NVIDIA Corporation Overview
NVIDIA is no longer usefully described as a graphics chip company. It is the general contractor of the global AI build-out: it designs the accelerators, CPUs, DPUs, switches, optics and rack-scale systems that constitute an "AI factory," writes the CUDA software layer that makes them programmable, and increasingly underwrites the financing and power infrastructure that lets customers buy them. In FY2026 it generated $215.9 billion of revenue at a 71.1% gross margin and $120.1 billion of net income — profitability without precedent at this scale. Its competitive position rests on three reinforcing assets: a two-decade CUDA software estate with roughly five million developers; annual full-stack platform cadence (Hopper → Blackwell → Vera Rubin → Feynman); and privileged access to constrained TSMC advanced packaging and HBM supply. Its principal vulnerabilities are the mirror image of that strength: extreme customer concentration, a supply chain anchored in Taiwan, exclusion from China, and a demand base increasingly financed by NVIDIA itself.
At FY2026 year end NVIDIA employed approximately 42,000 people across the United States and 37 other countries (38 in total), of whom approximately 31,000 were in research and development. More than 80% of the workforce holds technical roles and more than half hold an advanced degree. FY2026 voluntary turnover was 3.7%, against a semiconductor industry norm in the mid-teens; over 40% of new hires came through employee referrals (Source: FY2026 Form 10-K, Item 1, Human Capital).
The company's own characterisation
NVIDIA describes itself in its FY2026 Form 10-K and in every subsequent press release as "the world leader in AI and accelerated computing." The 10-K frames the business around the proposition that two simultaneous platform transitions are under way — from general-purpose CPU computing to accelerated computing, and from hand-written software to machine-learned software — and that NVIDIA supplies the full computing stack for both. Management's operative metaphor, repeated by Jensen Huang through FY2026 and FY2027 and reproduced in the FY2026 Sustainability Report, is that AI is a "five-layer cake": energy at the base, then chips, then infrastructure/systems, then models, then applications. NVIDIA's strategy is to participate in — and where possible set the standards for — every layer above energy, and to guarantee the layer below it.
Independent characterisation
NVIDIA is a fabless systems company with a software moat and, since FY2026, a balance-sheet financing arm. Four things are true simultaneously:
-
It sells systems, not chips. The transition from Hopper HGX boards to Blackwell full-scale rack solutions (GB200/GB300 NVL72) and now Vera Rubin NVL72 moved the average unit of sale from a ~$30,000 accelerator to a ~$3 million rack. The 10-K explicitly attributes the FY2026 gross-margin decline to this business-model transition. The corollary is that NVIDIA now captures roughly 70% of the bill of materials of a new AI data centre rather than 30–40%.
-
It monetises almost entirely through hardware. Despite an enormous software estate (CUDA, CUDA-X, NIM, NeMo, Omniverse, AI Enterprise, Dynamo, Isaac, DRIVE, Cosmos), NVIDIA does not separately disclose software revenue. Management has periodically indicated a software and services run-rate in the low single-digit billions — under 2% of revenue. Software's function is demand creation and lock-in, not P&L. Deferred revenue of $6.4 billion at 26 July 2026 (of which $2.8 billion is customer advances) is small relative to revenue; NVIDIA is a transactional hardware business, not a subscription business.
-
It is becoming a capital allocator in its own value chain. By 26 July 2026 NVIDIA held $47.9 billion of non-marketable (privately held) equity securities, up from $3.4 billion eighteen months earlier, plus $42.8 billion of marketable equity securities and $5.0 billion of long-dated locked-up public equity in Other assets. It had $25 billion of remaining equity-investment commitments, $3.5 billion notional of land/power/shell guarantees for AI-cloud partners, and — announced August 2026 — up to $105 billion of residual-value guarantees supporting an OpenAI-affiliated tenant in Ohio. Gains from equity securities contributed $23.7 billion of pre-tax income in H1 FY2027, roughly 17% of pre-tax income.
-
Its customers are few and enormous. Five direct customers represented 22%, 14%, 13%, 11% and 10% of accounts receivable at 26 July 2026 — 70% of receivables in five names. Three customers had represented 25%, 18% and 13% at 25 January 2026. NVIDIA also disclosed in the FY2026 10-K that a single "AI research and deployment company" contributed a meaningful amount of revenue indirectly, by purchasing cloud services from NVIDIA's direct customers.
