NVIDIA Corporation (NVDA) | The Buildout — AI Infrastructure
The Verdict
NVIDIA designs accelerated computing platforms and full-stack AI factory systems. Its technology stack includes the CUDA development platform and domain-specific software libraries, and its products span GPUs, CPUs, networking, systems, and software. The company sells to hyperscalers, AI clouds, sovereigns, enterprises, and edge customers building AI infrastructure. Its role in the buildout is as a supplier of the compute, networking, and software layer that those customers use to stand up AI factories.
| Market Cap | — |
| Revenue (TTM) | $303.0B |
| Revenue Growth | +83.4% |
| EBITDA Margin (TTM) | 66.4% |
| Net Cash | $60.5B |
| Earnings Beats | 7 of 7 |
| P/E (TTM) | — |
| EV/EBITDA (TTM) | — |
What We Like
- Revenue re-accelerated to +106% y/y in Q2 FY2027 from +85% y/y in Q1 FY2027, which management called the fourth consecutive quarter of acceleration.
- Data center was about 93% of Q2 FY2027 revenue, with ACIE up 138% y/y; networking set a record, up 18% q/q.
- Management points to $1 trillion in cumulative Blackwell and Rubin revenue from 2025 through calendar 2027.
- Vera Rubin production shipments commenced in August 2026, and management guides it to about 20% of Q3 data center revenue.
- Capital return scaled: dividend at $0.25 per quarter, an $80B buyback authorization, and 60% of free cash flow returned year-to-date versus a 50%+ plan.
What We’re Watching
- Q3 FY2027 earnings on Nov 17, 2026 tests the $108B revenue guide, Vera Rubin at ~20% of data center revenue, and the gross-margin path toward the Q4 71–72% bottom.
- Supply is expected to remain a bottleneck at least through the end of FY2028; management says it has supply for about 70% of demand.
- Customer concentration is high: three direct customers were 21%, 17%, and 16% of Q1 FY2027 revenue, and management says AI labs where NVIDIA leverages its balance sheet are expected to be roughly a quarter of next year's business.
- Memory pricing is the named cause of the gross-margin reset; FY2028 gross margin is guided to settle at 72–73% as executed price increases take effect in Q1.
The revenue and product-ramp evidence strengthens the case: growth accelerated for four consecutive quarters, Vera Rubin shipped on schedule, and management now guides FY2028 revenue growth of about 70% on a supply-constrained basis. The margin reset and the financing-dependent customer mix are the two facts that weaken the quality read. The key open question is whether the NeoCloud and Frontier-lab demand NVIDIA is helping finance proves as durable as hyperscaler demand when the backstop is tested.
Earnings Beat
Q2 FY2027, for the quarter ended 2026-07-26, revenue was $96.2B, up 18% sequentially and 106% y/y. Data center revenue was $89B, with Hyperscale at $49B and ACIE at $40B. GAAP and non-GAAP gross margins were both 75.0%. GAAP EPS was $2.46 and non-GAAP EPS was $2.22.
| Metric | Q2 FY2027 | Q1 FY2027 | Q2 FY2026 | YoY |
|---|---|---|---|---|
| Revenue | $96.2B | $81.6B | $46.7B | +105.9% |
| Gross margin | 75.0% | 74.9% | 72.4% | +260bps |
| EBITDA | $64.9B | $54.5B | $29.1B | +122.8% |
| EPS | $2.46 | $2.39 | $1.08 | +128.2% |
| ACIE revenue | $40B | $37.4B | n/a | +138% y/y |
we are experiencing extreme pricing conditions in memory. The magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year. As a result, we are resetting expectations today.. We want to be direct about this rather than let it linger as an open question.— Colette Kress, 2026-08-26
Management tone: Management shifted from reaffirming mid-70s gross margin in Q1 FY2027 to pre-announcing a memory-driven margin reset in Q2 FY2027. It quantified the supply constraint at about 70%, named the circular-financing critique, and introduced a full-year FY2028 revenue guide of about 70% growth.
Management Guidance
For Q3 FY2027, management guides revenue of $108B ±2%, gross margin of 74% ±50 bps, GAAP OpEx of about $9.2B, and non-GAAP OpEx of about $9.0B. It expects Q4 FY2027 gross margin to bottom at 71–72% and FY2028 gross margin to settle at 72–73% as executed price increases take effect in Q1. FY2028 revenue growth is guided to approximately 70% y/y, described as supply-constrained, with supply expected to remain a bottleneck through the end of FY2028. FY2027 OpEx growth is guided to the low 50s and the tax rate to 16–18%. No China data center compute revenue is in the forward outlook.
