BUSINESS OVERVIEW

NVIDIA Corporation

NASDAQ: NVDA

6 September 2026

Evidence base: NVIDIA SEC filings FY2017–FY2026 (10-Ks), 10-Qs through fiscal Q2 2027 (filed 26 Aug 2026),

8-Ks through 3 Sep 2026 and proxy statements; supplemented by primary outside sources — peer company

results releases, US Bureau of Industry and Security, Federal Register, SIA, WSTS, SEMI and NERC.

Not a valuation and not a recommendation.

1. Executive Snapshot

2. What the Company Does

The customer problem is capacity. Training and serving a large AI model requires far more computation than any single processor can deliver, so the work must be spread across thousands of chips that have to behave as though they were one machine. Splitting the work is the easy part; moving data between the chips fast enough that they do not sit idle is the hard part. NVIDIA sells the answer to that second problem, and the chips come with it.

This changes what is actually being sold. Through FY2024 NVIDIA largely shipped HGX boards — components that a server maker built into a machine. The FY2026 10-K describes Data Center offerings as "typically delivered to customers as rack-scale systems, subsystems, or modules," and management attributes the FY2026 gross-margin decline to the business model transitioning "from offering Hopper HGX systems to Blackwell full-scale datacenter solutions." NVIDIA now sells the rack, not the part.

The unit, traced from order to cash

The unit of economics is one rack-scale AI system: a GB300 NVL72 or, from the third quarter of FY2027, a Vera Rubin NVL72 — 72 GPUs and 36 CPUs wired together by NVLink switches, with ConnectX network adapters and Ethernet or InfiniBand switching attached. NVIDIA has never disclosed how many it ships or what one costs; unit volumes and average selling prices are NOT FOUND in any filing, and this is the single largest gap in the disclosure.

What can be traced is the cash cycle around that unit. It begins more than a year before revenue: NVIDIA commits to wafer capacity at TSMC, to CoWoS advanced packaging, and to HBM memory, describing "extended lead times of more than 12 months" for which it has "paid premiums, provided deposits, and entered into long-term supply agreements and capacity commitments" (FY2026 10-K). Contract manufacturers — Hon Hai, Wistron and Fabrinet are the named ones — assemble the systems. NVIDIA then bills a direct customer that is usually not the operator of the machine: an ODM, OEM, system integrator or distributor. The FY2026 10-K estimates that 76% of Data Center revenue billed to Taiwan-headquartered customers was attributable to end customers in the United States and Europe. Cash arrives roughly two months later, on a receivables balance of $63.1bn at 26 July 2026, and the Q2 FY2027 10-Q notes that "for investment-grade customer purchases, we have and may in the future provide longer payment terms ranging from 90 days up to one year to assist customers with large data center builds."

Revenue is transactional, not recurring. "Most of our sales are made on a purchase order basis," the FY2026 10-K states, and "our customers can generally cancel, change, or delay product purchase commitments with little notice to us and without penalty." Contractual remaining performance obligations extending beyond one year were $3.2bn at 26 July 2026 — against $96.2bn of revenue in that quarter alone. There is no backlog cushioning this business in any accounting sense.

Segments, platforms, and what changed in the reporting

Two reportable segments carry the business. Compute & Networking generated $193.5bn of revenue and $130.1bn of segment operating income in FY2026; Graphics generated $22.5bn and $9.2bn (FY2026 10-K). By market platform, FY2026 Data Center revenue of $193.7bn splits into compute of $162.4bn (up 59%) and networking of $31.4bn (up 142%, on NVLink compute fabric for GB200 and GB300 systems plus Ethernet and InfiniBand). Gaming contributed $16.0bn, Professional Visualization $3.2bn, Automotive $2.3bn and OEM & Other $0.6bn.

That platform table no longer exists. From the first quarter of FY2027 NVIDIA reports only Data Center — split into Hyperscale (48.7bn in Q2) and AI Clouds, Industrial & Enterprise (40.3bn) — and Edge Computing ($7.2bn), into which Gaming, Professional Visualization, Automotive and OEM have been folded. Prior periods were recast, and one customer was reclassified between the two Data Center lines in Q2 with a further recast. The practical consequence is that an investor can no longer see Gaming separately, and the FY2017–FY2026 platform series cannot be bridged forward from the filings alone.

