Financial Performance and Outlook

FY27Q2 Financial Performance

• Revenue: $96.2B, up 106% YoY and up 18% sequentially, exceeding prior guidance of $91B (midpoint).
• Data Center revenue: $89.0B, up 117% YoY and up 18% sequentially.
  • Hyperscale: $48.7B, up 13% sequentially.
  • ACIE (AI Clouds, Industrial & Enterprise): $40.3B, up 25% sequentially and up 138% YoY.
• Edge Computing revenue: $7.2B, up 27% YoY and up 13% sequentially.
• GAAP and non-GAAP gross margins: both 75.0%, up 2.5-2.6 pts YoY and approximately flat sequentially.
• Non-GAAP operating income: $64.0B, up 124% YoY.
• Non-GAAP diluted EPS: $2.22, up 120% YoY.
• Free cash flow: $21.3B in Q2; $69.9B YTD through Q2.
• China data center compute revenue: less than 1% of Data Center revenue in Q2.

Gross Margin Reset — Key Issue for the Quarter

• Q3 FY27 gross margin guidance: 74% (GAAP and non-GAAP), down ~100 bps from Q2's 75%.
• Q4 FY27 gross margin guidance: expected to bottom in the 71-72% range.
• FY28 gross margin guidance: expected to recover to 72-73% as executed price increases take effect in Q1 FY28.
• Driver: memory prices have risen to levels that "exceeded prior expectations" and are expected to increase further. Memory scarcity is management's primary stated driver.
• This is a reset from NVIDIA's prior "mid-70s" gross margin target that held through the Blackwell ramp.

FY27Q3 Guidance

• Revenue: $108.0B, plus or minus 2%.
• Growth expected to be driven primarily by ACIE, with Hyperscale growth expected to reaccelerate in Q4 and into FY28 as Vera Rubin supply grows.
• Vera Rubin expected to represent approximately 20% of Data Center revenue in Q3.
• GAAP and non-GAAP gross margins: 74%, plus or minus 50 bps.
• GAAP operating expenses: ~$9.2B; non-GAAP: ~$9.0B.
• No China data center compute revenue assumed in guidance.

FY28 Preliminary Guidance

• Revenue expected to grow approximately 70% YoY — the first time NVIDIA has provided a full-year forward outlook.
• Management explicitly characterized this as supply-constrained; unconstrained demand growth would be materially higher.
• Supply expected to remain a bottleneck "at least through the end of fiscal year 2028."
• CPU revenue expected to more than double in FY28, positioning NVIDIA as "one of the world's leading server CPU suppliers."
• Full-year FY27 non-GAAP OpEx expected to grow in the low 50s YoY.

Business Drivers

Demand Inflection: Agentic AI and the Compute-as-Revenue Equation

• Jensen Huang stated that AI has crossed a threshold: most AI workloads are now agentic rather than single-shot, requiring orders of magnitude more compute per task. Agents spawn sub-agents, use tools, and reason across extended contexts — each step consuming GPU cycles.
• Management estimates that for agentic workloads, compute required per user interaction is 15 to 100 times higher than traditional one-shot inference, depending on task complexity.
• Global VC funding in AI exceeded $400B in the first half of 2026, surpassing the full-year 2025 total of $265B. Nearly 20 AI native companies now exceed $1B in annualized revenue (up from 13 in Q4 FY27Q1), with vertical enterprise software logging the fastest growth.
• Sovereign AI revenue grew 35% sequentially and more than tripled YoY in Q2.

Vera Rubin: Production and Early Adoption

• Vera Rubin entered full production and commenced shipments this month (August 2026). Purchase orders have been received from every major hyperscaler, AI cloud, and system OEM.
• Management expects Vera Rubin to be the fastest product ramp in NVIDIA's history. Every frontier model company is committed from day one, unlike the Blackwell transition.
• Vera Rubin delivers 30x higher throughput per megawatt and 35x lower token costs versus Grace Blackwell Ultra. Revenue opportunity per gigawatt increases from $25B (Blackwell) to $40B (Vera Rubin).
• Vera Rubin is expected to represent approximately 20% of Data Center revenue in Q3 FY27.
• Google's XGS instances can support up to 960,000 Rubin GPUs across multiple sites. AWS is deploying an additional 2 million GPUs (Blackwell and Rubin) starting this quarter through Q2 FY29.

Vera CPU: Standalone Ramp and New TAM

• The Vera CPU is in full production as a standalone product. Management has visibility to nearly $20B in standalone Vera CPU revenue in FY27, and expects CPU revenue to more than double in FY28.
• Jensen Huang framed Vera as opening a $200B TAM NVIDIA has never previously addressed — agentic orchestration, tool use, IO, and memory management all run on CPUs.
• Vera is being deployed by every major hyperscaler and system maker; AWS is a new adopter starting this quarter.
• Vera standalone CPU revenue is not included in the $1T Blackwell and Rubin GPU revenue outlook, making it one of the largest incremental upside vectors.

