NVIDIA 2026 10-K: what changed from 2025
NVIDIA's fiscal 2026 10-K against the 2025 report: the new counterparty risk, the H20 charge, China export licenses and the drop in gross margin.
NVIDIA filed its annual report for fiscal year 2026, the year that ended January 25, 2026, on February 25, 2026. filings.tools compares it with the fiscal 2025 report, filed February 26, 2025, heading by heading and sentence by sentence. Across the 18 items that differ, 375 sentences are new, 474 are gone, 869 are rewritten and 1,381 are unchanged. Every number below comes from that comparison, the NVIDIA 10-K diff, or from the filing's own pages.
The short version
- Risk factors (Item 1A): 24 headings. 1 is new, 2 are reworded and 21 are unchanged. No fiscal 2025 heading was removed. In the text, 74 sentences are new, 54 are gone and 90 are rewritten.
- The new risk factor: "Commercial arrangements expose us to counterparty risks."
- The two reworded headings: the supply risk now speaks of uncertain supply and "capacity" availability, where it said "component" availability. The other is the sustainability risk, which now reads "Scrutiny from shareholders, regulators and others regarding our corporate sustainability practices".
- China: a license requirement for H20 chips from April 2025, and a $4.5 billion charge for H20 inventory and purchase obligations in the first quarter of fiscal 2026.
- Margins: gross profit was 71.1% of revenue in fiscal 2026, against 75.0% in fiscal 2025.
- Investments: new sentences on NVIDIA's investments in other companies, including an investment and partnership agreement with OpenAI that NVIDIA says it is finalizing.
The one new risk factor: counterparty risk
The only new heading in Item 1A is about the deals NVIDIA makes to help customers build data centers. The new text says NVIDIA has entered into long-term capacity purchase obligations and financial guarantees, and has been asked to offer financing to customers and partners. It also says NVIDIA has not entered into any financing arrangements.
The risk it names is that customers or partners cannot meet their financial commitments, cannot secure financing or infrastructure, run into significant project delays, or fall into financial distress or insolvency. The text adds that financing, if NVIDIA offers it, could mean lower upfront cash flows over multi-year payment terms and more credit risk.
Also changed in Item 1A: the fiscal 2025 report described one indirect customer, buying mainly through system integrators and distributors, estimated at 10% or more of revenue. The fiscal 2026 report replaces that sentence: NVIDIA now estimates that some indirect customers individually represent 10% or more of revenue. Read the section yourself: Item 1A, Risk Factors.
China and export controls: what was added and what was dropped
New in fiscal 2026:
- In April 2025 the U.S. government told NVIDIA that it needs a license to export H20 chips, and any chip that matches the H20's memory or interconnect bandwidth, to China (including Hong Kong and Macau) and to D:5 countries.
- That requirement led to a $4.5 billion charge in the first quarter of fiscal 2026 for excess H20 inventory and purchase obligations.
- On September 15, 2025, China's antitrust regulators published a preliminary finding that NVIDIA's compliance with U.S. export controls, which required it to offer degraded products in China, discriminated against Chinese customers and so violated the terms of China's approval of the Mellanox acquisition.
The management discussion (Item 7) adds the licensing history after that. In August 2025 the U.S. granted licenses to ship certain H20 products to certain China-based customers, and NVIDIA generated about $60 million in H20 revenue under them. In February 2026 it granted a license to ship small amounts of H200 products to specific China-based customers. NVIDIA says it has not generated revenue under the H200 program and does not yet know whether any imports will be allowed into China, that the chips must pass an inspection in the United States before shipment, and that any H200 shipped under it is subject to a 25% tariff on import into the United States.
Dropped from the fiscal 2025 text:
- The sentence that China's regulators were investigating whether NVIDIA's compliance with U.S. export controls was discriminatory. The preliminary finding above replaces it.
- The passages on the AI Diffusion IFR (interim final rule). The fiscal 2025 report said the rule, unless modified, would impose a worldwide license requirement on chips such as the A100, H100, H200, B200 and GB200.
- The sentences saying that Blackwell systems need a license for certain shipments, and that NVIDIA had not received licenses to ship the restricted products to China.
The diff page shows an excerpt of the changes with every added and removed word marked; each sentence in full is in Item 1A of either report. NVIDIA's risk factor page tracks the headings across years.
Investments and OpenAI
Item 1A has new sentences on NVIDIA's investments in its ecosystem. They say these investments may produce realized and unrealized gains or losses, that NVIDIA could lose up to the value of the investments, and that many of these companies may not reach profitability. They also say NVIDIA is finalizing an investment and partnership agreement with OpenAI, and that there is no assurance it will be signed or completed.
Margins in the management discussion
The results table in Item 7 shows each line as a share of revenue:
| Share of revenue | Fiscal 2026 | Fiscal 2025 |
|---|---|---|
| Cost of revenue | 28.9% | 25.0% |
| Gross profit | 71.1% | 75.0% |
| Research and development | 8.6% | 9.9% |
Gross margin fell 3.9 points of revenue. The report gives the reasons in Item 7; this comparison counts what changed, it does not assign causes.
Get the same sections for any company
Every 10-K and 10-Q on filings.tools splits into its numbered Items, and annual reports with an earlier one on file have the same comparison against the year before. Open NVIDIA's latest 10-K or any other company the same way. The API returns the same sections as Markdown or JSON, for a model or a script: see the API docs.