NVIDIA Through the Lens of Its Annual Report: The Triple Metamorphosis Behind $216 Billion in Revenue

If you only read the quarterly earnings, NVIDIA's FY2026 is a straightforward story: seven consecutive beat-and-raise quarters, Data Center revenue on an unstoppable tear, Blackwell demand far outstripping supply. We've dissected those numbers in every quarterly earnings review already.

But the 10-K is a different animal. This is a document led by lawyers and auditors, not a carefully choreographed performance by the IR team. It carries none of management's rhetorical polish, none of the analyst day's strategic narrative arc. What it does carry is everything the SEC mandates a company must disclose — including the things the company would rather you not dwell on.

When you spread out every number in NVIDIA's FY2026 annual report, what emerges is not "a chip company that sold a lot of GPUs." What emerges is an entity undergoing three simultaneous transformations: a phase transition in financial scale, a paradigm shift in strategic logic, and a fundamental recast of its geopolitical role. None of these are visible in a quarterly snapshot.

I. The Gravity of Numbers: When Growth Ceases to Be Linear

Start with a set of figures worth pausing over.

FY2024 (ending January 2024) full-year revenue: $60.9 billion. FY2025: $130.5 billion. FY2026: $215.9 billion. In three fiscal years, revenue more than tripled. This is not a growth curve. This is a phase transition.

The net income trajectory is equally striking: $29.8 billion → $72.9 billion → $120.1 billion. Operating cash flow climbed from $28.1 billion to $102.7 billion — a single company generating more annual cash flow than the GDP of most countries.

But what the annual report truly underscores is the transformation of the balance sheet. Total assets leapt from $111.6 billion in FY2025 to $206.8 billion — nearly doubling in a single year. Stockholders' equity surged from $79.3 billion to $157.3 billion. This is not incremental asset accumulation. This is a company whose financial center of gravity shifted to an entirely different order of magnitude within twelve months.

NVIDIA Five-Year Cumulative Total Return (NVDA vs. S&P 500 vs. Nasdaq 100, 1/31/2021 – 1/25/2026)

source: NVIDIA Corporation, FY2026 10-K Annual Report (SEC Filing, February 25, 2026)

The five-year cumulative total return chart included in the filing tells the same story from a different angle: $100 invested in January 2021 became $1,449 by the end of FY2026 — while the S&P 500 grew to just $200 and the Nasdaq 100 to $206. The steep inflection between FY2024 and FY2025 marks the precise temporal coordinates of the AI demand explosion.

A note on gross margin. It declined from 75.0% in FY2025 to 71.1%, a 3.9 percentage point compression. Part of this reflects the product mix shift from Hopper HGX systems to full-stack Blackwell data center solutions — which bundle networking equipment, liquid cooling, and rack assemblies, naturally carrying lower per-unit margins than standalone GPU sales. Another significant factor is the $4.5 billion H20-related charge, which we will return to shortly.

Yet even with margins under pressure, operating expense leverage remains firmly in NVIDIA's favor: R&D as a percentage of revenue fell from 9.9% to 8.6%; SG&A from 2.7% to 2.1%. In absolute terms, R&D spending grew 43% to $18.5 billion — but against this revenue base, the expense ratio actually declined. A textbook demonstration of scale economics at work.

II. A Paradigm Shift in Capital Allocation: From Chipmaker to Ecosystem Architect

If quarterly earnings tell you how much NVIDIA earned, the 10-K tells you where it spent the money. And FY2026's capital allocation picture reveals an entity that bears little resemblance to a traditional semiconductor company.

The single most arresting line item: NVIDIA invested $17.5 billion in private companies and infrastructure funds during FY2026. The comparable figure in FY2025 was $1.5 billion. An eleven-fold increase in a single year. By fiscal year-end, the carrying value of non-marketable equity securities had swollen from $3.4 billion to $22.3 billion.

Simultaneously, publicly traded equity holdings surged from less than $400 million to $17.7 billion — of which $10.5 billion remains subject to lock-up restrictions. The filing discloses a transaction previously mentioned only in passing: in the first quarter of FY2026, an investment was reclassified from non-marketable to marketable securities (as the portfolio company began public trading), generating $6.6 billion in unrealized gains. This single event propelled "other income, net" from $1.0 billion in FY2025 to $9.0 billion — an almost entirely investment-driven new variable in the income statement.

