The AI super-cycle’s toll booth.

NVIDIA FY2026 (ended 2026-01-25) total revenue was $215.9B, up 65.5% YoY; the Data Center market platform accounted for 89.7% of revenue, with quarterly run-rate already at the $50–70B range. The year carried two structural shocks: a Q1 $4.5B H20 inventory writedown (new US export-control rules), and China’s share of revenue retreating from 19.2% to 9.1%. This report is built from three SEC filings — FY2026 10-K, Q3 FY26 10-Q, Q2 FY26 10-Q — and walks through the Data Center–dominated revenue mix, customer concentration, the inventory build for the Blackwell ramp, and key governance points.

Quick numbers

§01 · Key Metrics — the year at a glance

FY2026 marks the inflection where NVIDIA evolves from “supplier to the AI infrastructure cycle” into something closer to a “supercompute infrastructure operator.” Data Center revenue grew $78.5B YoY; the top two customers combined for 36% of revenue; $40B in non-public equity (incl. Groq) plus $22.7B of long-term leases not yet commenced — the balance sheet has expanded from a semiconductor company toward something closer to an “AI compute platform” capital structure.

— Synthesis of 10-K / 10-Q disclosures; data not independently audited

§02 · Business — platforms, geography, customers

Segment results

Two reporting segments · USD millions.

Customer concentration · 10-K disclosure

Direct customers >10% of total revenue.

1. Customer A · 22% of total revenue. 12% in FY25 · 13% in FY24 — share has roughly doubled. The name is not disclosed; the market widely assumes Microsoft or another hyperscale CSP. Attributed to the Compute & Networking segment.
2. Customer B · 14% of total revenue. 11% in FY25 — second-largest customer. Also Compute & Networking. Q3 FY26 disclosures still showed this customer at the 15% level.
3. Q3 single quarter: Customer C 13% · Customer D 11%. The Q3 FY26 10-Q disclosed four direct customers each above 10% in the quarter (22% / 15% / 13% / 11%), 61% of revenue combined. CSP concentration is extreme — among the top tail risks.
4. Indirect end customers. The Q2 FY26 10-Q separately disclosed two indirect end customers (each shipping through A/B) above 10%. The company describes them as “AI research and deployment companies” contributing a meaningful share — broadly read as OpenAI/Anthropic-type customers procuring via the cloud providers.

§03 · P&L — operating leverage at scale

Income statement (quarterly + FY)

Per SEC disclosure · USD millions.

* Q1 FY26 derived from 6M FY26 (Q2 10-Q) less Q2; Q4 FY26 derived from FY26 less Q1/Q2/Q3; not separately disclosed. Q1 includes the $4.5B H20 inventory writedown (in cost of revenue).

§04 · Balance — inventory build for Blackwell

Balance sheet comparison

Jan 25 2026 vs Jan 26 2025 · USD millions.

Capital structure · single class

Single-class common stock · 1 share = 1 vote. As of 2026-01-25, 24,304M common shares outstanding.

Unlike most founder-led Silicon Valley public companies, NVIDIA uses a single-class capital structure — founder Jensen Huang owns ~3.5% (non-controlling), and governance follows a standard majority-rule model. Huang has actively chosen to keep this structure long-term.

§05 · Cash Flow — cash returns and Groq

Cash flow takeaways

1. OCF $102.7B · FCF $96.7B. Operating cash flow grew 60.3% from $64.1B; FCF is 80.5% of net income — high-quality cash conversion. Capex doubled to $6.0B, 2.8% of revenue.
2. Buybacks $40.1B · dividends $1.0B. FY26 buybacks $40.1B (vs FY25 $33.7B); dividends $974M. Total shareholder return $41.1B, 42.5% of FCF. Buyback authorization remaining is large.
3. $13B invested in Groq. A one-off $13B strategic investment in Groq (AI inference chips) in FY26. Non-public equity surged from $3.4B to $22.3B — the canonical move in the “the company is becoming an AI infrastructure fund” thesis.
4. SBC $6.4B. Stock-based comp at $6.4B is just 3.0% of revenue (FY25: 3.6%) — extremely low among large-cap tech, partly because Huang himself takes a small base salary plus performance shares.

§06 · Leadership — the Jensen era

Named Executive Officers

The following are the named executives identified in the FY2026 10-K (filed 2026-02-25) signature page and Item 10 disclosures. There were no executive departures or new appointments during the period. Founder Jensen Huang has served as CEO since co-founding the company in 1993, one of the longest tenures of any sitting Silicon Valley CEO.

Executive & governance highlights

FY26 10-K · Item 10 / 11 disclosures.