Software is described but not measured. NVIDIA sells paid licences to NVIDIA AI Enterprise and vGPU, and its own risk factors concede that "we may fail to sell any meaningful standalone software or services." No software or DGX Cloud revenue figure appears in any filing (NOT FOUND). Every dollar of disclosed revenue is hardware.

3. Industry, Competitive Position and Moat

The industry converts scarce manufacturing capacity into AI compute. It runs from EDA and IP, through TSMC's leading-edge fabrication and CoWoS advanced packaging, through HBM memory from SK hynix, Micron and Samsung, to accelerator designers, then to ODM and OEM assemblers, then to clouds and model developers. The useful question is not how large the industry is but which link keeps the profit, and each participant's own reported margins answer it.

Each figure is from that company's own results release for the period shown; fiscal periods differ and margins are GAAP unless the company reports otherwise. Broadcom's 54% is company-wide GAAP operating margin, not an AI-only figure. SK hynix does not disclose a gross margin.

The pattern is unambiguous and it is not the one the industry narrative implies. The highest margins in the chain are not at the accelerator designer — they are at the memory makers, whose operating margins of 76% and 80% exceed NVIDIA's 66.2%, and at TSMC. The lowest are wherever a participant can be replaced with capital and effort: Dell's infrastructure business earns 15.0% and Supermicro earned a 10.8% gross margin across its 2026 fiscal year. Value accrues to whoever controls a physically constrained step, and it is thin everywhere else.

Where NVIDIA actually sits

NVIDIA's scale is not in question. Data Center revenue of $89.0bn in a single quarter compares with roughly $25.6bn of quarterly AI-relevant silicon at its three nearest rivals combined — Broadcom's AI semiconductor revenue of $16.7bn, AMD's Data Center segment at $6.7bn and Marvell's data-centre revenue of $2.2bn. Intel's $6.3bn data-centre business is overwhelmingly Xeon CPUs; its Q2 2026 release does not mention an accelerator product at all.

The barriers that genuinely bind are narrower than the ones usually cited. The scale-up domain is one: making 72 GPUs act as a single accelerator over NVLink is not something a merchant network vendor currently replicates, and it is the reason a rack is the sellable unit. Pre-purchased supply is a second: NVIDIA has committed $279bn to manufacturing capacity and memory, and no entrant can buy comparable access to CoWoS or HBM inside two years when TSMC's chief executive says on his own earnings call that "our packaging capacity is so tight that now it's limiting my customers' growth" and SK hynix states in its results release that "customer demand exceeds supply capabilities." The software installed base is a third, though it is the softest of them.

Two barriers leak, and the evidence is specific rather than speculative. The scale-out network layer has already gone to Ethernet, where Broadcom's Tomahawk switches are deployed across the hyperscalers building their own accelerators, and NVIDIA's own answer is an Ethernet product — Spectrum-X — rather than a defence of InfiniBand. And CUDA does not bind at the top of the market: AMD's Data Center segment moved from a $155m operating loss to $2.1bn of operating income in four quarters, and OpenAI has separately committed to 6 gigawatts of AMD Instinct and 10 gigawatts of Broadcom-built accelerators of its own design, while Anthropic has contracted for up to a million Google TPUs. Customers with their own kernel engineers are demonstrably able to leave. Whether the long tail of enterprises can is unverified from outside evidence.

Testing NVIDIA's account of itself

Outside evidence supports NVIDIA on two claims and works against it on two others. It supports the claim of accelerator dominance, and it supports the claim that demand exceeds supply — that one is corroborated at four independent points in the chain: TSMC on packaging, SK hynix on memory, Dell's $95bn AI backlog, and $166bn of combined capital expenditure by Microsoft, Amazon, Alphabet and Meta in calendar Q2 2026. It contradicts the claim that China is an addressable opportunity: despite the January 2026 licensing liberalisation, Chinese shipments were under 1% of Data Center revenue and NVIDIA's own Q3 guidance assumes no China data-centre compute revenue at all. And it works against the durability of the margin structure, which Section 5 takes up.

