NVIDIA's fiscal 2026 revenue rose 65% to $215.9 billion, but gross margin fell 3.9 points on a $4.5 billion H20 write-off and the shift to selling full racks, and roughly $8.9 billion of pre-tax profit came from unrealized gains on its own equity investments.
Revenue
$215.9B
+65.5% YoY
Net income
$120.1B
+64.7% YoY
Diluted EPS
$4.90
+66.7% YoY
Operating margin
60.4%
Revenue up 65% to $215.9 billion — but the cost of shipping whole AI systems, not chips, shows up in the margin
NVIDIA's fiscal year 2026 ended January 25, 2026. Revenue was $215,938 million (about $215.9 billion), up 65% from $130,497 million in fiscal 2025. Net income reached $120,067 million and diluted earnings per share $4.90, up 67%.
The growth itself is not the interesting part of this filing. Two things are: gross margin fell almost four percentage points even as revenue grew 65%, and GAAP earnings per share grew faster than the company's own adjusted (non-GAAP) measure — a reversal that says something specific about where fiscal 2026's profit came from.
Gross margin is the share of revenue left after the direct cost of making and shipping the product. Operating margin is what's left after research and selling costs too, but before interest and tax.
Networking, not GPUs, was the fastest-growing line
Data Center revenue of $193,737 million was 90% of the company. Inside it, the split matters:
Compute (the GPUs themselves): $162,361 million, up 59%, which management attributes to demand for the Blackwell platform.
Networking (the switching fabric that ties GPUs together into one machine): $31,376 million, up 142%. The filing names the cause directly — "the introduction and continued ramp of NVLink compute fabric for GB200 and GB300 systems and the growth of Ethernet and InfiniBand platforms."
Networking grew from 11% of Data Center revenue to 16%. That is the clearest financial evidence of the business-model change management describes elsewhere in the filing: NVIDIA has moved "from offering Hopper HGX systems to Blackwell full-scale datacenter solutions" — from selling accelerator boards that a customer assembles into a server, to selling the rack.
Outside the data center, the smaller lines all grew but are now rounding errors against it: Professional Visualization $3,191 million (+70%, helped by the DGX Spark launch), Automotive $2,349 million (+39%), OEM and other $619 million. Gaming at $16,042 million (+41%) is now 7% of revenue.
By reported segment, Compute & Networking revenue was $193,479 million (+67%) with operating income of $130,141 million (+57%), while Graphics revenue was $22,459 million (+57%) with operating income of $9,156 million (+80%). Graphics profit grew faster than Graphics revenue; Compute & Networking profit grew slower than its revenue. The reason is in the next section.
The margin decline is two separate things, and only one of them reverses
Gross margin fell from 75.0% to 71.1%. Cost of revenue rose 91% ($32,639M to $62,475M) against 65% revenue growth. The filing gives two causes:
1. A $4.5 billion H20 write-off. In April 2025 the US government told NVIDIA that exporting its China-market H20 product required a license. NVIDIA took a $4.5 billion charge in the first quarter of fiscal 2026 for "excess inventory and purchase obligations, as the demand for H20 diminished." Licenses were granted in August 2025, but NVIDIA generated only "approximately $60 million in H20 revenue under those licenses" — roughly 1.3% of what it had written off. In February 2026 a further license was granted for small quantities of H200; no revenue has come from it yet, and any H200 shipped under it must be inspected in the US first and will carry a 25% tariff on importation. This is a one-time hit to the fiscal 2026 comparison, but China remains effectively closed: the Q1 FY2027 outlook assumes zero Data Center compute revenue from China.
Total inventory provisions were $7.2 billion in fiscal 2026 against $3.7 billion in fiscal 2025, with $1.5 billion released as previously written-down inventory sold. Net, inventory write-downs cost 2.6 percentage points of gross margin in fiscal 2026 versus 2.3 points in fiscal 2025 — so even excluding the H20 event, write-downs are a persistent margin drag at this scale.
2. Selling systems instead of components structurally lowers the margin percentage. A full rack contains memory, power, chassis, cabling and third-party content that NVIDIA buys and resells. Revenue per unit goes up, but so does the bought-in cost, which compresses the percentage even when gross profit dollars rise — gross profit still grew from $97,858 million to $153,463 million.
The fourth quarter shows the first effect washing out. Q4 revenue was $68,127 million (up 73% year over year, and by itself more than half of all of fiscal 2025) at a 75.0% gross margin — back to the prior-year full-year level, versus 73.0% in Q4 fiscal 2025.
Operating expenses grew 41% to $23,076 million, slower than revenue, so operating leverage was positive. R&D was $18,497 million (+43%), driven by a 29% rise in compensation and benefits and a 79% rise in compute and infrastructure costs — NVIDIA is now a large consumer of its own product class.
GAAP earnings beat non-GAAP — that's backwards, and it matters
GAAP net income was $120,067 million; the company's own non-GAAP net income was lower, at $116,997 million. GAAP diluted EPS grew 67% while non-GAAP diluted EPS grew 60% ($4.77 versus $2.99).
Adjusted earnings are normally higher than GAAP, because companies exclude costs. Here the direction flips because of one line: "Other income, net" was $9,022 million, against $1,034 million a year earlier. The filing attributes this to "unrealized gains in non-marketable and publicly-held equity securities, including gains from our previously announced investment in Intel's common stock," and the non-GAAP reconciliation strips out $8,918 million of such gains.