Revenue model mix
| Revenue type | Characterisation | Approximate share |
|---|---|---|
| Product (hardware/systems) | Accelerators, CPUs, DPUs, NICs, switches, optics, rack-scale systems, boards, SoCs | >95% of revenue |
| Software licences & support | AI Enterprise, Omniverse, DRIVE, vGPU; hardware/software support contracts | Low single-digit % (not separately disclosed) |
| Cloud/DGX Cloud services | Capacity resold from CSPs; NVIDIA holds $29 billion of cloud service agreement commitments as of 26 July 2026 | — |
| Licensing / IP | Historically Intel/other cross-licences; Groq LPU technology licence (in-bound, not out-bound) | Immaterial |
Value-chain position and customer types
NVIDIA is fabless. It designs silicon; TSMC fabricates it (N4P/N3P today, TSMC A16 1.6nm targeted for Feynman in 2028); TSMC CoWoS and, increasingly, alternative advanced packaging assemble it; SK hynix, Micron and Samsung supply HBM; Foxconn, Wistron, Quanta, Dell, HPE, Supermicro and Lenovo integrate systems. NVIDIA's direct customers are add-in-board makers, distributors, ODMs, OEMs, cloud service providers, AI model makers and system integrators. End markets served: hyperscale cloud, AI-native labs, neoclouds, sovereign/national AI programmes, enterprise IT, telecommunications (AI-RAN), scientific computing/HPC, automotive and robotics, healthcare and life sciences, financial services, media and entertainment, gaming, and professional design/visualisation.
Strategy
Stated strategy — themes from FY2026 disclosure and management commentary
Management's articulated strategy rests on five themes, all verbatim or near-verbatim from FY2026 and FY2027 disclosure:
-
Two platform transitions. From general-purpose to accelerated computing, and from human-written to machine-learned software. NVIDIA positions itself as the only vendor able to serve both.
-
The AI factory. Huang's framing at Q4 FY2026: AI infrastructure is "the factories powering the AI industrial revolution." The unit of sale is a factory, not a chip. NVIDIA DSX, launched Q2 FY2027, productises this as a complete build-and-operate playbook.
-
Compute is revenue. Huang, Q2 FY2027: "AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue." The claim is that AI infrastructure has moved from a research cost centre to a revenue-generating asset, which changes the durability of the demand curve.
-
Ubiquity. Huang, Q1 FY2027: NVIDIA is "the only platform that runs in every cloud, powers every frontier and open-source model, and scales everywhere AI is produced — from hyperscale data centers to the Edge."
-
Broadening beyond hyperscalers. The ACIE construct is a deliberate strategic signal that AI clouds, industrial deployments, enterprise and sovereign programmes are durable long-term markets, not secondary businesses.
Strategic initiatives announced in the last 24 months
| Date | Initiative | Detail |
|---|---|---|
| Sep 2025 | OpenAI strategic partnership | LOI for at least 10 GW of NVIDIA systems; intent to invest up to $100 billion. Subsequently scaled to $30 billion equity and 5 GW |
| Dec 2025 | Groq transaction | Approximately $20 billion cash for assets plus non-exclusive inference IP licence; acquihire of Ross, Madra and core engineering. Entry into non-GPU inference silicon |
| Jan 2026 | CoreWeave investment | $2 billion in Class A shares at $87.20 |
| Feb 2026 | Anthropic Series G | Up to $10 billion pledged into a $30 billion round at a $380 billion post-money valuation, alongside Microsoft's up to $5 billion |
| Jan 2026 | xAI Series E | Participation as strategic investor in a $20 billion round; separately $7.5 billion equity plus $12.5 billion debt structure through an SPV for Colossus 2 GPU purchases |
| Mar 2026 | GTC 2026 | Vera Rubin in production; Feynman roadmap to 2028; CPO to volume-production planning; $1 trillion Blackwell + Vera Rubin order outlook through 2027 |
| May 2026 | Reporting framework change | Data Center (Hyperscale/ACIE) and Edge Computing; explicitly to "better reflect current and future growth drivers" |
| May 2026 | Capital return reset | Dividend raised from $0.01 to $0.25 per quarter (25x); $80 billion added to buyback authorisation; commitment to return approximately 50% of free cash flow |
| Jun 2026 | Debt capital markets entry | $25 billion across seven tranches; $25 billion commercial paper programme |
| Aug 2026 | Third-party financing platforms | Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR — over $500 billion of third-party capital, subject to definitive agreements |
| Aug 2026 | Ohio / SB Energy | Up to $105 billion of residual-value guarantees for an OpenAI affiliate at PORTS-Pike; $1.5 billion equity in SB Energy; SB Energy and SoftBank to build 10 GW of power and invest at least $4.2 billion in regional grid infrastructure |
| Aug 2026 | AWS expansion | 2 million additional GPUs plus Vera CPUs, integrated and standalone |
| Aug 2026 | Sovereign AI | Japan national AI infrastructure (a world first); Korea gigawatt-scale build with SK Telecom, NAVER and Brookfield; 35 new European AI/HPC supercomputers |
| Aug 2026 | SK hynix | Multiyear technology partnership on next-generation memory |
| Aug 2026 | Open Secure AI Alliance | Industry consortium on AI safety and security |
| Sep 2026 | Hugging Face | $12.93 billion acquisition — moves NVIDIA up the stack into the open-model distribution layer |
Management's medium-term financial targets
| Guidance item | Value | Source |
|---|---|---|
| Q3 FY2027 revenue | $108.0 billion ±2% | Q2 FY2027 release, 26 Aug 2026 |
| Q3 FY2027 gross margin (GAAP and non-GAAP) | 74.0% ±50 bp | Q2 FY2027 release |
| Q3 FY2027 operating expenses | $9.2 billion GAAP / $9.0 billion non-GAAP | Q2 FY2027 release |
| FY2027 tax rate | 16.0%–18.0% | Q2 FY2027 release |
| Q4 FY2027 gross margin trough | 71%–72% | CFO commentary, Q2 FY2027 call |
| FY2028 revenue growth | approximately 70% — the first year-ahead guidance in company history | CFO Kress, Q2 FY2027 call |
| FY2028 gross margin | 72%–73% | CFO Kress, Q2 FY2027 call |
| China Data Center compute in outlook | zero assumed | All FY2027 guidance to date |
| Capital return policy | approximately 50% of free cash flow | Committed May 2026 |
| Order outlook | $1 trillion of Blackwell and Vera Rubin through 2027 | GTC 2026 keynote |
Applying the 70% growth guidance to consensus FY2027 revenue of approximately $396 billion implies FY2028 revenue of roughly $673–700 billion, versus a prior consensus near $570 billion. Huang framed this as a supply ceiling, not a demand estimate: "Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%." Kress said supply would "remain a bottleneck" at least through the end of FY2028, and projected $1.3 trillion of FY2028 capex from the top five hyperscalers against a cloud backlog above $2 trillion.