ACIE Segment: NeoCloud and Sovereign AI Leading Growth

• ACIE revenue of $40.3B grew 25% sequentially and 138% YoY — faster than Hyperscale in both periods.
• NeoCloud partners are expected to exit FY27 with 8 GW of total installed capacity, up from approximately 3 GW at end-2025. NeoCloud capacity additions are described as the primary driver of sequential ACIE growth in Q3.
• Revenue sharing structure: NVIDIA is offering take-or-pay commitments (minimum revenue guarantees) to help NeoClouds secure financing for facilities. In exchange, NVIDIA earns a share of rental revenue generated above the guarantee floor. Jensen Huang described this as "getting paid twice" — on the hardware sale and on recurring usage revenue.
• Hyperscaler customer demand also spills into ACIE as large hyperscalers purchase NeoCloud capacity to supplement their own build-outs.

Memory Price Inflation: The Driver of the Margin Reset

• "Extreme pricing conditions in memory" are the primary driver of the gross margin reset. Management stated that price increases have "exceeded prior expectations" and are "headed even higher into next year."
• Memory scarcity is itself driven by the AI buildout: the same demand surge that drives NVIDIA's revenue is tightening memory supply. Management framed this as a symptom of the overall demand environment, not a structural cost impairment.
• NVIDIA has deep relationships with all three major memory suppliers and is working to increase capacity aligned with its roadmap. Executed price increases take effect in Q1 FY28, which management expects to drive margin recovery.

Networking: Record Quarter

• Networking revenue grew 18% sequentially to a record level in Q2. Spectrum-X Ethernet grew 2.6x YoY.
• NVIDIA now describes itself as the largest and fastest-growing network company in the world.
• Spectrum-6 switch systems (pluggable and co-packaged optics) are arriving across gigascale AI factories as part of the Vera Rubin platform.
• Groq 3 LPX (the interactive AI inference accelerator) is in full production and is expected to ship in volume later in Q3 FY27. Management characterized LPX as a niche, premium-token product bolt-on for a subset of providers; Nebius is the first adopter.

Ecosystem and Frontier AI Lab Investments

• NVIDIA has invested nearly $50B in frontier AI labs (OpenAI, Anthropic, xAI, Mistral, and others). Jensen Huang explicitly stated his only regret is "not investing more and sooner."
• Six infrastructure capital providers — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR — have formed financing platforms expected to mobilize over $500B of third-party capital for AI infrastructure buildout.
• The PORTS-Pike campus partnership with SB Energy (4.25 GW, Ohio) is the largest single commitment: NVIDIA's guarantee exposure is capped at $105B. OpenAI has committed to the facility under 20-year leases, and each infrastructure generation could represent ~1.5 million NVIDIA GPUs or $150-200B in NVIDIA revenue.

One-Offs

Gross Margin Reset and Memory Inflation

• Management proactively reset gross margin expectations, guiding Q3 to 74% and Q4 to bottom in the 71-72% range, before recovering to 72-73% in FY28. This is materially below the mid-70s target the company maintained throughout the Blackwell cycle.
• The driver is extreme HBM and DRAM pricing conditions driven by the AI buildout itself tightening memory supply. The magnitude of the increase exceeded management's prior expectations.
• Investors should expect gross margin comparability issues across FY27: Q4 will be the trough, and FY28 recovery depends on NVIDIA's ability to execute price increases on its systems.

Balance Sheet and Supply Commitments Surge

• Supply and capacity commitments jumped from $119B at end of Q1 to $279B at end of Q2, primarily driven by memory procurement. The first three fiscal years ($92B in remainder of FY27, $87B in FY28, $88B in FY29) account for the vast majority.
• Inventory rose to $31.6B from $25.8B sequentially, driven by preparation for the Vera Rubin launch.
• DSO rose to 60 days from 45 days sequentially, reflecting extended payment terms granted to investment-grade customers for large, multi-quarter purchases.
• NVIDIA issued $25B of senior unsecured notes in Q2 to be used for general corporate purposes, a significant increase in long-term debt from $7.5B to $32.4B.

Free Cash Flow Decline Despite Higher Revenue

• Free cash flow declined to $21.3B in Q2 from $48.6B in Q1, driven by higher working capital (especially accounts receivable up $22.3B sequentially) and higher cash taxes. The sequential FCF decline should be understood in the context of timing, not operational deterioration.