Then there is the Groq transaction. In our prior deep dive, we characterized this deal as "a strategic defensive acqui-hire." But the financial statement notes reveal more precise figures: $13.0 billion in cash at closing, with an additional $4.0 billion (including imputed interest) due within one year. The transaction generated $14.4 billion in goodwill and $2.5 billion in developed technology intangible assets. Crucially, NVIDIA did not acquire any of Groq's customer contracts, existing products, or equity — it purchased purely people and technology. The $14.4 billion in goodwill is tax-deductible, meaning the after-tax cost is meaningfully below the nominal consideration.

Add to this the $3.5 billion in land, power, and shell facility guarantees extended to early-stage companies (with 5-7 year terms), plus the ongoing investment negotiations with OpenAI — though the filing explicitly states that "there can be no assurance that definitive agreements will be reached" — and NVIDIA's capital allocation strategy now resembles less a semiconductor company's playbook than that of an AI-domain Berkshire Hathaway: deploying cash flow's gravitational pull to draw an entire ecosystem into orbit.

Another number overlooked in quarterly reporting: NVIDIA has committed to $27.0 billion in multi-year cloud service agreements to support its own R&D. This is not a trivial sum — it signals that NVIDIA itself has become a large-scale consumer of AI compute, operating internal AI supercomputer clusters valued in the billions.

Assembling these pieces, a coherent picture emerges:

• $17.5 billion in strategic equity investments — cultivating and binding the AI ecosystem
• $17.0 billion in the Groq transaction — fortifying the inference technology moat
• $3.5 billion in facility guarantees — lowering the barrier for customers to build AI infrastructure
• $27.0 billion in cloud service commitments — building internal AI R&D infrastructure
• $40.4 billion in share repurchases — returning capital to existing shareholders
• $6.1 billion in capital expenditures — expanding physical infrastructure

These six items alone sum to a capital deployment machine spending over $110 billion annually. The "chip company" narrative no longer contains what NVIDIA has become.

III. The Export Control Chronicle: A Geopolitical Epic Still Being Written

In quarterly reports, export controls are typically compressed into a paragraph or two of risk disclosure. But the 10-K's mandate is comprehensive disclosure, and when you read the full export control section of NVIDIA's FY2026 annual report, what unfolds is a four-year geopolitical chronicle.

August 2022 : the U.S. government announced the first export restrictions targeting China's semiconductor and supercomputing industries, affecting the A100 and H100. July 2023 : restrictions expanded to cover parts of the Middle East. October 2023 : the control scope widened again, adding an extensive product list — A100, A800, H100, H800, L4, L40, L40S, RTX 4090, GB200 NVL72, B200 — covering virtually NVIDIA's entire data center product line.

The inflection point came in April 2025 : the U.S. government imposed a license requirement on H20 exports to China. The H20 was a product NVIDIA had specifically designed as a compliance-friendly, downgraded version for the Chinese market. This single restriction triggered a $4.5 billion inventory and purchase commitment charge in Q1 FY2026 — a single-quarter write-down exceeding the entire annual revenue of many semiconductor companies.

What followed grew only more convoluted. In August 2025, NVIDIA obtained limited licenses to ship H20 to certain Chinese customers, generating approximately $60 million in revenue — but the U.S. government simultaneously indicated it expected to receive 15% or more of licensed sales revenue. In February 2026, an additional license was granted to ship small quantities of H200 to specific Chinese customers, but the products must first be shipped to the United States for inspection before re-export, triggering a 25% import duty.

One sentence in the filing stands out with particular weight: "As of the end of FY2026, the company has been substantially excluded from the China data center compute market."

The gravity of that statement requires the context of geographic revenue distribution. FY2026 China (including Hong Kong) revenue was $19.7 billion, down 21.5% year-over-year — and the vast majority came from non-data center businesses. Revenue from customers outside the U.S. declined from 41% to 31% of total, while U.S. domestic revenue surged to 69% ($149.6 billion). Taiwan-headquartered customers contributed $42.3 billion, but the filing specifically notes that approximately 76% of the end users of Taiwan customers' Data Center revenue are located in the United States and Europe.

In other words, measured by ultimate end users, NVIDIA's revenue concentration is significantly higher than the surface numbers suggest.

The deeper repercussions lie in the competitive landscape. The filing states candidly that export controls have "helped competitors build larger developer and customer ecosystems globally." Meanwhile, on September 15, 2025, China's antitrust regulator issued a preliminary ruling: NVIDIA's compliance with U.S. export controls — resulting in the provision of downgraded products to the Chinese market — constitutes "unfair discrimination" and violates the conditions under which China approved the Mellanox acquisition.