Products & Services
Data Center — GPUs and accelerators
NVIDIA Blackwell (B100 / B200 / GB200 Grace Blackwell Superchip; GB200 NVL72) — Announced GTC March 2024, volume ramp from Q4 FY2025. Dual-reticle die, 208 billion transistors, second-generation Transformer Engine, FP4/FP6 support, fifth-generation NVLink. GB200 NVL72 is a liquid-cooled rack of 72 Blackwell GPUs and 36 Grace CPUs presented as one NVLink domain. Target customer: hyperscalers and frontier labs for training and large-scale inference. Priced at system level; rack ASPs in the low millions of dollars. Blackwell led every category of MLPerf Training 6.0 and of AgentPerf, the first agentic-AI infrastructure benchmark (Q2 FY2027 release).
NVIDIA Blackwell Ultra (B300 / GB300 NVL72) — Ramped through FY2026 and FY2027; NVIDIA described GB300 NVL72 as one of its fastest product ramps. Blackwell Ultra infrastructure was the named driver of Q2 FY2027 Data Center revenue. B300 lead times improved from approximately 36 weeks to 18 weeks by early 2026 as the Rubin transition approached.
NVIDIA Vera Rubin (Rubin GPU + Vera CPU) — the current flagship. Announced at GTC 2025, in full production as of Q2 FY2027. The platform packages seven chip types into five rack-scale computers operating as a single AI supercomputer: Vera CPUs, Rubin GPUs, NVLink 6 switches, ConnectX-9 NICs, BlueField-4 DPUs, Spectrum-X co-packaged-optical NICs, and Groq 3 LPUs. Headline platform specifications disclosed at GTC 2026: 3.6 exaflops of compute and 260 TB/s of all-to-all NVLink bandwidth, with NVIDIA claiming up to a 10x reduction in inference token cost versus Blackwell; third-party analysis (SemiAnalysis) put the gain at roughly 50x tokens per watt versus Hopper H200. Deployed at or being deployed by CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius as of July 2026; AWS added Vera Rubin instances.
NVIDIA Vera (CPU) — Unveiled Q2 FY2027 as "the first CPU built for AI agents." Arm-based successor to Grace. Adopted standalone as well as coupled to Rubin: SpaceXAI committed to Vera CPUs for agentic AI, and AWS agreed to deploy Vera CPUs both integrated with Rubin and standalone (August 2026). This is a material new revenue pool — NVIDIA has not previously sold CPUs decoupled from GPUs at scale.
NVIDIA Groq 3 LPX — In full production as of 24 August 2026. Interactive-inference accelerator derived from the Groq LPU architecture licensed in December 2025. Rack form factor houses 256 LPUs and sits alongside Vera Rubin racks. Coupled to Rubin through NVIDIA Dynamo, which disaggregates inference — prefill to Rubin, decode to the LPU — for a claimed 35x throughput per megawatt versus Blackwell alone. This is NVIDIA's first non-GPU compute architecture at scale.
NVIDIA Hopper (H100 / H200 / GH200) — Prior generation, still shipping. H20 was the China-compliant Hopper derivative; production of China-configured H200 was halted in March 2026. Hopper shipments to China were under 1% of Data Center revenue in Q2 FY2027.
Roadmap beyond Rubin. Rubin Ultra (2027) — four reticle-sized compute dies per package, roughly 100 PFLOPS NVFP4, 1 TB HBM4e, approximately 32 TB/s memory bandwidth, approximately 3.6 kW per package, on the Kyber rack architecture (Vera Rubin Ultra NVL576, approximately 600 kW per rack with 800V DC distribution). Feynman (2028) — first NVIDIA use of 3D-stacked GPU dies, custom HBM (HBM4E/HBM5 class), fabricated on TSMC's A16 1.6nm node, paired with the Rosa CPU and the LP40 LPU built with the Groq team, BlueField-5, NVLink-8, and Kyber scaled to NVL1152. Intel supplies EMIB advanced packaging for the Feynman I/O die. Reports in July 2026 suggested Kyber could slip toward 2028; NVIDIA responded that its "roadmap is intact" without publishing a revised schedule.
