NVIDIA Corporation (NVDA)

NVIDIA reported revenue of $96.2 billion for the quarter that ended 26 July 2026, up 106% from a year earlier 1. Data Center, the business that sells the computers behind artificial intelligence (AI), brought in $89.0 billion of it 1. Net income was $59.7 billion 1.

The finance chief's commentary credits the ramp of Blackwell Ultra, the newest chips shipping in volume, and says inventory rose to prepare for Vera Rubin, the next generation 2. To keep up, NVIDIA raised its commitments to suppliers from $119 billion to $279 billion in one quarter, mostly for memory 2.

In August it agreed to guarantee, up to $105 billion, the leases of an Ohio data center campus that OpenAI rents to run NVIDIA systems 6. In September it agreed to buy Hugging Face, the platform where developers share open AI models, for about $11.9 billion 5.

On 26 October it said it will invest $3 billion in Meta's equity offering, as Meta expands its use of Vera Rubin systems 18. The open question is the quarter under way: the company guided revenue of $108.0 billion, plus or minus 2%, assuming no data center compute sales in China 1.

NVIDIA was incorporated in California in April 1993 to make graphics chips, and public trading of its stock began on 22 January 1999 4. Its annual report dates its invention of the GPU, the graphics processing unit, to 1999 4.

The turn beyond games began with CUDA in 2006, software that lets programmers use a GPU's many small cores for any heavy calculation 4. In 2012 a neural network trained on NVIDIA GPUs, AlexNet, won the ImageNet image recognition contest, which the company calls the "Big Bang" moment of AI 4.

In April 2020 it completed the purchase of Mellanox, an Israeli maker of data center networking, for $7.13 billion, adding the links that tie many chips into one computer 15104. In fiscal 2021 revenue rose 53% to $16.68 billion, and Mellanox alone contributed 10% of it 13.

Fiscal 2022 brought $26.91 billion, up 61%, as gamers bought RTX 30 Series cards and some buyers used them to mine cryptocurrency 14. In February 2022 NVIDIA and SoftBank called off the planned purchase of Arm, the chip designer, because of significant regulatory challenges 15.

Fiscal 2023 was flat at $26.97 billion 15. Gaming revenue fell 27% as partners worked down unsold cards, and the company took $2.17 billion of inventory charges plus $1.35 billion for the failed Arm deal 15.

Data Center still grew 41% that year to $15.01 billion, led by the big cloud companies, as the first Hopper H100 chips began to ramp 15.

Then generative AI took off: fiscal 2024 revenue more than doubled to $60.9 billion, and Data Center revenue rose 217% to $47.5 billion on Hopper chips such as the H100 1615. Fiscal 2025 reached $130.5 billion, up 114%, and the first Blackwell systems shipped in its fourth quarter 17.

The fiscal 2024 annual report estimated that about 40% of that year's Data Center revenue came from inference, which means running trained models 16. Large cloud providers took more than half of it in the fourth quarter 16.

Fiscal 2026 brought $215.9 billion, up 65%, with Blackwell the majority of Data Center revenue 4. Gross margin, the share of revenue left after production costs, slipped as sales moved from Hopper boards to whole Blackwell racks 4. In December 2025 it paid for a licence to Groq's inference chip technology and hired some of its staff, rather than buying the company 4.

Export rules cost it too: from April 2025 the US government required a licence to ship the H20 data center chip to China 4. NVIDIA took a $4.5 billion charge as demand for the chip dried up 4. By the end of fiscal 2026 it said it was effectively shut out of China's data center market 4.

Data center computers

NVIDIA's main product is the AI factory: racks of its GPUs joined to its own central processors, switches and software, which clouds, AI labs and companies use to train and run AI models 4. Data Center revenue was $89.0 billion of the latest quarter's $96.2 billion, about 93% (computed) 1.

Hyperscale, the largest cloud companies, bought $48.7 billion, and AI clouds, industrial and enterprise buyers $40.3 billion, up 138% from a year earlier 2. Blackwell Ultra drove the quarter, and Vera Rubin began production shipments in the third quarter of fiscal 2027 23.

