NVIDIA Q2 FY2027 Earnings: Revenue by Segment and AI Growth Drivers

NVIDIA (NVDA) reported Q2 FY2027 revenue of $96 billion, more than double the figure from the same quarter a year earlier, marking the fourth consecutive quarter of accelerating growth. The result extended a streak that has made NVIDIA's data center segment the defining financial story across the [semiconductor sector](/groups/semiconductors).

Data Center Breakdown: Hyperscale and ACIE

Data center revenue reached $89 billion in Q2, up 18% sequentially from Q1. Within that total, hyperscale customers, principally large cloud providers, contributed $49 billion, a 13% sequential increase. The faster-moving segment was the AI Cloud and Infrastructure Enterprise category, or ACIE, which encompasses NeoClouds, enterprise AI deployments, and sovereign AI programs.

ACIE revenue came in at $40 billion for the quarter, up 25% sequentially and 138% year over year. That growth rate placed ACIE at roughly half of all data center revenue, a position it had not occupied in prior quarters. The gap between ACIE's 138% year-over-year growth and hyperscale's comparatively measured expansion reflects a broadening of the customer base funding AI infrastructure beyond the largest cloud operators.

Sovereign AI Accelerates Within ACIE

Sovereign AI programs, in which national governments fund domestic AI infrastructure, posted the sharpest acceleration within ACIE. Sovereign AI revenue tripled year over year, reflecting coordinated investment across multiple countries that treat AI capability as a strategic national priority. These programs tend to operate on multi-year procurement timelines driven by policy decisions rather than commercial demand cycles.

Enterprise AI adoption contributed the remaining ACIE growth, as corporations moved workloads from pilot deployments to production-scale infrastructure. The combination of sovereign programs and enterprise rollouts elevated ACIE from a smaller portion of data center revenue to parity with hyperscale within a single fiscal year, a shift that took fewer than four quarters.

Supply Constraints Cap Reported Growth Below Actual Demand

NVIDIA guided Q3 FY2027 revenue at $108 billion, plus or minus 2%. For the full fiscal year 2028, management projected revenue growth of approximately 70%. That figure carries a critical qualifier: the 70% represents a supply ceiling, not a demand ceiling. CEO Jensen Huang indicated that actual demand growth runs closer to 100%, with the gap between the two rates attributable entirely to manufacturing and supply chain constraints on GPU production capacity.

The distinction between supply-constrained and demand-constrained growth carries different analytical weight than a headline growth rate alone would suggest. When demand exceeds available supply, unfulfilled orders accumulate in backlog rather than disappear. Revenue grows at 70% while demand grows at 100% because production capacity is the binding constraint, not customer appetite. For the broader AI infrastructure cycle, this framing implies that the volume of capital committed to GPU procurement likely exceeds what quarterly shipment data currently reflects.

Gross Margin Headwinds: Trough Trajectory and Recovery Path

The transition to Blackwell-generation products introduced a gross margin headwind tied primarily to elevated DRAM prices embedded in the architecture's memory subsystem. NVIDIA guided Q3 gross margin to 74%, down from higher levels in prior quarters. Management projected a trough of approximately 71% to 72% in Q4 before a recovery to 72% to 73% in the following fiscal year, as DRAM supply normalizes and Blackwell ramp costs are absorbed.

The compression trajectory follows a pattern observed in prior NVIDIA generation transitions: margins compress during the initial production ramp and retrace as component costs normalize and manufacturing yields improve. The guided recovery to 72% to 73% in the next fiscal year would return gross margins to a range consistent with the company's historical post-ramp economics, meaning the trough is presented as a duration-limited phenomenon rather than a structural repricing.

Vera Rubin and the Revenue-Per-Gigawatt Expansion

The multi-year revenue picture is defined by how much dollar value each GPU generation can extract from a fixed unit of data center power. Hopper-based infrastructure generated approximately $18 billion per gigawatt of data center capacity. Blackwell raised that figure to $25 billion per gigawatt. The forthcoming Vera Rubin architecture, pairing the Vera CPU with the Rubin GPU, is projected to reach $40 billion per gigawatt, more than doubling the Hopper-era yield from the same physical footprint.

The performance gains underlying that revenue expansion are substantial. Vera Rubin delivers 30x higher throughput per megawatt and 35x lower token cost compared with Blackwell Ultra. For hyperscalers and enterprises running large-scale inference workloads, those efficiency ratios translate directly into reduced per-query operating costs, which in turn makes deploying additional capacity economically viable at a lower cost basis per unit of AI output.

The generational revenue-per-gigawatt progression, from $18 billion to $25 billion to $40 billion, also illustrates why hyperscalers continue to upgrade rather than pause at existing generations. Each successive architecture delivers a step-change in the economics of running AI workloads, not an incremental refinement, which sustains procurement motivation across product cycles rather than concentrating it within a single generation window.

AWS Deployment Plans Illustrate Hyperscaler Commitment Depth

Amazon Web Services (AWS) provided a concrete illustration of hyperscaler procurement depth. AWS announced plans to deploy an additional 2 million NVIDIA GPUs through fiscal 2029, with those deployments to include Vera CPUs paired with the Rubin GPU upon its availability. The scale of that commitment, spanning multiple hardware generations over several years of capital allocation, reflects the multi-year planning horizon that large cloud operators are applying to AI infrastructure.

The cross-generational nature of the AWS commitment is notable. These customers are not pausing Blackwell orders ahead of Vera Rubin's arrival; they are adding Vera Rubin commitments on top of existing Blackwell deployments. That pattern points to demand continuity across the product transition rather than the demand pause that typically precedes a new-generation ramp, and it supports the broader observation that supply rather than demand is the operative constraint on reported revenue growth.

Key Q2 FY2027 Figures

• Total revenue: $96 billion, more than double year over year
• Data center revenue: $89 billion, up 18% sequentially
• Hyperscale revenue: $49 billion, up 13% sequentially
• ACIE (NeoClouds, enterprise, sovereign AI) revenue: $40 billion, up 25% sequentially and 138% year over year
• ACIE share of data center revenue: approximately half
• Sovereign AI revenue: tripled year over year
• Q3 FY2027 revenue guidance: $108 billion, plus or minus 2%
• FY2028 supply-constrained growth guidance: approximately 70% (estimated demand growth near 100%)
• Q3 gross margin guidance: 74%; Q4 trough estimate 71-72%; next fiscal year recovery target 72-73%
• Revenue per gigawatt: Hopper $18B, Blackwell $25B, Vera Rubin $40B
• Vera Rubin vs. Blackwell Ultra: 30x throughput per megawatt, 35x lower token cost
• AWS incremental GPU deployment: 2 million units through fiscal 2029, including Vera Rubin generation

Full earnings call analysis is available in the Beta Finch [NVIDIA Q2 FY2027 episode](/podcasts/NVDA_Q2_2027). Additional coverage of the broader sector is available on the [semiconductors page](/groups/semiconductors).