The kind of company that wins here owns a physically constrained step that cannot be reproduced in under three years. TSMC owns one; the memory oligopoly owns another; both out-earn NVIDIA today. NVIDIA is only partly that kind of company. It rents both constraints, and its own scarce assets — the rack-scale interconnect and the developer installed base — are real but are engineering leads rather than physical scarcity. That distinction is what makes the position formidable and, at the same time, contestable in a way that TSMC's is not.

4. Growth Engine

Revenue grew from $60.9bn in FY2024 to $130.5bn in FY2025 to $215.9bn in FY2026, and reached $96.2bn in the single quarter ended 26 July 2026 — up 106% year on year and 18% sequentially. Guidance for the following quarter is $108.0bn, plus or minus 2% (Q2 FY2027 results release, 26 Aug 2026).

Reported growth versus organic growth

Almost none of this is acquired, which is unusual enough to state plainly. NVIDIA has made exactly one revenue-additive acquisition of scale in the decade: Mellanox, closed April 2020 for $7.13bn, which contributed 10% of FY2021 revenue and created the networking line. Since then the deals have bought capability rather than revenue. The December 2025 Groq transaction — $13.0bn paid at closing plus $4bn payable within a year — was a non-exclusive technology licence plus employee hires, and the FY2026 10-K is explicit that "no customer contracts, existing products, or equity interests were purchased." The Hugging Face agreement signed on 2 September 2026, at approximately $11.9bn plus up to $1.0bn of retention equity, has not closed and post-dates every figure in this document. FY2026's 65% growth and FY2027's first-half growth are therefore organic.

The drivers, ranked

• Data Center compute volume and content per rack — structural. Compute revenue rose 59% to $162.4bn in FY2026. The shift from selling boards to selling racks raises the revenue captured per unit of end-customer capacity, because NVIDIA now supplies the CPUs, switches and interconnect that a system builder previously supplied.
• Networking attach — structural, and the clearest management-driven gain. Networking grew 142% to $31.4bn in FY2026 on NVLink compute fabric for GB200 and GB300 systems plus Ethernet and InfiniBand. It is the Mellanox asset compounding six years after purchase, and it is the part of the rack a customer is least able to source elsewhere.
• Hyperscaler and AI-cloud capital expenditure — cyclical, and the largest single dependency. Microsoft, Amazon, Alphabet and Meta together spent $166bn in calendar Q2 2026; Meta guides to $130–145bn for the full year; AWS said on 26 August 2026 it plans to deploy 2 million additional NVIDIA GPUs across 2027 and 2028. This is a budget decision made annually by four companies, not a contracted revenue stream.
• The annual architecture cadence — management-driven. Blackwell, then Blackwell Ultra shipping from Q2 FY2026, then Vera Rubin entering production shipments in Q3 FY2027. Each generation raises performance per rack and resets the competitive comparison, but it also means the ramp risk recurs every year; NVIDIA's FY2025 gross margin was hit by "inventory provisions for low-yielding Blackwell material" on exactly this pattern.
• Edge Computing — cyclical, and now small. The consolidated Gaming, visualisation and automotive businesses grew 27% year on year to $7.2bn in Q2 FY2027, 7.5% of revenue. NVIDIA has also warned that supply constraints are a headwind to Gaming, because the same wafers earn far more in a data centre.
• China — regulatory, and currently a subtraction rather than a driver. Chinese revenue fell from $25.0bn in FY2025 to $19.7bn in FY2026 and Hopper and H200 shipments were each under 1% of Data Center revenue in Q2 FY2027. Section 6 treats the mechanism.