Data Center — networking, storage and optics
| Product | Description | Status |
|---|---|---|
| NVLink / NVLink Switch | Scale-up GPU-to-GPU fabric; NVLink 6 in Vera Rubin, NVLink-8 in Feynman | Shipping / roadmap |
| Quantum InfiniBand | Scale-out low-latency fabric for training clusters | Shipping |
| Spectrum-X Ethernet | AI-optimised Ethernet; Spectrum-6 switch systems supporting both pluggable and co-packaged optics announced Q2 FY2027 and shipping into gigascale AI factories | Shipping |
| ConnectX-9 SuperNIC | Rubin-generation NIC | Shipping |
| BlueField-4 DPU / Vera BlueField-4 STX | Data processing unit; STX variant announced Q2 FY2027 adds agentic-AI storage processing with in-silicon security | Shipping |
| Co-packaged optics (CPO) | Announced at GTC 2026 as entering volume-production planning; debuts on Rubin Ultra NVL576 and Feynman NVL1152 | Ramping |
| NVIDIA DSX | Complete reference "playbook" platform for designing, building and operating AI factories; underpins the Ohio and Korea deployments | Launched Q2 FY2027 |
Data Center networking revenue reached $14.8 billion in Q1 FY2027, up 199% year on year — the fastest-growing disclosed line in the company.
Edge Computing — Gaming, PC and consoles
| Product | Description | Latest generation |
|---|---|---|
| GeForce RTX | Consumer discrete GPUs; ray tracing, DLSS, Tensor Cores | RTX 50 Series (Blackwell), launched 2025. No RTX 60 announcement as of September 2026; a Rubin-based consumer line (GR20x die family) is rumoured for H2 2027 but unconfirmed |
| NVIDIA RTX Spark | 1-petaflop superchip for Windows PCs with the full CUDA and RTX stack; joint reinvention of the Windows PC with Microsoft | Announced Q2 FY2027; shipping October 2026 |
| GeForce NOW | Cloud gaming subscription service | Tiered subscription |
| Console SoCs | Nintendo Switch family SoCs; reported in OEM & Other | Shipping |
| Local AI stack | RTX/DGX optimisations for DeepSeek v4 Flash, Diffusion Gemma, Nemotron 3.5 Lightning, Qwen 3.8; agent harnesses Hermes Agent and OpenClaw | Launched Q2 FY2027 |
Edge Computing — Professional Visualization and workstations
| Product | Description | Notes |
|---|---|---|
| RTX PRO / Quadro workstation GPUs | Desktop and mobile professional graphics | ProViz revenue reached a record in Q4 FY2026 |
| DGX Station for Windows | Deskside AI supercomputer; GB300 Grace Blackwell Ultra Desktop Superchip, 748 GB coherent memory, up to 20 PFLOPS FP4 | Announced Q2 FY2027; OEM availability (ASUS and others) expected Q4 CY2026 |
| DGX Spark / Vera Rubin Spark | Personal AI development systems | Vera Rubin Spark targeted 2027–2028; Rosa Feynman Spark 2029–2030 |
| Omniverse / Omniverse Enterprise | Simulation, digital twin and USD-based 3D collaboration platform | Licensed per-seat/enterprise |
| vGPU software | Virtualised GPU for VDI and enterprise | Licensed |
Edge Computing — Automotive and robotics
| Product | Description | Notes |
|---|---|---|
| DRIVE AGX Hyperion 10 | Reference compute and sensor architecture for safe, scalable Level 4 fleets | Introduced Q3 FY2026; expanded in Q2 FY2027 to a "robotaxi-ready" global platform with Foxconn, VinFast, Uber and HUMAIN |
| DRIVE Thor / DRIVE OS | Centralised AV compute SoC and operating system | Shipping |
| Alpamayo / Alpamayo 2 Super | Open reasoning models, simulation tools and datasets for autonomous driving; Alpamayo 2 Super released for commercial use | Alpamayo unveiled Q4 FY2026; 2 Super in Q2 FY2027 |
| Cosmos 3 | Fully open frontier omnimodel for physical AI (world simulation) | Launched Q2 FY2027 |
| Isaac / Isaac GR00T Reference Humanoid Robot | Robotics development platform; first open humanoid robot reference design, built on Jetson Thor | Announced Q2 FY2027 |
| Jetson Thor / Jetson Orin Nano 2 | Edge AI compute modules; Orin Nano 2 redefines entry-level edge AI for robots, drones and vision AI | Orin Nano 2 announced 25 August 2026 |
| Halos for Robotics | Industry-first full-stack safety system for physical AI, unifying AI compute and safety | Announced Q2 FY2027 |
| AI-RAN | Accelerated radio access network platform for telecom operators | Included in Edge Computing |
Uber partnership (announced Q3 FY2026): scale the world's largest Level 4-ready mobility network starting 2027, targeting 100,000 vehicles.