Buyers are few and large: one direct customer took 16% of revenue in the quarter, and the filing estimates that one unnamed AI research company bought a meaningful amount through cloud providers 3.

Its named rivals include AMD, Huawei and Intel, and cloud giants such as Alphabet, Amazon and Microsoft that design AI chips of their own 4. AMD agreed to supply OpenAI with 6 gigawatts of its GPUs, starting with 1 gigawatt in the second half of 2026 29.

Networking

Networking ties many GPUs into one machine, through NVLink inside a rack and InfiniBand or Ethernet switches between racks 4. It earned $31.4 billion in fiscal 2026, up 142%, the fastest growing line 4. The newest Spectrum-6 switches support co-packaged optics, light-based links built into the switch itself 1.

Edge computing: gaming, PCs and workstations

Edge Computing, the new name for everything outside the data center, brought in $7.2 billion in the quarter, up 27% 2. Blackwell workstations sold strongly, while consumer PC sales slowed as memory and system prices rose 2.

GeForce RTX 50 Series cards are its gaming line, and Tom's Hardware called the top RTX 5090 the "Fastest GPU around (usually)", with a $1,999 price tag 26. The reviewer docked half a star because the Founders Edition card ran hot 26.

RTX Spark, a laptop superchip pairing a Blackwell GPU with a Grace processor and up to 128GB of shared memory, went on sale on 16 October 2026 21. PCWorld called Asus's $7,000 ProArt P16 model "utterly stunning hardware", and Tom's Guide said its 6,144 CUDA cores are "bridging the gap between serious productivity and high-end gaming" 2827.

It also sells RTX PRO chips for design workstations and DGX Spark and DGX Station desktop AI computers for developers 41.

Software, models and Hugging Face

CUDA, its programming platform, is used by more than 7.5 million developers, and NVIDIA AI Enterprise is a paid software suite for companies running AI 4. It also releases open models of its own, such as Nemotron for AI agents and Cosmos for robots and other physical AI 4.

The pending Hugging Face purchase brings the platform where developers upload and download open models and datasets, and NVIDIA has committed to keep it open and to support other chip makers 5.

Cars and robots

Automotive revenue was $2.3 billion in fiscal 2026, up 39%, from DRIVE computers and software for self-driving cars 4. In the latest quarter it widened its DRIVE Hyperion robotaxi platform with Foxconn, VinFast, Uber and HUMAIN, and announced a reference design for humanoid robots 1.

Factories, suppliers and people

NVIDIA is fabless, meaning it designs chips and hires others to make them: TSMC and Samsung make its wafers, SK Hynix, Micron and Samsung supply memory, and Hon Hai, Wistron and Fabrinet assemble products 4. On 21 October it announced capacity deals with six suppliers, including a $1.1 billion advance to KYEC for new testing plants in the US 20.

Its headquarters is in Santa Clara, California, about 3 million square feet it owns and leases 4. At the end of fiscal 2026 it had about 42,000 employees in 38 countries, 31,000 of them in research and development, and its Israel operations employ about 6,200 people 43.

Jen-Hsun (Jensen) Huang has been chief executive since he co-founded the company in 1993, and Colette Kress is its finance chief 4. Ajay Puri, head of worldwide sales for 21 years, is retiring, and Nicholas Parker, after 26 years at Microsoft, was named to succeed him from 24 August 2026 7.

Problems and probes

Export rules still keep it out of most of China: US licences allow small H200 shipments, which were less than 1% of Data Center revenue last quarter, and each carries a 25% US tariff 3. China's government has also discouraged its companies from buying NVIDIA data center products 4.

Reuters reported on 9 September, citing the New York Times, that the Justice Department is examining whether NVIDIA structured its Groq licensing deal to avoid antitrust review 25. NVIDIA said the deal shows "the American system working as designed" 25.