5. Margin, Cash and Capital Allocation

Gross margin is set by two things NVIDIA controls and one it does not. It controls mix — a full rack carries a different margin from a board — and it controls how much inventory it has to write down. It does not control the price of HBM memory. All three moved in FY2026: gross margin fell 3.9 points to 71.1% as the model shifted from Hopper HGX systems to Blackwell full-rack solutions and as a $4.5bn charge was taken on H20 inventory and purchase obligations after the April 2025 export-licence requirement. Inventory provisions totalled $7.2bn in FY2026 against $1.5bn of releases, a net 2.6-point drag.

The first half of FY2027 reversed the charge effect: gross margin returned to 75.0% with a provision drag of only 1.0 point. The next move is the informative one. NVIDIA guides Q3 FY2027 gross margin to 74.0%, and the Q2 10-Q states that supply availability and rising prices for memory and other components "may drive the prices for data center buildouts higher," with the $279bn commitment increase described as "primarily related to the procurement of memory." Input cost has turned from tailwind to headwind, and it is set by suppliers NVIDIA has no ability to substitute.

Operating leverage runs the other way and is the reason margins have held. Research and development fell from 14.2% of revenue in FY2024 to 8.6% in FY2026 while rising 43% in dollars; selling and administrative expense fell from 4.4% to 2.1%. Roughly 42,000 employees, 31,000 of them in R&D, now support a revenue base that has trebled in two years. Operating margin reached 65.9% in the first half of FY2027 against a prior-decade peak of 37.3% in FY2022.

Sources: FY2020, FY2023 and FY2026 10-Ks; Q2 FY2027 10-Q. Cash flow less capex is computed, not a reported line. Four caveats on comparability: reportable segments changed from GPU/Tegra to Compute & Networking/Graphics in FY2021, so segment series do not span this table; the market-platform taxonomy was rebuilt in Q1 FY2027 and the FY2027 H1 Data Center figure is not defined identically to earlier years; FY2023 operating income includes a $1,353m Arm acquisition-termination charge, roughly 5 points of margin; and two stock splits (4-for-1 in July 2021, 10-for-1 in June 2024) make as-reported per-share figures across these columns non-comparable, which is why none are shown.

Cash conversion is deteriorating even as cash grows

FY2026 produced $102.7bn of operating cash flow against $6.0bn of capital expenditure, and the first half of FY2027 produced $74.4bn against $4.4bn. This remains a business that requires almost no fixed capital of its own — the fabs belong to TSMC. But the working-capital picture has changed markedly. Receivables rose from $38.5bn to $63.1bn and inventory from $21.4bn to $31.6bn in six months, with raw materials alone going from $3.8bn to $11.3bn ahead of the Vera Rubin ramp. Days sales outstanding is roughly 60, but the 10-Q's disclosure of payment terms extending to a year for investment-grade buyers means that number is a policy choice as much as a measurement.

Reported net income should not be read as operating performance from FY2026 onward. Other income of $24.1bn in the first half of FY2027 was 20.5% of net income of $118.0bn and consists largely of unrealised marks on an equity portfolio that reached $99bn. NVIDIA's own non-GAAP diluted earnings per share of $2.22 in Q2 FY2027 is below its GAAP $2.46 for this reason. Operating income is the comparable series.

Where the cash has gone

Ranked over FY2017 to the first half of FY2027, capital went first to share repurchases ($135.6bn), then to purchases of equity securities in other companies (roughly $62.3bn), then to employee stock-plan tax withholding ($28.9bn), then to capital expenditure ($20.6bn), the Groq licence ($15.9bn), acquisitions ($11.8bn) and dividends ($11.0bn). Two features of that ranking are informative. Buybacks were suspended entirely in FY2020, FY2021 and FY2022 while the Arm acquisition was pending and resumed at scale in FY2023; the authorisation was raised by $80bn on 18 May 2026, with $99.3bn remaining at 26 July 2026. And the second-largest use of capital is now buying stakes in other companies rather than building anything — a shift Section 7 treats as a risk rather than an allocation preference.