Software and AI platform
| Product | Description |
|---|---|
| CUDA / CUDA-X | Parallel computing platform and 400+ accelerated libraries; expanded in Q2 FY2027 alongside PhysicsNeMo |
| NVIDIA AI Enterprise | Supported enterprise AI software suite; per-GPU annual subscription |
| NIM / NeMo / Nemotron | Inference microservices, model customisation framework, and NVIDIA's open model family (Nemotron 3.5 Lightning) |
| Dynamo | Disaggregated inference serving layer; the software that makes the Rubin + Groq LPX pairing work |
| NVIDIA Agent Toolkit | Agent construction toolkit, expanded with PhysicsNeMo in Q2 FY2027 |
| BioNeMo / BioNeMo Agent Toolkit | Life-sciences domain models and agent tooling; launched Q2 FY2027 |
| NemoClaw | Enterprise agentic stack introduced at GTC 2026 |
| Confidential Computing | GPU-level confidential inference; disclosed Q2 FY2027 as in use in Apple's Private Cloud Compute |
| Hugging Face (pending) | 3 million models, 500,000 datasets, 1 million applications, 18 million developers, 200,000 companies; to remain an open platform per Huang |
Financial Narrative
Income statement: notes and sources
FY2022 total operating expenses exclude the $1,353 million Arm acquisition termination cost recorded in FY2023; FY2023 total operating expenses of $11,132 million include it. EBITDA is an analyst computation (operating income plus D&A) and is not a NVIDIA-reported measure.
Margin profile (%)
Revenue CAGR. FY2021 ($16,675 million) to FY2026 ($215,890 million): five-year CAGR of 66.9%. FY2022 to FY2026 (four years): 68.3%. Net income CAGR FY2022–FY2026: 87.4%.
Balance sheet
Post-FY2026 balance-sheet transformation (as of 26 July 2026): total assets $320,272 million; shareholders' equity $228,984 million; cash and equivalents $22,443 million; marketable debt securities $34,143 million; marketable equity securities $42,783 million; non-marketable securities $51,157 million (including $47,898 million of privately held equity and $3.3 billion of equity-method investments in infrastructure financiers); accounts receivable $63,059 million; inventories $31,575 million (raw materials $11,341 million — a 3.0x increase in six months, consistent with the Rubin ramp and memory pre-buying); total debt $33,366 million following the June 2026 issuance; total liabilities $91,288 million.
Cash flow: notes and sources
FY2026 dividends and buybacks are derived from the disclosed $41.1 billion of total FY2026 capital returned and the $0.01 quarterly dividend on approximately 24.4 billion shares.
H1 FY2027 (six months to 26 July 2026): operating cash flow $74,421 million; capex $4,434 million; free cash flow $69,895 million; buybacks $39,044 million; dividends $6,290 million; $24,896 million of net debt proceeds; $2,944 million paid to Groq, Inc. under the licence agreement (classified in financing activities); purchases of equity securities $42,404 million.
Ratio analysis: notes and sources
ROIC computed as operating income after the effective tax rate divided by total debt plus ending equity. Net debt is negative (net cash) in every year, so net debt/EBITDA is shown as a negative multiple. Interest coverage uses reported interest expense.
Commentary on trends, inflections and drivers
FY2022 — the crypto-inflated peak of the old NVIDIA. Revenue of $26.9 billion was still 46% Gaming. Gross margin of 64.9% was a graphics-company margin. The Arm acquisition was in flight and would fail.
FY2023 — the trough. Revenue was flat but the composition and profitability collapsed. Gaming fell 27% as crypto demand evaporated and channel inventory had to be cleared; gross margin fell 800 basis points to 56.9%; the $1,353 million Arm termination charge and $2.2 billion of inventory provisions crushed operating income to $4.2 billion. This is the single most important year for understanding NVIDIA's downside: operating income fell 58% on flat revenue. The operating leverage that works so violently upward works equally violently downward.
FY2024 — the AlexNet moment monetised. Data Center revenue tripled to $47.5 billion on H100. Gross margin recovered to 72.7% on mix. Operating income rose 7.8x. Free cash flow of $27.0 billion exceeded the prior three years combined.
FY2025 — scale. Revenue more than doubled again to $130.5 billion; gross margin peaked at 75.0%; net income of $72.9 billion made NVIDIA the most profitable semiconductor company in history by a wide margin. Buybacks stepped up to $33.7 billion.