1. Management. Jensen Huang (CEO), Colette Kress (CFO), Debora Shoquist (Ops), Ajay Puri (Sales), Tim Teter (GC) + Donald Robertson (Chief Accounting Officer) — no Item 5.02 events in FY26.
2. CEO comp. Huang receives base salary plus performance shares, with a low cash-bonus weighting. The classic “founder CEO with low cash compensation and high equity alignment” profile.
3. Capital structure. Single-class common stock (1 vote per share). Huang owns ~3.5%; no super-voting and no controlling stockholder.
4. Headcount. About 42,000 employees in 38 countries at FY2026 end; 31,000 in R&D, 11,000 in sales / operations / G&A.

Board of Directors

13 directors · 2026 proxy slate.

Huang is the sole inside director; the other 12 are independent, with Dawn Hudson as Lead Independent Director. Unlike Meta or Alphabet’s founder-controlled models, NVIDIA has 12/13 independent directors — a structure that “looks more like a traditional blue chip,” and a quiet reason actively managed funds tend to overweight the name.

§07 · Risk — what could derail the trade

1. China export controls + H20-style risk. FY26 Q1 booked a $4.5B writedown of inventory and purchase commitments after the new H20 licensing requirement in 2025-04. China’s revenue share has fallen from 19.2% (FY25) to 9.1% (FY26). If a future tightening hits a China-specific Blackwell / Rubin variant, a single writedown could be several times the H20 case. The 10-K places this first among structural risks.
2. Customer concentration · four CSPs each >10%. Q3 FY26 single-quarter: Customer A 22% · B 15% · C 13% · D 11% — four customers, 61% of revenue combined. For the full year FY26, A+B alone are 36%. If any one shifts toward in-house ASICs (Google TPU, Meta MTIA, AWS Trainium), single-quarter shipping cadence is affected. Among indirect end customers, the renewal capacity of “AI model companies” like OpenAI/Anthropic is a hidden variable.
3. In-house and competitive substitution · AMD / Intel / custom silicon. AMD MI300/MI400 ramping, Intel Gaudi, and CSP custom ASICs continue to chip away at AI training/inference share. The CUDA software moat remains, but on the inference side (latency-sensitive, cost-sensitive) substitution pressure is highest. The $13B Groq investment can be read as NVIDIA’s hedge against the “inference ASICs displace general-purpose GPUs” path.
4. Supply chain + ramp execution. TSMC advanced packaging (CoWoS), HBM supply (SK Hynix / Micron / Samsung), and back-end test are the core bottlenecks. Blackwell Ultra is shipping; Rubin is still ramping — any link in the chain slipping affects quarterly cadence. With inventory up from $10B to $21B, capital is committed ahead of demand, and impairment risk is amplified if the demand cadence shifts.
5. Legal + regulatory · antitrust / export / tax. The US FTC/DOJ, the European Commission, France, and China have all opened or are watching antitrust / tying investigations on NVIDIA; the CUDA ecosystem and InfiniBand bundling are core flashpoints. Beyond that, the global minimum tax, the sustainability of the $3.1B Israel-segment tax benefit, and the expansion of the US “Foreign Direct Product Rule” on semiconductor exports are all structural unknowns.

Bottom line · The biggest beneficiary of the AI super-cycle — and the biggest single point.

NVIDIA’s FY2026 has triple-locked the franchise: scale (revenue $216B), margins (operating margin 60%), and customer relationships (multi-year commitments from four CSPs). But three new structural questions are starting to come into view:

• The pace at which customers substitute their own ASICs
• The “never-zero, never-saturated” tail of the China market
• The effective capitalization rate of inventory + long-term leases + the equity portfolio

“Net income $120B · FCF $96.7B · single-class capital · 12/13 independent directors” — a rare combination of traditional blue-chip governance and a hyper-growth curve.

§08 · Valuation — valuation in context

NVDA current valuation

Data as of 2026-04-20 · close.

Peer benchmark

TTM multiples · latest.

On a TTM basis, NVDA trades at 40.5× P/E and 36.4× EV/EBITDA — actually the lowest earnings multiple among the AI Big Four, because $120B of real net income inflates the denominator, while AMD (104×) and AVGO (79×) read more like “discounted AI expectations.” On a sales basis, 22.9× P/S is still a premium, but lower than AVGO’s 28.1×, reflecting the market awarding AVGO a higher growth premium for its ASIC + software combination. FCF Yield of 2.0% is the highest of the four, meaning that by current cash flow, NVDA is the “cheapest” AI bellwether — partly offsetting the tail discount from China export controls (China revenue already retreated from 19.2% to 9.1%). INTC’s P/E is N/M and FCF is negative; its multiples have broken down and it is no longer in the same comparable bucket — it sits in a special restructuring state inside foundry recapitalization. At $200.98, NVDA is just -5.3% from its 52-week high of $212.19; multiple repair and earnings delivery have happened simultaneously, so further upside hinges on the pace of the Rubin ramp and the slope of customer ASIC substitution.