The balance sheet changed shape in the same period. Total debt went from $8.5bn to $33.4bn on a $25.0bn seven-tranche senior note issue completed 18 June 2026 at coupons of 4.250% to 5.625%, against $99.4bn of cash and marketable securities. The quarterly dividend was raised from $0.01 to $0.25 per share on 18 May 2026. A company generating $70bn of free cash flow per half-year chose to borrow $25bn, which is a statement about the pace at which it intends to deploy capital into supply, equity stakes and repurchases rather than about need.

6. Cyclicality, Constraints and What to Monitor

NVIDIA is at or through its own prior peak on every axis that matters. Gross margin of 75.0% matches the FY2025 record; operating margin of 66.2% is a decade high against a prior peak of 37.3% in FY2022; revenue is growing 106% year on year. Nothing in this business is currently below trend. That is the necessary context for every margin and growth figure in this document, and it is the opposite of the situation in which the same numbers would be encouraging.

The company's own history shows what the other end looks like. In FY2019 a crypto-driven channel-inventory correction cut fourth-quarter revenue 24% year on year, took gross margin from 61.9% to 54.7%, and produced $185m of DRAM and component charges; FY2020 revenue fell 7% and operating income 25%. In FY2023 gross margin fell to 56.9% and operating margin to 15.7%. Both episodes followed periods in which demand looked entirely secure.

The transmission mechanism today is larger and runs through the balance sheet rather than the channel. Four companies fund most of the demand: $166bn of hyperscaler capital expenditure in a single quarter. Against that, NVIDIA holds $279bn of non-cancellable purchase and capacity commitments — $92bn due in the remainder of FY2027, $87bn in FY2028 and $88bn in FY2029 — and $31.6bn of inventory, while its customers can cancel purchase orders without penalty. A capital-expenditure pause would not reduce NVIDIA's obligations; it would convert them into charges, exactly as the $4.5bn H20 write-down did when a single export rule removed one market. The industry-level precedent is Micron in FY2023, whose revenue fell 49.5% and whose gross margin went from 45.2% to minus 9.1% in four quarters. That is what happens to whoever holds the scarcity position when the cycle turns; today those holders are the memory makers, and NVIDIA has just committed $279bn to buying from them.

Three constraints bind on the supply side and are documented outside NVIDIA. Advanced packaging capacity at TSMC is limiting customer growth on the company's own statement, and TSMC allocates only 10–20% of its $60–64bn 2026 capital budget to packaging, testing and masks. HBM memory is contracted forward, with SK hynix reporting long-term agreements with around ten customers. And electricity is the constraint furthest from anyone's control: NERC's 2025 Long-Term Reliability Assessment classifies 2026–2030 as high risk, with planned resources falling short of established criteria and summer peak demand projections nearly doubling year over year.

Durable versus borrowed

Leading indicators, and where each is published

7. Risks, Unknowns and Questions for Deeper Work

The cyclical risk is in Section 6. What follows is what cyclicality does not capture, ordered by how badly each compounds with the others.