FY2026 — growth with two visible costs. Revenue rose 65% to $215.9 billion, but gross margin fell 390 basis points to 71.1%. Two drivers, both disclosed: (i) the business-model transition from selling Hopper HGX boards to selling full-scale Blackwell data-centre solutions, which lowers percentage margin while raising absolute dollars per deployment; (ii) the $4.5 billion H20 charge in Q1 FY2026 following the April 2025 export licensing requirement, part of $7.2 billion of total FY2026 inventory and excess-purchase-obligation provisions (versus $3.7 billion in FY2025), a net 2.6% unfavourable margin impact. Margin recovered sharply through the year, reaching 75.0% in Q4. Operating expenses rose 41%, driven by a 29% increase in compensation and benefits and a 79% increase in compute and infrastructure spend — NVIDIA is now one of the largest internal consumers of its own product. Goodwill rose from $5.2 billion to $20.8 billion and non-marketable securities from $3.4 billion to $22.3 billion, reflecting the Groq transaction and the start of the equity-investment programme.
FY2027 to date — acceleration and a new margin problem. H1 FY2027 revenue of $177.8 billion already exceeds all of FY2026 by 82%. Gross margin has held at 75.0%. But management guided Q3 FY2027 gross margin down to 74.0% and told investors margin will trough at 71–72% in Q4 FY2027 before settling at 72–73% in FY2028, attributing this explicitly to extreme memory pricing driven by the AI build-out itself. This is the first NVIDIA-specific manifestation of a genuinely reflexive risk: the company's own demand creation is inflating its input costs. Offsetting it, "other income, net" of $24.1 billion in H1 FY2027 — of which $23.7 billion is gains on equity securities — now contributes materially to earnings and introduces mark-to-market volatility that did not previously exist in the P&L. Note the divergence in Q2 FY2027: GAAP EPS of $2.46 versus non-GAAP EPS of $2.22, the gap being investment gains.
Working capital is the pressure point. In H1 FY2027 accounts receivable consumed $24.6 billion of cash and inventories $10.2 billion. Receivables at $63.1 billion are 65% of a single quarter's revenue. The 10-Q discloses that NVIDIA has extended payment terms of "90 days up to one year" to investment-grade customers "to assist customers with large data center builds." Combined with $2.8 billion of customer advances on the other side, this is a company beginning to use its balance sheet as a sales tool.
Unlock Company Financials
Unlock this company profile to view its financial data.
View Subscription PlansFinancial Detail
Segment Revenue
| Segment | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Data Center | 10,613 | 15,005 | 47,525 | 115,186 | 193,740 |
Gaming | 12,462 | 9,067 | 10,447 | 11,350 | 15,991 |
Professional Visualization | 2,111 | 1,544 | 1,553 | 1,878 | 3,190 |
Automotive | 566 | 903 | 1,091 | 1,694 | 2,350 |
OEM & Other | 1,162 | 455 | 306 | 389 | 619 |
Total revenue | 26,914 | 26,974 | 60,922 | 130,497 | 215,890 |
Segment Revenue — Data Center
| Segment | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Data Center | 39.4 | 55.6 | 78.0 | 88.3 | 89.7 |
Gaming | 46.3 | 33.6 | 17.2 | 8.7 | 7.4 |
Professional Visualization | 7.8 | 5.7 | 2.5 | 1.4 | 1.5 |
Automotive | 2.1 | 3.3 | 1.8 | 1.3 | 1.1 |
OEM & Other | 4.3 | 1.7 | 0.5 | 0.3 | 0.3 |
Segment Revenue (FY2023–FY2026)
| Segment | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|
Data Center | 41.4 | 216.7 | 142.4 | 68.2 |
Gaming | -27.2 | 15.2 | 8.6 | 40.9 |
Professional Visualization | -26.9 | 0.6 | 20.9 | 69.9 |
Automotive | 59.5 | 20.8 | 55.3 | 38.7 |
OEM & Other | -60.8 | -32.7 | 27.1 | 59.1 |
Total | 0.2 | 125.9 | 114.2 | 65.4 |
Segment Revenue (FY2025–FY2026)
| Metric | FY2025 | FY2026 |
|---|---|---|
Data Center compute | 100,000 | 156,000 |
Data Center networking | 15,186 | 37,740 |
Segment Revenue (Quarterly)
| Segment | Q2 FY2026 | Q1 FY2027 | Q2 FY2027 |