Here lies a profound paradox: NVIDIA has been found to violate Chinese law precisely because it complied with American law. Regardless of the final verdict, this "squeezed from both sides" predicament is geopolitical risk in its most tangible form.

The filing also reveals a detail rarely discussed: U.S. legislative proposals for chip tracking and restriction mechanisms have prompted the Chinese government to publicly question whether H20 products contain "built-in backdoors." The erosion of trust runs in both directions.

IV. The Silent Gravity: Taxes, Regulation, and Open-Source AI

The annual report contains a set of trend lines that lack the drama to make headlines, yet may matter more to long-term investors than any single quarterly beat.

The climbing effective tax rate. FY2024: 12.0%. FY2025: 13.3%. FY2026: 15.1%. The direction is singular. Although NVIDIA continues to benefit from FDII (Foreign-Derived Deduction Eligible Income) deductions, Israeli preferential tax rates, and R&D tax credits, these benefits are diluting as a proportion of pre-tax income each year. FY2026 income tax expense reached $21.4 billion — a figure that by itself exceeds the total annual revenue of most technology companies. The "One Big Beautiful Bill Act" (OBBBA), enacted in July 2025, introduced multiple federal income tax changes; the filing confirms their impact was recognized but does not quantify the details. The OECD's Pillar One and Pillar Two frameworks could further increase the tax burden going forward.

The global proliferation of antitrust investigations. The filing presents a striking regulatory roster: the European Union, the United States, the United Kingdom, China, South Korea, and France — competition regulators from virtually every major economy have sent NVIDIA "broad information requests." France's competition authority is investigating the graphics card and CSP market. The EU AI Act took effect on August 1, 2024, with full applicability following a two-year transition period. Multiple U.S. states enacted AI-related legislation effective January 1, 2026.

Open-source AI models as a competitive variable. One risk disclosure passage deserves close reading: "The rise of high-quality open-source foundation models makes advanced AI capabilities broadly accessible — if deployed on competitors' platforms, they could reduce demand for NVIDIA products." The filing goes further, specifically naming DeepSeek, Qwen, and KIMMI. This marks the first time NVIDIA has identified specific Chinese open-source models as risk factors in a formal SEC document.

These "silent gravitational forces" — rising tax rates, tightening regulation, open-source AI lowering hardware barriers — individually pose no near-term threat. But their direction is uniform, and each is accelerating.

V. A Moat Built by 42,000 People

One of the most inconspicuous yet potentially most significant data points in the entire annual report: NVIDIA's employee attrition rate stands at just 3.7%.

In an industry where talent competition is white-hot, 3.7% is almost implausibly low. For context, the average voluntary turnover rate at Silicon Valley technology companies typically ranges from 13% to 15%.

NVIDIA employs approximately 42,000 people across 38 countries. Over 80% hold technical roles; more than half possess advanced degrees. 31,000 work in R&D; 11,000 in sales, marketing, and operations. In FY2026, over 40% of new hires came from internal referrals — itself a strong signal of employee satisfaction.

The sentence "cumulative R&D investment exceeding $76.7 billion" is easy to gloss over, but its implications warrant a pause. This represents not merely the accumulation of capital expenditure but the compounding of institutional knowledge — the 350-plus libraries in the CUDA ecosystem, the network effects of over 6 million developers, and decades of architectural backward compatibility. These are the "interest payments" on cumulative R&D.

FY2026 stock-based compensation expense totaled $6.4 billion, but the fair value of RSUs and PSUs that vested during the year reached $22.2 billion. Put differently, NVIDIA employees realized wealth through equity awards worth 3.5 times the SBC expense recorded on the income statement. When your workforce grows wealthy by holding company stock, they possess the strongest possible incentive to preserve and expand that wealth — creating a self-reinforcing positive feedback loop.

VI. $95.2 Billion in Commitments: Supply Chain as Strategy

NVIDIA's supply chain commitments have reached a scale that warrants standalone discussion.

As of FY2026 year-end, outstanding inventory purchase and long-term supply and capacity obligations totaled $95.2 billion — with substantially all due for payment in FY2027. PwC, in its critical audit matters paragraph, flagged this area as requiring "substantial professional judgment and effort," given that management must make significant assumptions about future demand and market conditions.