A shareholder class action filed in 2018 over cryptocurrency mining demand is still pending, and competition regulators in the EU, the US, the UK, China and South Korea have sent it broad requests for information 34.

Its guarantees of others' obligations now reach up to $108.5 billion, almost all for the Ohio campus, and are payable only if a tenant defaults 3.

The umbrella, in words

NVIDIA has one class of stock, common stock listed on the Nasdaq Global Select Market 6. Its subsidiaries exhibit lists three significant subsidiaries, all wholly owned: Mellanox Technologies in Israel, NVIDIA International in Delaware and NVIDIA Singapore, and leaves out the smaller ones 10.

It reports stakes in two AI cloud companies whose racks run its Vera Rubin systems: 11.5% of CoreWeave's Class A stock and 9.3% of Nebius Group's Class A shares 11121.

It owns 2.90% of Nokia after a $1.0 billion investment, and agreed to invest $5 billion in Intel common stock alongside joint chip work 3022.

It has invested in AI model makers: $30 billion in OpenAI's latest round, as Sam Altman announced, and a commitment of up to $10 billion in Anthropic 2423. In all, the quarterly report counts $99 billion of equity investments and $25 billion more committed 3.

The Groq deal of December 2025 is a licence, not a purchase: NVIDIA bought no shares, paid $13.0 billion at closing, owed $4 billion more and hired some Groq staff 4.

Control sits with ordinary shareholders: as of 23 March 2026 Huang owned 3.58% of the stock, BlackRock 7.43% and Vanguard Capital Management 7.31% 9.

For the quarter ending in late October, NVIDIA guided revenue of $108.0 billion, plus or minus 2%, and gross margin of 74.0%, plus or minus 50 basis points 1. It expects a tax rate of 16.0% to 18.0% for fiscal 2027 1.

The company says it will ship Blackwell and Rubin systems side by side and faces some supply constraints 3. Its annual report said Rubin is built to cut the cost per token, the unit of AI output, by up to 10x against Blackwell 4.

On 27 October Oracle and NVIDIA said Oracle plans to grow its Vera Rubin clusters to multi-gigawatt scale and to offer HGX Rubin NVL8 servers with eight GPUs 19.

The Hugging Face deal is expected to close in the first half of 2027, once regulators approve 5. The Ohio campus is built in nine phases on 20-year leases, and the first is expected in fiscal 2029 3.

NVIDIA says each generation of its systems at the Ohio site could mean about 1.5 million GPUs and $150 billion to $200 billion of its revenue 2.

Its suppliers have set dates: KYEC targets high-volume testing in the US from April 2027, and Amkor plans a more than 300% rise in interposer wafer capacity by the end of 2028 20.

On 18 May 2026 the board added $80.0 billion to the buyback authorization and raised the quarterly dividend to $0.25 a share from $0.01 3. In June it borrowed $25.0 billion through notes maturing from 2028 to 2056 82.

Its upcoming data center leases run up to twenty years and are expected to start between the third quarter of fiscal 2027 and fiscal 2033 2.

With Hugging Face it also plans a share-based retention program of up to about $1.0 billion for the staff who join NVIDIA 5. Puri is to stay on as a senior adviser during the handover to Parker 7.

How it operates

NVIDIA designs the processors, networking and software behind data-center computing, above all for artificial intelligence. It mostly sells them as whole systems: racks, boards and modules that pair its graphics processors (GPUs) with its own central processors, switches and network adapters, running on CUDA, the software platform that programmers write for. It also sells GeForce graphics cards for gaming PCs, RTX PRO chips for design workstations and the DRIVE computers for self-driving cars. 1213141516

It sells to a short list of buyers. Its direct customers are mostly the companies that build and resell its hardware (server makers, contract manufacturers, distributors, system integrators), alongside cloud providers and AI model makers. Two direct customers alone took 22% and 14% of fiscal 2026 revenue. Behind them stand the indirect customers who end up running the machines: cloud providers, neocloud builders, AI model makers, enterprises and governments. The filing estimates that one unnamed AI research company, buying cloud services from its customers, contributed a meaningful amount of revenue. 17181920