|---|---|---|---|
Hyperscale | 24,200 | 37,900 | 48,700 |
ACIE | 16,900 | 37,400 | 40,300 |
Data Center total | 41,100 | 75,200 | 89,000 |
Edge Computing | 5,643 | 6,415 | 7,200 |
Total revenue | 46,743 | 81,615 | 96,221 |
Segment Revenue (Q1 FY2027 YoY–Q2 FY2027 YoY)
| Metric | Q1 FY2027 YoY | Q2 FY2027 YoY | Q2 FY2027 QoQ |
|---|---|---|---|
Hyperscale | 92 | 101 | 13 |
ACIE | 122 | 138 | 25 |
Data Center total | 92 | 117 | 18 |
Edge Computing | 15 | 27 | 13 |
Total revenue | 85 | 106 | 18 |
Income Statement and Per-Share Data
| Metric | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Revenue (USD Mn) | 26,914 | 26,974 | 60,922 | 130,497 | 215,890 |
Cost of revenue (USD Mn) | 9,439 | 11,618 | 16,621 | 32,639 | 62,384 |
Gross profit (USD Mn) | 17,475 | 15,356 | 44,301 | 97,858 | 153,506 |
Research and development (USD Mn) | 5,268 | 7,339 | 8,675 | 12,914 | 18,400 |
Sales, general and administrative (USD Mn) | 2,166 | 2,440 | 2,654 | 3,491 | 4,731 |
Total operating expenses (USD Mn) | 7,434 | 11,132 | 11,329 | 16,405 | 23,131 |
Operating income (USD Mn) | 10,041 | 4,224 | 32,972 | 81,453 | 130,400 |
Depreciation and amortization (USD Mn) | 1,174 | 1,544 | 1,508 | 1,864 | 2,900 |
EBITDA, computed (USD Mn) | 11,215 | 5,768 | 34,480 | 83,317 | 133,300 |
Other income (expense), net (USD Mn) | -33 | 1,401 | 846 | 4,180 | 21,200 |
Pre-tax income (USD Mn) | 9,941 | 4,181 | 33,818 | 84,026 | 144,800 |
Income tax expense (USD Mn) | 189 | -187 | 4,058 | 11,146 | 24,700 |
Net income (USD Mn) | 9,752 | 4,368 | 29,760 | 72,880 | 120,100 |
Basic EPS (USD) | 0.39 | 0.18 | 1.21 | 2.97 | 4.94 |
Diluted EPS (USD) | 0.39 | 0.17 | 1.19 | 2.94 | 4.90 |
Dividends per share (USD) | 0.016 | 0.016 | 0.016 | 0.034 | 0.040 |
Diluted weighted-average shares (M) | 25,350 | 24,940 | 24,940 | 24,804 | 24,500 |
Margin Analysis (%)
| Metric | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Gross margin | 64.9 | 56.9 | 72.7 | 75.0 | 71.1 |
Operating margin | 37.3 | 15.7 | 54.1 | 62.4 | 60.4 |
EBITDA margin, computed | 41.7 | 21.4 | 56.6 | 63.8 | 61.7 |
Pre-tax margin | 36.9 | 15.5 | 55.5 | 64.4 | 67.1 |
Net margin | 36.2 | 16.2 | 48.8 | 55.8 | 55.6 |
R&D as % of revenue | 19.6 | 27.2 | 14.2 | 9.9 | 8.5 |
Effective tax rate | 1.9 | -4.5 | 12.0 | 13.3 | 17.1 |
Financial Analysis
| Metric | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Cash and cash equivalents | 1,990 | 3,389 | 7,280 | 8,589 | 10,605 |
Marketable debt securities | 19,218 | 9,907 | 18,704 | 34,621 | 39,065 |
Marketable equity securities | 0 | 0 | 0 | 0 | 12,886 |
Accounts receivable, net | 4,650 | 3,827 | 9,999 | 23,065 | 38,466 |
Inventories | 2,605 | 5,159 | 5,282 | 10,080 | 21,403 |
Total current assets | 28,829 | 23,073 | 44,345 | 80,126 | 125,605 |
Property and equipment, net | 2,778 | 3,807 | 3,914 | 6,283 | 10,383 |
Goodwill | 4,349 | 4,372 | 4,430 | 5,188 | 20,832 |
Intangible assets, net | 2,339 | 1,676 | 1,112 | 807 | 3,306 |
Non-marketable securities | 267 | 288 | 1,546 | 3,387 | 22,251 |
Total assets | 44,187 | 41,182 | 65,728 | 111,601 | 206,803 |
Accounts payable | 1,783 | 1,193 | 2,699 | 6,310 | 9,812 |
Total current liabilities | 4,335 | 6,563 | 10,631 | 18,047 | 32,163 |
Short-term debt | 0 | 1,250 | 1,250 | 0 | 999 |
Long-term debt | 10,946 | 9,703 | 8,459 | 8,463 | 7,469 |
Total debt | 10,946 | 10,953 | 9,709 | 8,463 | 8,468 |
Total liabilities | 17,575 | 19,081 | 22,750 | 32,274 | 49,510 |
Total shareholders' equity | 26,612 | 22,101 | 42,978 | 79,327 | 157,293 |
Net cash (negative = net debt) | 10,262 | 2,343 | 16,275 | 34,747 | 54,088 |
Working capital | 24,494 | 16,510 | 33,714 | 62,079 | 93,442 |
Margin Analysis (%) — Net cash from operating activities
| Metric | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Net cash from operating activities | 9,108 | 5,641 | 28,090 | 64,089 | 102,700 |
Capital expenditures | 976 | 1,833 | 1,069 | 3,236 | 6,100 |
Free cash flow | 8,132 | 3,808 | 27,021 | 60,853 | 96,600 |