The trajectory of product warranty liability is equally telling: $306 million at FY2024 year-end → $1.29 billion → $2.81 billion. A roughly nine-fold increase in two years. The filing notes the growth is primarily attributable to the Compute & Networking segment. This does not reflect deteriorating product quality — it reflects the fact that NVIDIA now sells complete data center systems rather than discrete GPU cards. When you graduate from a $10,000 graphics card to a $2 million NVL72 rack, warranty obligations naturally scale with it. This transformation is explored in greater detail in our earlier analysis of NVIDIA's shift from selling engines to building cars.

Inventories doubled to $21.4 billion, with work-in-process surging from $3.4 billion to $8.8 billion — reflecting the complex assembly cycles of Blackwell systems. Also noteworthy: NVIDIA is expanding its supply chain from near-total dependence on Asia toward the United States and Latin America. Yet long-lived assets in Taiwan (primarily property, plant, and equipment) grew from $1.5 billion in FY2025 to $3.2 billion — an increase exceeding 100%. This suggests that even within the "reshoring" narrative, dependence on the Taiwanese manufacturing base continues to deepen.

VII. In the Footnotes: Signals Hidden in the Fine Print

Every annual report contains details that never surface in financial summaries but are essential to understanding the complete picture. NVIDIA's FY2026 10-K is no exception.

The Microsoft Xbox clause: Under NVIDIA's agreement with Microsoft relating to Xbox, should any party launch a tender offer for more than 30% of NVIDIA's outstanding common stock, Microsoft holds a right of first and last refusal. This is an anti-takeover mechanism that has received virtually no market attention. Given NVIDIA's current market capitalization approaching $3 trillion, this clause may never be triggered in practice — but its very existence speaks to the depth and historical reach of the relationship between these two companies.

Delaware anti-takeover provisions: The board holds the right to issue preferred stock, alter board size, and impose supermajority voting requirements. Stockholders cannot act by written consent and must provide advance notice. This comprehensive anti-takeover toolkit ensures management's continued control of the corporation.

Israel's role: NVIDIA maintains approximately 6,000 employees in Israel, primarily supporting R&D, operations, and sales for networking products (a legacy of the Mellanox acquisition). Long-lived assets in Israel total $1.47 billion — third only to the United States and Taiwan. Israel's tax incentives saved NVIDIA $3.06 billion in income taxes — making geopolitical stability in the region a non-trivial risk variable.

ERP system migration: The filing notes that NVIDIA is implementing a new ERP system, which may introduce quality issues or programming errors. For a company generating $216 billion in annual revenue, ERP migration is a "boring but potentially fatal" operation — history offers no shortage of large enterprises that suffered significant financial losses from failed ERP implementations.

The $27 billion cloud commitment maturity schedule: $7.0 billion due in FY2027, $6.0 billion in FY2028, $5.0 billion in FY2029, $5.0 billion in FY2030, tapering thereafter. This means NVIDIA will pay $5-7 billion annually to cloud service providers over the next four years to support its own R&D infrastructure — the financial manifestation of its dual identity as both an AI hardware supplier and an AI compute consumer.

Reading Between the Annual Lines

Every quarterly report is a snapshot. An annual report is a full portrait.

The portrait painted by NVIDIA's FY2026 10-K is not that of "a chip company that sold a lot of GPUs." It depicts a new species — an entity simultaneously playing the roles of hardware platform provider, AI ecosystem investor, full-stack software company, and involuntary geopolitical actor. The $17.5 billion in strategic investments reveals it is building an ecosystem. The $4.5 billion write-down demonstrates that the costs of geopolitics are not hypothetical. The $95.2 billion in purchase commitments signals that its operational complexity has reached sovereign scale. The 3.7% employee attrition rate suggests a talent moat far deeper than any patent portfolio.

These stories, the quarterly earnings will never tell you.

For readers who have been following NVIDIA's quarterly performance, the annual report's value lies not in updating the numbers — you already know those. Its value lies in forcing you to step back and reassess what this company actually is, what it is becoming, and the unprecedented complexity of the environment it now navigates.

This is a company that earns $120 billion in net income, invests $17.5 billion annually in ecosystem building, faces scrutiny from every major antitrust authority on the planet, and is being pressured by the U.S. and Chinese governments in diametrically opposite directions. Its annual report reads less like a semiconductor company's financial filing and more like a field report on the power structures of twenty-first-century technology.

MD&A

Open this more visual friendly version in a new tab/点击跳转查看原文,左上角切换中文

Financials

Open this more visual friendly version in a new tab/点击跳转查看原文,左上角切换中文

Risk and Overview

Open this more visual friendly version in a new tab/点击跳转查看原文,左上角切换中文