, industrial and enterprise
PredictabilityHigh, multi-year capex plansMedium to low, project-based and lumpy
Sustainable growth bandMid-teens to roughly 30%High teens to high growth, but volatile
Sensitivity to financing conditionsLowHigh
Sensitivity to inference scalingHighVery high

If inference becomes the main driver in 2027, both legs benefit, but differently: hyperscale through utilization and replacement cycles, AI cloud through project count and project size.

Conclusion in three sentences ​

First, NVIDIA's growth story is shifting from four customers to a crowd. That is good, because lower customer concentration means more cycle resilience.

Second, customer quality is declining. Hyperscalers pay from their own cash flow; neoclouds pay from financing and long-term contracts. Their ability to pay and their project completion rates are the biggest uncertainties in that leg.

Third, margin is the price tag both legs share. The 71%-72% guidance reflects not only the new platform ramp but also the customer mix moving toward price-sensitive buyers. That number says more about the economics of this cycle than the revenue growth rate does.

What cannot be compared directly ​

  • The $60.1 billion and $47.5 billion split comes from the company's CFO commentary; the press release does not break it out. The two categories sum to the reported $107.6 billion data center revenue, but the definition is the company's and may change over time.
  • Prior-year figures of roughly $35.1 billion and $16.1 billion are approximations back-solved from disclosed growth rates, not reported numbers.
  • Buyer categorization is an industry judgment based on public information; the company does not disclose customer lists by category.
  • Purchase commitments, lease commitments and guarantee exposure cover the whole business and cannot be attributed entirely to either leg.
  • This article makes no valuation judgment and does not forecast the share price.

Sources ​