• NVIDIA increasingly finances the demand for its own product. It held $99bn of equity investments and $25bn of further investment commitments at 26 July 2026, guarantees with maximum gross exposure of $108.5bn — $105bn of residual-value guaranties to SB Energy covering roughly 4.25 gigawatts leased to OpenAI for twenty years, plus $3.5bn of land, power and shell guarantees to AI clouds — and $36bn of six-year cloud service commitments. The 10-Q states the mechanism directly: "if AI clouds do not successfully sell committed capacity to third-party customers, we have agreed to purchase that capacity." In August 2026 NVIDIA signed memoranda with six capital providers to mobilise more than $500bn of third-party capital. Each arrangement is defensible alone; together they mean a downturn in end demand arrives at NVIDIA three times over — as lost revenue, as impaired investments, and as assumed obligations.
• Its customers' funding is the constraint, and NVIDIA says so. The Q2 FY2027 10-Q states that AI clouds and model makers "have significant demand for training and inference compute and currently lack the ability to secure long-term infrastructure contracts and investment-grade financing capacity." The buyers of the most expensive capital equipment in the world cannot finance it conventionally. This is why the guarantees exist, and it makes NVIDIA's revenue a function of credit conditions in a way a hardware business normally is not.
• Concentration compounds in three directions at once. Two direct customers were 22% and 14% of FY2026 revenue; three receivable balances were 25%, 18% and 13% of the total. Those customers order on cancellable purchase orders while NVIDIA holds $279bn of non-cancellable commitments. The asymmetry is the risk, not the concentration by itself.
• Reported earnings now contain a large mark-to-market component. Roughly a fifth of first-half FY2027 net income came from unrealised gains on holdings whose value moves with the same AI capital cycle that drives the operating business. NVIDIA disclosed that a 10% decline in its public equity holdings would have cost $3.9bn as at April 2026. The portfolio does not diversify the earnings stream; it levers it.
• The Groq transaction bought no revenue. $15.9bn has been paid or committed for a non-exclusive technology licence and a team, producing $14.4bn of goodwill and a $2.5bn intangible with no acquired customer contracts, products or equity. The 10-K concedes the payments were "nonrefundable" and that "we may be unable to recover the associated costs." It is the largest capital commitment in the company's history outside supply, and its return is unverifiable from disclosure.
• The regulatory position is a pincer, and one side has already closed. The FY2026 10-K no longer frames China as a risk but as an accomplished loss: NVIDIA was "effectively foreclosed from competing in China's data center computing/compute market," and states that this foreclosure "helped our competitors build larger developer and customer ecosystems to challenge us worldwide." A 25% tariff applies to H200 units imported into the US and "we have been unable to pass along any of the tariff to our customers." Chinese antitrust regulators published a preliminary finding on 15 September 2025 that NVIDIA's compliance with US export controls violated the conditions of their Mellanox approval, which reaches the networking business. Two US bills, the GAIN AI Act and the Remote Access Security Act, are named in the filings as further exposure.
• Disclosure is narrowing as the stakes rise. From Q1 FY2027 Gaming, Professional Visualization and Automotive are no longer reported separately, and one customer moved between the two Data Center sub-lines with prior periods recast again. An investor has less visibility into the composition of this business than a year ago.

What the sources could not answer

These are findings, not gaps in the research. Several would change how an investor sizes the business.

• Unit volumes and average selling prices. NVIDIA has never disclosed how many systems it ships or at what price, so the growth decomposition between volume, price and content per rack cannot be made from filings.
• Software, services and DGX Cloud revenue. Described in the business section, quantified nowhere.
• The maintenance-versus-growth split of capital expenditure, and the identity of the 22% and 14% customers.
• Dollar amounts for several investments the filings reference: the OpenAI arrangement is described in the FY2026 10-K only as being "finalized," with no figure, and the Intel and xAI stake sizes are not disclosed in filings.
• Whether the $1 trillion Blackwell-and-Rubin figure management has cited on earnings calls has any contractual basis. Filed remaining performance obligations beyond one year are $3.2bn, which suggests it does not, and no filing repeats the number.
• The share of hyperscaler AI compute now served by custom silicon rather than merchant GPUs. Neither Amazon nor Meta discloses deployed Trainium or MTIA scale, so the substitution rate cannot be measured from primary sources.

8. Investor Takeaways

• This is an AI data-centre systems business, not a chip business. The sellable unit is a rack of 72 GPUs, and the interconnect that makes those GPUs act as one machine is the part competitors have not reproduced.
• The economic engine is turning pre-purchased scarce capacity — TSMC packaging and HBM memory — into integrated systems at a 75% gross margin, sold to a handful of buyers on cancellable orders.
• The main growth lever is Data Center content per rack and networking attach, both structural; the pace at which they convert to revenue is set by four customers' annual capital budgets, which is not.
• What could break the story is the combination of $279bn of non-cancellable supply commitments and $108.5bn of guarantees meeting a pause in customer funding — a sequence in which NVIDIA absorbs the correction rather than passing it on.
• Monitor gross margin against memory cost, the commitments and guarantee balances, and hyperscaler capital expenditure. Those three lines describe the whole risk.