By the company's own geographic tagging, the United States brought in 149.6 billion dollars, 69.3% of fiscal 2026 revenue, Taiwan 19.6% and China including Hong Kong 9.1%, the one region that shrank, down 21.4%. The Taiwan figure mostly shows where the assemblers sit, not where the chips end up: the company estimates that 76% of data-center revenue from Taiwan-headquartered customers went to end customers in the United States and Europe. 2122

NVIDIA owns no chip factories. It is fabless, meaning it designs the chips and hires suppliers for every step of making them. TSMC and Samsung make its silicon wafers; SK Hynix, Micron and Samsung supply the memory; Hon Hai, Wistron and Fabrinet assemble, test and package finished products. The supply chain sits mainly in Asia, and the company says it is expanding into the United States and Latin America. 2324252627

The specifics

Export rules shape this business more than most. Since 2022 the US government has required licenses for its most capable data-center chips bound for China and other listed countries. A license requirement placed on its H20 chips in April 2025 led to a 4.5 billion dollar charge for excess inventory and purchase obligations, and by the end of fiscal 2026 the company said it was effectively shut out of China's data-center market. 28293031

In its own telling, software holds its customers as much as silicon does. Over 7.5 million developers use CUDA and its other tools, and its platforms are sold by virtually every major server maker and cloud provider. It has put more than 76.7 billion dollars into research and development since it was founded, dates its invention of the GPU to 1999, and got its networking business by buying Mellanox in 2020. 3233343536

It names its competitors plainly: AMD, Huawei and Intel in chips; the large cloud companies, among them Alphabet, Amazon, Microsoft, Alibaba and Baidu, designing AI chips of their own; and Arista, Broadcom, Cisco and Marvell, among others, in networking. 373839

What drives growth, line by line

Compute. The largest line by far: 162.4 billion dollars, 75.2% of fiscal 2026 revenue, up 58.9% on the year. The filing credits demand for the Blackwell computing platform, which made up the majority of data-center revenue. 214041

Networking. 31.4 billion dollars, 14.5% of revenue and the fastest-growing line, up 141.5%. The filing credits NVLink, the fabric that ties processors together inside its GB200 and GB300 systems, and growth in its Ethernet and InfiniBand networking. 2142

Gaming. 16.0 billion dollars, 7.4% of revenue, up 41.3% on demand for Blackwell graphics cards. The company expects supply constraints to hold gaming back from the first quarter of fiscal 2027. 2143

Professional Visualization. 3.2 billion dollars, 1.5% of revenue, up 69.9%, on demand for Blackwell workstation chips and the launch of its new DGX Spark. 2144

Automotive. 2.3 billion dollars, 1.1% of revenue, up 38.7%, which the filing puts down to continued adoption of its self-driving platforms. 2145

OEM and Other. The smallest line, at 619.0 million dollars, 0.3% of revenue. 21

What comes next, in the company's own words: new data-center architectures on a one-year cadence, with the Rubin platform expected to begin production shipments in the second half of fiscal 2027. 4647

The read's section on the people who run it is on the People tab.

What the desk infers Inferred

1. Inferred

   The two direct customers that took 22% and 14% of revenue are most likely contract manufacturers that build NVIDIA's rack systems, not cloud companies buying for themselves.

   Because

   • The filing puts both customers mainly in Compute & Networking, the data-center segment. 18
   • Its list of direct customers leads with board makers, distributors, ODMs, OEMs and system integrators, the companies that assemble hardware for others. 17
   • Taiwan-headquartered customers brought in 19.6% of revenue, yet most of their data-center purchases ended up with users in the United States and Europe, the pattern of assemblers in Taiwan shipping racks to American buyers. 2122
   • It names Hon Hai and Wistron among the contract manufacturers it works with. 26
   Confidence 65 of 100Wrong if A filing or a disclosure by either customer that names it as a cloud provider or an AI model maker buying directly for its own use.
2. Inferred

   The data center is now the whole company in all but name: its fortunes follow the building plans of a few cloud and AI buyers far more than the PC and gaming cycle it grew up on.

