NVIDIA Q2 FY2027 Earnings: $96B Revenue, Vera Rubin Ramp, and Margin Headwinds

A Fourth Consecutive Quarter of Accelerating Growth

NVIDIA (NVDA) reported Q2 FY2027 total revenue of $96 billion, more than double year-over-year and a figure that marks the fourth consecutive quarter of accelerating growth at this scale. Data center revenue reached $89 billion, up 18% sequentially, as hyperscale customers and a rapidly expanding enterprise base both drew on NVIDIA's GPU supply.

The result continues a pattern in which demand has materially outpaced what NVIDIA can ship. Management characterized fiscal year 2028 revenue growth as approximately 70%, explicitly framing that figure as supply-constrained. Actual demand growth, per management commentary on the earnings call, is running closer to 100% year-over-year.

Hyperscale and ACIE: Two Distinct Growth Vectors

Within the $89 billion data center total, hyperscale revenue was $49 billion, up 13% sequentially. The hyperscale segment reflects orders from the largest cloud providers, whose capital expenditure cycles are well-documented. Amazon, for example, is deploying an additional 2 million GPUs through fiscal 2029, alongside new Vera CPUs, some of which are paired with the Rubin GPU.

The second component, which NVIDIA refers to as ACIE (AI Cloud Infrastructure and Enterprise, encompassing NeoClouds, enterprise customers, and sovereign AI programs), posted $40 billion in Q2, up 25% sequentially and 138% year-over-year. ACIE now represents roughly half of NVIDIA's total data center revenue, a share that stood far lower just two years ago.

Sovereign AI revenue tripled year-over-year, reflecting national-level commitments to build domestic AI infrastructure. These programs tend to involve longer procurement cycles and are less sensitive to short-term cloud cost optimization, which contributes to ACIE's relative stability as a revenue base.

Q3 Guidance and the Supply Gap

NVIDIA guided Q3 FY2027 revenue to $108 billion, plus or minus 2%. The sequential step from $96 billion to $108 billion represents continued expansion, but management was explicit that the guide reflects available supply rather than the full extent of customer demand. The gap between what can be shipped and what customers are requesting is a structural feature of this cycle, not a signal of softening end-market conditions.

Listeners to the NVIDIA Q2 FY2027 earnings podcast can hear management address this dynamic directly, including commentary on how allocation decisions across hyperscalers and ACIE customers are being managed. The full earnings discussion is available at /podcasts/NVDA_Q2_2027.

Gross Margin Headwinds: DRAM Costs Take Center Stage

The most significant near-term concern from the quarter is gross margin trajectory. NVIDIA guided Q3 FY2027 gross margin to 74%, with a further decline to a trough of 71-72% expected in Q4. The primary driver is a spike in DRAM prices, which affects the bill of materials for high-bandwidth memory used in NVIDIA's data center GPUs.

Management projected recovery toward 72-73% gross margin in the following fiscal year as DRAM pricing normalizes and product mix shifts toward newer architectures. The compression is therefore presented as transitional rather than structural, tied to a specific input cost cycle rather than a change in NVIDIA's pricing power or competitive position. For historical context on NVIDIA's margin profile across earnings cycles, the semiconductor earnings archive at /groups/semiconductors provides quarter-by-quarter comparisons.

The margin dynamic is worth contextualizing against the revenue scale. A 71-72% gross margin on $108 billion in quarterly revenue still implies gross profit of roughly $77-78 billion per quarter at the trough, a figure that most semiconductor companies will not approach in total annual revenue. The percentage decline is real, but the absolute gross profit base remains large.

Vera Rubin: Production Begins, Efficiency Claims Are Substantial

NVIDIA began production shipments of its Vera Rubin architecture during the same month as the earnings report. CEO Jensen Huang described it as the fastest product ramp in company history, with purchase orders already in place from all major hyperscalers. The transition from Blackwell Ultra to Vera Rubin is not merely a generational refresh; the efficiency claims are materially larger than prior generation-to-generation steps.

Vera Rubin delivers 30x higher throughput per megawatt compared with Blackwell Ultra, and reduces the token cost of inference by 35x. For data center operators whose capital expenditure and operating cost decisions are increasingly driven by energy constraints, those ratios carry significant weight in procurement discussions. A system that produces 30 times the useful output per unit of power consumption changes the economics of AI inference at scale.

The ramp timing also matters for NVIDIA's revenue trajectory. Vera Rubin commands a higher average selling price than prior generations and, as the revenue-per-gigawatt data below illustrates, enables substantially more revenue capture per unit of deployed infrastructure capacity.

The Revenue-Per-Gigawatt Flywheel

NVIDIA provided a framework for understanding how its revenue opportunity scales with data center capacity across chip generations. Under the Hopper architecture, one gigawatt of data center capacity corresponded to roughly $18 billion in GPU revenue opportunity. With Blackwell, that figure rose to approximately $25 billion per gigawatt. With Vera Rubin, the estimate reaches approximately $40 billion per gigawatt.

This progression illustrates a compounding dynamic. As data centers expand their total power footprint to meet AI workload demand, the revenue NVIDIA can capture per unit of that capacity keeps rising. The combination of global capacity expansion and a higher revenue yield per gigawatt means NVIDIA's addressable revenue grows faster than raw capacity additions alone would suggest.

The Vera Rubin ramp is therefore not just a product launch; it represents the mechanism by which NVIDIA monetizes the next wave of infrastructure build-out. If hyperscalers follow through on announced capacity expansions, including Amazon's 2 million incremental GPUs through fiscal 2029, the per-gigawatt revenue step-up from Blackwell to Vera Rubin translates into a meaningful expansion of NVIDIA's total addressable market within an already contracted customer base.

Key Figures at a Glance

• Q2 FY2027 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 revenue: $40 billion, up 25% sequentially and 138% year-over-year
• Sovereign AI revenue: tripled year-over-year; ACIE is now ~50% of data center revenue
• Q3 FY2027 revenue guidance: $108 billion, plus or minus 2%
• Q3 gross margin guidance: 74%; Q4 trough guidance: 71-72%
• FY2028 revenue growth projection: approximately 70% (supply-constrained; underlying demand ~100%)
• Vera Rubin vs. Blackwell Ultra: 30x throughput per megawatt, 35x lower token cost
• Revenue opportunity per GW: Hopper $18 billion, Blackwell $25 billion, Vera Rubin $40 billion
• Amazon incremental GPU deployment: 2 million GPUs through fiscal 2029

For a detailed discussion of these results, including management commentary on supply allocation, DRAM cost dynamics, and the Vera Rubin ramp timeline, listen to the NVIDIA Q2 FY2027 earnings episode at /podcasts/NVDA_Q2_2027.