Trajectory
Revenue rose from $81.6B in Q1 FY2027 to $96.2B in Q2 FY2027, with y/y growth accelerating from 85% to 106% and q/q growth easing from 20% to 18%. The Q3 FY2027 guide of $108B ±2% implies further sequential growth at a slower rate. Gross margin has been roughly flat near 75% in recent quarters, but management guides it down to 74% in Q3, a Q4 bottom of 71–72%, and 72–73% in FY2028 on memory costs. EBITDA was $64.9B in Q2 FY2027, 67.4% of revenue, while free cash flow fell to $21.4B from $48.6B in Q1 as inventory rose to $32B ahead of the Vera Rubin launch and days sales outstanding reached 60 days.
The Model
The model's locked projections show FY+1 revenue of $410,036M and EBITDA of $271,444M, a 66.2% EBITDA margin. For FY+2, the model projects revenue of $704,918M and EBITDA of $467,713M, a 66.35% EBITDA margin. The near-term anchor is the supply-constrained FY2028 outlook of approximately 70% revenue growth, with Vera Rubin ramping and ACIE growing faster than hyperscale. FY+2 depends on supply availability, memory-cost normalization or price pass-through, and whether NeoCloud, sovereign, and enterprise demand compounds.
| Metric | FY2026 | Next FY (E) | Following FY (E) |
|---|---|---|---|
| Revenue | $215.9B | $410.0B | $704.9B |
| YoY Growth | — | +89.9% | +71.9% |
| EBITDA | $133.2B | $271.4B | $467.7B |
| EBITDA Margin | 61.7% | 66.2% | 66.3% |
Projections are the median of 5 independent model runs. The model’s revenue sits 25.6% above analyst consensus.
For Q3 FY2027, management guides revenue of $108B ±2%, gross margin of 74% ±50 bps, GAAP OpEx of about $9.2B, and non-GAAP OpEx of about $9.0B. It expects Q4 FY2027 gross margin to bottom at 71–72% and FY2028 gross margin to settle at 72–73% as executed price increases take effect in Q1. FY2028 revenue growth is guided to approximately 70% y/y, described as supply-constrained, with supply expected to remain a bottleneck through the end of FY2028. FY2027 OpEx growth is guided to the low 50s and the tax rate to 16–18%. No China data center compute revenue is in the forward outlook.
What Could Go Right — and Wrong
- Supply gap closes faster than guided; management says demand is closer to 100% while supply supports about 70%.
- ACIE and sovereign demand keep compounding; NeoCloud installed capacity is targeted at 8GW exiting the year, up from about 3GW at end-2025.
- Vera Rubin reaches about 20% of Q3 data center revenue and the $40B per gigawatt revenue opportunity versus $25B for Blackwell drives content growth.
- Memory prices normalize or executed price increases stick, allowing gross margin to settle at the guided 72–73% in FY2028.
- CPU revenue more than doubles in FY2028 against a stated $200B TAM, and physical AI, already >$9B TTM, adds a second demand pool.
- Memory costs stay extreme; Q4 gross margin bottoms below 71–72% or the FY2028 72–73% recovery does not arrive.
- Supply bottleneck extends beyond FY2028, capping revenue below the approximately 70% FY2028 guide.
- A Frontier-lab credit event: roughly a quarter of next year's business is from AI labs where NVIDIA leverages its balance sheet, alongside ~$50B invested, $500B+ third-party financing platforms, and ~2GW credit enhancement.
- Custom silicon takes high-volume inference workloads; Meta has 1+GW Broadcom and 6GW AMD Instinct commitments, Amazon has $225B Trainium commitments, and Google is considering external TPU sales.
- Customer concentration and working capital pressure: three direct customers were about 54% of Q1 FY2027 revenue, DSO is 60 days, and inventory is $32B.
Looking Ahead
Over the next 12 months, the key tests are the Q3 FY2027 earnings on Nov 17, 2026 and the ramp of Vera Rubin into customer data centers. Management also has public appearances at Goldman Sachs Communacopia on Sept 10, 2026 and GTC Berlin on Oct 21, 2026. Supply remains the stated limiter through the end of FY2028, and the Australia ecosystem buildout of up to 2GW by 2027 and the AWS 2M-GPU deployment through Q2 FY2029 are the largest dated capacity commitments in the record.
- Sept 10, 2026Goldman Sachs Communacopia — Jensen keynote/fireside; management visibility update.
- Oct 21, 2026GTC Berlin keynote — Product and platform roadmap signals.