   Because

   • Compute and Networking together made up nearly nine tenths of revenue. 21
   • Two direct customers took more than a third of revenue between them. 18
   • The indirect buyers the filing names are cloud providers, neocloud builders, AI model makers and enterprises. 19
   Confidence 80 of 100Wrong if A fiscal year in which gaming, professional visualization and automotive together return to more than a fifth of revenue.
3. Inferred

   Manufacturing capacity at a few Asian suppliers, not demand, has been the limit on how fast NVIDIA can ship, which is why it pays premiums and deposits for capacity far ahead.

   Because

   • It says it may place non-cancellable orders ahead of its usual lead times, pay premiums or provide deposits to secure supply and capacity. 48
   • Some of its lead times have run beyond twelve months. 49
   • It expects supply constraints to hold back gaming. 43
   • Its wafers come from TSMC and Samsung and its supply chain is mainly in Asia. 2427
   Confidence 60 of 100Wrong if A year in which the company cuts purchase commitments and books large inventory write-downs for weak demand, showing demand rather than supply set the pace.

NVIDIA pioneered accelerated computing to help solve the most challenging computational problems. NVIDIA is now a data center scale AI infrastructure company reshaping all industries.

We have invested over $76.7 billion in research and development since our inception, yielding inventions that are essential to modern computing.

What it warns about, in its own words

• Long manufacturing lead times and uncertain supply and capacity availability, combined with a failure to estimate customer demand accurately has led and could lead to mismatches between supply and demand.
• A significant amount of our revenue stems from a limited number of partners and distributors and we have a concentration of sales to customers who purchase directly or indirectly from us, and our revenue could be adversely affected if we lose or are prevented from selling to any of these customers.
• Competition could adversely impact our market share and financial results.
• We are subject to complex laws, rules, regulations, and political and other actions, including restrictions on the export of our products, which may adversely impact our business.
• Defects in our products have caused and could cause us to incur significant expenses to remediate and could damage our business.

Past

Revenue grew from $5.0 billion in fiscal 2016 to $216 billion in fiscal 2026, up 45.8% a year, rising in 9 of 9 years with no fall from a peak, and net income was positive in all 10 years. Over the last three fiscal years, the diluted share count shrank about 0.7% a year, and dividends and buybacks returned about 44% of operating cash flow. Officers and directors filed 262 sales, with 234 under scheduled 10b5-1 plans, and 0 purchases of their own stock on the open market in the year to 2026-10-13. The record shows a company that multiplied revenue eighteenfold in eight filed years while narrowing to one engine: nine of every ten dollars now arrive from data centers, and its leaders sold their own shares 482 times against zero purchases; both facts are the past. 12

Present

The annual report filed 2026-02-25 for the fiscal year ended 2026-01-25 carries the full account. Return on equity stands at about 76 cents of profit for every dollar of shareholders' money, gross profitability at about 74 cents per dollar of everything the company owns, leverage at about 24 cents owed per dollar of assets, and the share count shrank 1.2% over the year. Revenue of $216 billion in fiscal 2026 ranks in the top 1% by size among its industry family, a net margin of 55.6% ranks in the top 5%, and revenue growth of 100.5% a year over the last three filed years ranks in the top 9%. Jen Hsun Huang is president and CEO, filing under that title since 2006-03-14, 20.6 years; the 19 officers and directors filing in the last fifteen months have been on the record a median of 11.6 years; 9 officer or director changes were reported on Form 8-K since 2023-10-21, against the 19 people in charge now. 31