In plain terms: roughly $8.9 billion of fiscal 2026 pre-tax GAAP profit — equal to about 7% of reported net income — came from the mark-to-market value of NVIDIA's investment portfolio going up, not from selling anything. Those gains are unrealized, reverse if the holdings fall, and are not cash. The $4.90 GAAP EPS headline is genuine under the accounting rules but overstates the operating result; $4.77 is the cleaner read.
This portfolio is growing deliberately. NVIDIA invested $17.5 billion in private companies and infrastructure funds during the year, "primarily to support early-stage startups," and notes that these "include AI model makers that purchase our products directly or through CSPs." It also provided $3.5 billion in land, power, and shell guarantees to early-stage companies. The filing states it is "finalizing an investment and partnership agreement with OpenAI" with no assurance it will be completed. An arrangement where the supplier funds the customer is not necessarily unsound, but it makes a portion of reported demand dependent on NVIDIA's own capital, and it is the mechanism that produced most of the GAAP-versus-non-GAAP gap above.
The effective tax rate also rose, from 13.3% to 15.1%, because fixed-size benefits (stock-compensation deductions, the FDDEI export deduction, the federal research credit) shrank as a percentage against much larger pre-tax income. A $711 million state valuation-allowance release partly offset it.
Cash is real; the commitments behind it are large
Operating cash flow was $102,718 million against $64,089 million, up 60% — closely tracking operating income, which is the check that the profit is converting to cash. Capital expenditure was $6.1 billion (from $3.4 billion), and management expects it to rise again in fiscal 2027.
Two balance-sheet items deserve attention:
Inventory doubled to $21,403 million from $10,080 million — growth of 112% against 65% revenue growth. Accounts receivable rose 67% to $38,466 million, roughly in line with sales, so collections are not the issue; inventory is building ahead of demand.
Inventory purchase and long-term supply and capacity obligations stood at $95.2 billion at year end (disclosed in the auditor's critical audit matter). That is a commitment to buy roughly 44% of a full year of current revenue in future supply.
Both are what you would expect from a company racing to meet demand on an annual product cadence, and both are exactly the exposure that produced the $4.5 billion H20 charge when one market closed abruptly. PricewaterhouseCoopers singled out inventory valuation as its critical audit matter for this reason.
Capital returned to shareholders: $40.4 billion of buybacks (282 million shares) plus $974 million of dividends, against $102.7 billion of operating cash flow. A further $60.0 billion authorization was added in August 2025, leaving $58.5 billion available at year end.
Concentration got worse, not better
For fiscal 2026, one direct customer was 22% of total revenue and a second was 14% — 36% of the company in two accounts. In fiscal 2025 the top three were 12%, 11% and 11%. The largest relationship nearly doubled in share in one year.
Geographic disclosure is harder to read than it looks. Revenue from customers headquartered outside the US fell to 31% from 41%, but NVIDIA changed its methodology in Q3 fiscal 2026 to report by customer headquarters location and recast prior periods; it also estimates that 76% of Data Center revenue from Taiwan-headquartered customers (largely contract manufacturers) ultimately served end customers in the US and Europe. The shift is more a change in how a mostly-American demand base is attributed than a change in where the demand is.
Outlook
Management's guidance for Q1 fiscal 2027, given with these results:
Revenue of $78.0 billion, plus or minus 2% — about 15% above Q4's $68.1 billion — assuming no Data Center compute revenue from China at all.
GAAP gross margin of 74.9% (non-GAAP 75.0%), plus or minus 50 basis points: essentially holding Q4's recovered level rather than extending it.
GAAP operating expenses of about $7.7 billion, including $1.9 billion of stock-based compensation.
A full-year fiscal 2027 tax rate of 17.0% to 19.0% — up again from 15.1%.
Two notes on comparability going forward. From Q1 fiscal 2027, NVIDIA will stop excluding stock-based compensation from its non-GAAP figures, which makes adjusted earnings more conservative and breaks comparison with prior adjusted numbers. And management flags that "supply constraints [will] be a headwind to Gaming in the first quarter of fiscal 2027 and beyond" — the gaming business is being rationed behind data center.
Our read on trajectory: the revenue line is guided to keep compounding, and the 74.9% gross margin guide implies the H20 charge and the early Blackwell system transition were the whole of the margin decline, with no further structural erosion expected near term. The risks are not in the demand line. They are (a) the $95.2 billion of supply commitments against inventory that grew nearly twice as fast as revenue, which converts any demand pause into write-offs rather than just slower growth, as H20 already demonstrated; (b) 36% of revenue in two customers; and (c) the growing share of reported profit and of funded demand that flows through NVIDIA's own investment book rather than through operations. Watch the gap between GAAP and non-GAAP EPS in fiscal 2027: if it widens again, more of the earnings story is portfolio marks.
Takeaway: Fiscal 2026's headline — revenue up 65%, GAAP EPS up 67% — flatters the operating result twice over. About $8.9 billion of pre-tax profit came from unrealized gains on NVIDIA's own equity investments, which is why GAAP EPS of $4.90 exceeded the company's non-GAAP $4.77 rather than trailing it; and the 3.9-point gross-margin decline was driven by a $4.5 billion write-off on China-bound H20 inventory that Q4's 75.0% margin shows has already passed through. The number to track from here is not revenue growth but the $95.2 billion of supply and capacity commitments sitting against $21.4 billion of inventory that doubled while revenue grew 65% — that is where a demand pause would show up first, and fastest.