- Nov 17, 2026Q3 FY2027 earnings — Tests revenue guide, Vera Rubin mix, gross-margin path.
- By 2027Australia 2-GW buildout — Firmus, Sharon AI, IREN, Megaport, ResetData, CDC, NEXTDC, AirTrunk.
- Through Q2 FY2029AWS 2M-GPU deployment — Additional GPUs plus Vera CPUs and Nemotron on AWS.
- Through 2030OpenAI ~12GW commitment — Existing and planned NVIDIA compute commitments.
Financials
Annual Summary
| Metric | FY2025 | FY2026 | TTM | YoY |
|---|---|---|---|---|
| Revenue | $130.5B | $215.9B | $303.0B | +65.5% |
| Gross Margin | 75.3% | 70.3% | 74.7% | 495bps |
| EBITDA | $83.3B | $133.2B | $201.3B | +59.9% |
| EBITDA Margin | 63.8% | 61.7% | 66.4% | 215bps |
| Net Income | $72.9B | $120.1B | $192.9B | +64.7% |
| Free Cash Flow | $60.9B | $96.7B | $127.0B | — |
| Net Cash | — | — | — | — |
Key Ratios (Trailing)
- P/E TTM—
- EV/EBITDA TTM—
- EV/Revenue TTM—
- Price/FCF TTM—
- Gross Margin (TTM)74.7%
- EBITDA Margin (TTM)66.4%
- Net Margin (TTM)63.7%
- ROIC92.6%
- FCF Conversion63.1%
- SBC / Revenue2.4%
The Company
NVIDIA describes itself as having pioneered accelerated computing and as a data center scale AI infrastructure company. Its technology stack includes the CUDA development platform and hundreds of domain-specific software libraries. The company reports in two segments, Compute & Networking and Graphics, and management also speaks to two market platforms: data center and edge computing. Its products include Blackwell and Blackwell Ultra, Vera Rubin, Grace and Vera CPUs, Spectrum-X Ethernet and InfiniBand networking, Groq 3 LPX, NVIDIA AI Enterprise software, GeForce RTX, and DRIVE Hyperion.
NVIDIA operates a fab-lite model. It uses foundries such as TSMC and Samsung for wafers, buys memory from SK Hynix, Micron, and Samsung, and uses Hon Hai, Wistron, and Fabrinet for assembly, test, and packaging. The 10-K says dependency on third-party suppliers reduces control over product quantity, quality, yields, and delivery schedules. NVIDIA owns and leases about 3 million square feet for its Santa Clara headquarters, leases data center space in Santa Clara, and has facilities primarily in China, India, Israel, and Taiwan. Management stated on the Q2 FY2027 call that the company has roughly 40,000 employees.
Business Segments
Competitive Landscape
The 10-K lists competitors in accelerated compute and AI including AMD, Huawei, and Intel, and cloud companies designing internal silicon including Alibaba, Alphabet, Amazon, Baidu, and Microsoft. It also lists networking competitors such as AMD, Arista, Broadcom, Cisco, Hewlett Packard Enterprise, Huawei, Intel, Lumentum, and Marvell. The cross-stack theme file describes a custom-silicon parallel economy from Broadcom, Marvell, and Qualcomm XPU platforms, with Meta, Amazon, and Google commitments, and concludes that it sits alongside, not against, NVIDIA's position because total AI compute demand is larger than any single supplier can satisfy.
- AMDListed in the 10-K for accelerated compute/AI and networking; cross-stack themes cite Meta's 6GW AMD Instinct deal.
- HuaweiListed in the 10-K for accelerated compute/AI; the 10-Q says export controls may benefit certain competitors and encourage alternatives.
- IntelListed in the 10-K for accelerated compute/AI, networking, and SoC; HPE names Intel and AMD as single-source x86 suppliers.
- BroadcomListed in the 10-K for networking and is a supplier of 200G EML lasers; cross-stack themes cite Broadcom's 20+GW AI XPU platform with Apollo/Blackstone and a first trench valued at $35B.
- AmazonListed in the 10-K among cloud companies designing internal silicon; cross-stack themes cite a $20B+ chips run-rate and $225B Trainium revenue commitments, while AWS is also a customer for 2M additional GPUs.
Supply Chain
NVIDIA sits at the center of the AI hardware chain: it designs the accelerated computing platforms and relies on third-party foundries, memory makers, and assembly partners to build them. Its customers are hyperscalers, AI clouds, sovereigns, enterprises, and edge device makers.
More on NVDA: Earnings recap