Future

A quarterly report (Form 10-Q) for the quarter ending 2026-10-31 is due by 2026-12-10. An annual report (Form 10-K) for the fiscal year ending 2027-01-31 is due by 2027-04-01. Registered call: "The quarterly filing shows data centers and networking still above nine-tenths of revenue" (confidence 85 in 100; graded by when the 10-Q for the quarter is filed; because the last four annual filings show that line rising from six-tenths to nine-tenths of revenue). A 25% tariff on certain imported chips took effect 2026-01-15, with a data-center carve-out under review; imports for uses that expand the US technology supply chain, including data centers and consumer electronics, are exempt, and Commerce was asked to report on the data-center chip market by July 1, 2026 so the tariff could be modified. Registered call: "Filed revenue is higher year over year in the FY2027 annual report despite the tariff regime" (confidence 80 in 100; graded by when the FY2027 10-K is filed; because the two shocks on record, 2020 and 2023, each produced one flat or down year followed by growth, and the exemption currently covers its main market). 34

How the company operates From the company read

NVIDIA designs the processors, networking and software behind data-center computing, above all for artificial intelligence. It mostly sells them as whole systems: racks, boards and modules that pair its graphics processors (GPUs) with its own central processors, switches and network adapters, running on CUDA, the software platform that programmers write for. It also sells GeForce graphics cards for gaming PCs, RTX PRO chips for design workstations and the DRIVE computers for self-driving cars. 678910

It sells to a short list of buyers. Its direct customers are mostly the companies that build and resell its hardware (server makers, contract manufacturers, distributors, system integrators), alongside cloud providers and AI model makers. Two direct customers alone took 22% and 14% of fiscal 2026 revenue. Behind them stand the indirect customers who end up running the machines: cloud providers, neocloud builders, AI model makers, enterprises and governments. The filing estimates that one unnamed AI research company, buying cloud services from its customers, contributed a meaningful amount of revenue. 11121314

By the company's own geographic tagging, the United States brought in 149.6 billion dollars, 69.3% of fiscal 2026 revenue, Taiwan 19.6% and China including Hong Kong 9.1%, the one region that shrank, down 21.4%. The Taiwan figure mostly shows where the assemblers sit, not where the chips end up: the company estimates that 76% of data-center revenue from Taiwan-headquartered customers went to end customers in the United States and Europe. 1516

NVIDIA owns no chip factories. It is fabless, meaning it designs the chips and hires suppliers for every step of making them. TSMC and Samsung make its silicon wafers; SK Hynix, Micron and Samsung supply the memory; Hon Hai, Wistron and Fabrinet assemble, test and package finished products. The supply chain sits mainly in Asia, and the company says it is expanding into the United States and Latin America. 1718192021

New moves

2026-09-03: Form 8-K, Current report, item 8.01, another event the company chose to report. 2026-08-26: Form 10-Q, Quarterly report for the period ended 2026-07-26, with the quarterly money story and trend by quarter; Financial health reads the newest balance sheet. 2026-08-17: Form 8-K, Current report, item 1.01 (major agreement signed), item 2.03 (new debt or obligation), item 7.01 (investor disclosure under Regulation FD). 2026-07-02: Form 8-K, Current report, item 5.02 (officer or director change). 3

2026-10-21, MarketWatch: "Heckler disrupts Nvidia CEO Huang's speech at Berlin conference. " 2026-10-21, CNBC: "Nvidia announces plug and play Vera Rubin racks that can be shipped on a truck. " 2026-10-20, CNBC: "'Open your doors for AI': German chancellor urges businesses to embrace AI as Nvidia announces new investments. " 2026-10-15, CNBC: "TSMC chip capacity is 'fierce' battleground for AI customers, amid record quarter. " 2026-10-15, CNBC: "Nvidia CEO Jensen Huang urges U. S.-China dialogue on 'potentially dangerous' AI. " 2026-10-14, The Guardian: "Nvidia, worth $5.5tn, donates $250m to UK for AI safety research. " 2026-10-12, CNBC: "Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia. " 2026-10-15, MarketWatch: "Micron, Nvidia and other chip stocks face renewed pressure as AI fears swell. " 5