OpenAI Commercial Relationship Update

• OpenAI's existing and planned commitments now represent approximately 12 GW of NVIDIA compute through 2030. The PORTS-Pike partnership adds up to $105B in NVIDIA guarantee exposure. This deepens concentration risk: management estimates frontier AI labs for which NVIDIA expects to leverage its balance sheet will contribute roughly 25% of next year's business.
• Management characterized these labs as growing faster than their balance sheets can support, with NVIDIA filling the financing gap.

China: Still Zero Revenue, No Change

• Hopper 200 (H20) shipments to China were less than 1% of Data Center revenue in Q2, per licenses. No China data center compute revenue is assumed in any forward guidance. No expected timeline for resolution has been provided.

Thesis Updates

Gross Margin Trajectory: A Material Change in the Bull Case

• The mid-70s gross margin target that underpinned the bull case for NVIDIA's profitability profile has been formally reset. Management now guides Q4 FY27 margins to bottom in the 71-72% range and FY28 at 72-73% — roughly 200-300 bps below prior targets.
• The mechanism (memory price inflation driven by the AI buildout) is tied to the same demand surge driving revenue, which management uses to argue this is transitory. But the structural question is whether NVIDIA can pass through memory cost increases in its system pricing, and whether quarterly pricing power is as solid as generational pricing power.
• This is the first clear indication that the mid-70s gross margin level — achieved at Blackwell's peak — may not be a floor going forward.

Revenue Growth Acceleration: Fourth Consecutive Quarter of YoY Acceleration

• Revenue growth accelerated to 106% YoY in Q2 from 85% in Q1 — the fourth consecutive quarter of YoY acceleration. This is exceptional at NVIDIA's scale ($96B quarterly revenue base).
• The FY28 preliminary guidance of ~70% growth implies roughly $330-340B in annual revenue. While this would mark a deceleration from FY27's run rate, management is explicit that this is supply-constrained, not demand-constrained.
• The fact that NVIDIA is now providing full-year forward guidance for the first time is itself a signal: the company has sufficient demand visibility and supply chain confidence to make a public commitment 12 months out.

ACIE Segment: Emerging as NVIDIA's Most Defensible Growth Engine

• The ACIE segment (AI Clouds, Industrial, Enterprise, Sovereign) grew faster than Hyperscale in Q2 (25% QoQ vs. 13% QoQ) and now represents roughly 45% of Data Center revenue.
• This segment is structurally more defensible against the custom ASIC threat: these customers need full-stack, pre-integrated solutions and will not build bespoke chips. NeoCloud partners exiting FY27 at 8 GW of installed capacity (vs. 3 GW at end-2025) illustrates the pace of non-hyperscale infrastructure buildout.
• NVIDIA's new revenue-sharing model with NeoClouds — where NVIDIA earns both hardware revenue and recurring rental income — represents a potential shift in the business model toward more software-like recurring revenue streams.

Circular Financing Risk: Now Quantifiable and Material

• The PORTS-Pike guarantee ($105B maximum exposure) and the $36B AI cloud agreements and $20B third-party data center leases represent a material and growing contingent liability on NVIDIA's balance sheet.
• Management expects frontier AI labs for which NVIDIA will leverage its balance sheet to account for roughly 25% of FY28 revenue. Jensen Huang addressed the "circular financing" critique directly, defending it as investing in once-in-a-generation companies.
• Investors should weigh the risk that if one or more frontier AI labs encounters financial difficulty, NVIDIA's guarantee exposure could be called. The $105B SB Energy cap alone exceeds NVIDIA's current long-term debt balance.

Supply Constraints as Both a Ceiling and a Signal

• Management's statement that demand would support 100%+ growth in FY28 but supply limits the outlook to 70% is a notable signal in both directions.
• Bullishly, it means the demand backdrop is intact and NVIDIA is not capacity-constrained by customer interest — only by its ability to manufacture.
• Bearishly, it raises the question of whether competitors can close the supply gap during the 12-18 months when NVIDIA cannot fulfill demand. A supply-constrained NVIDIA is also an opportunity for customers to trial alternatives.
• The $279B in supply commitments now on NVIDIA's balance sheet — up from $119B last quarter — represents the largest single-quarter commitment expansion in the company's history. This scale of commitment works only if FY28 demand materializes as projected.

Competitive Landscape: Custom Chips Acknowledged, Not Feared

• OpenAI's Jalapeno ASIC was raised by analysts; Jensen Huang dismissed it by arguing NVIDIA is building a full-stack platform that runs across all clouds and all model types, while Jalapeno is an inference-specific chip for one cloud.
• NVIDIA's argument is consistent with prior quarters: ASICs optimize for narrow, stable workloads; NVIDIA handles the entire AI lifecycle and all model architectures. The resolution of this debate will play out over 2-3 years as ASIC deployments scale.
• The Groq 3 LPX launch is NVIDIA's own answer to ultra-low-latency inference, adding a niche interactive token service capability to the platform. Management continues to characterize this as a small fraction of the total market.