Dividends paid | 399 | 398 | 395 | 834 | 977 |
Share repurchases | 0 | 10,039 | 9,533 | 33,706 | 40,100 |
Total capital returned | 399 | 10,437 | 9,928 | 34,540 | 41,100 |
FCF conversion of net income (%) | 83.4 | 87.2 | 90.8 | 83.5 | 80.4 |
Returns & Leverage Ratios (%) and Financial Ratios (x)
| Ratio | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Return on equity, ending (%) | 36.6 | 19.8 | 69.2 | 91.9 | 76.4 |
Return on assets, ending (%) | 22.1 | 10.6 | 45.3 | 65.3 | 58.1 |
Return on invested capital, computed (%) | 26.2 | 11.5 | 55.1 | 80.4 | 66.1 |
Current ratio (x) | 6.65 | 3.52 | 4.17 | 4.44 | 3.91 |
Debt to equity (%) | 41.1 | 49.6 | 22.6 | 10.7 | 5.4 |
Net debt to EBITDA (x) | -0.92 | -0.41 | -0.47 | -0.42 | -0.41 |
Interest coverage, EBIT/interest (x) | 41.5 | 16.8 | 130.8 | 407.3 | 466.0 |
Asset turnover, on average assets (x) | 0.67 | 0.63 | 1.14 | 1.47 | 1.36 |
Days sales outstanding (days) | 63 | 52 | 60 | 65 | 65 |
Days inventory outstanding (days) | 101 | 162 | 116 | 113 | 125 |
Days payables outstanding (days) | 69 | 37 | 59 | 71 | 57 |
Cash conversion cycle (days) | 95 | 177 | 117 | 107 | 133 |
Region Overview
| Region | FY2026 (USD Mn) | % of total |
|---|---|---|
United States | 149,620 | 69.3 |
Taiwan | 42,350 | 19.6 |
China (including Hong Kong) | 19,680 | 9.1 |
All other | 4,300 | 2.0 |
Region Overview (FY2025 (USD Mn))
| Region | FY2025 (USD Mn) | % of total |
|---|---|---|
United States | 61,260 | 46.9 |
Singapore | 23,680 | 18.1 |
Taiwan | 20,570 | 15.8 |
China (including Hong Kong) | 17,110 | 13.1 |
Other | 7,880 | 6.0 |
Region Overview — United States
| Region | Direction | Reason |
|---|---|---|
United States | Fastest growing (+144% FY2026 on a like-for-like basis) | Hyperscaler and AI-lab capex concentration; OpenAI, Microsoft, Amazon, Google, Meta, Oracle, CoreWeave |
Taiwan | Growing (+106%) | ODM invoicing for US/European end customers, not underlying Taiwanese demand |
China | Structurally impaired | Export controls; NVIDIA's official outlook for Q1, Q2 and Q3 FY2027 assumes zero Data Center compute revenue from China. Q2 FY2027 Hopper shipments to China were under 1% of Data Center revenue |
Europe / Middle East | Growing off a small base | 35 new NVIDIA AI/HPC supercomputers in development across Europe (announced Q2 FY2027); sovereign AI programmes; HUMAIN in Saudi Arabia |
Korea and Japan | New growth vectors | National AI infrastructure partnerships with the Japanese government and industrial leaders; SK Telecom, NAVER, Brookfield and SK hynix in Korea |
Valuation Snapshot
| Metric | Value |
|---|---|
Share price (5 Sep 2026) | $229.49 |
Market capitalisation | ~$5.55 trillion (world's largest) |
Shares outstanding | 24.10 billion |
52-week high | $236.54 (14 May 2026; all-time closing high $235.47) |
52-week low | $164.07 (5 Sep 2025) |
1-year change | +34.2% |
3-year change | approximately +5x |
5-year change | in excess of +20x |
Beta | 2.22 |
10-day average volume | 155.7 million shares |
Stock split | 10-for-1, effective 10 June 2024 (all per-share figures herein are split-adjusted) |
Valuation Multiples
| Multiple | NVIDIA | Comment |
|---|---|---|
P/E (TTM) | 29.1x | On TTM EPS of $7.91 |
Forward P/E (NTM) | 19.1x | Reflects the FY2028 guide |
EV/Sales (TTM) | ~18.2x | On TTM revenue of $303.0 billion |
EV/EBITDA (TTM) | ~27.3x | On TTM EBITDA of $201.3 billion |
P/B | ~24.2x | On equity of $229.0 billion |
Dividend yield | 0.43% | $1.00 annualised |
ROE (TTM) | 117.2% | — |
Gross margin (TTM) | 74.7% | — |
Net margin (TTM) | 63.7% | — |
Debt to equity (MRQ) | 14.6% | — |
Per-Share Data and Dividends & Capital Returns
| Metric | FY2022 | FY2023 | FY2024 | FY2025 | FY2026 |
|---|---|---|---|---|---|
Dividends per share (USD) | 0.016 | 0.016 | 0.016 | 0.034 | 0.040 |
Dividends paid (USD Mn) | 399 | 398 | 395 | 834 | 977 |
Share repurchases (USD Mn) | 0 | 10,039 | 9,533 | 33,706 | 40,100 |
Total returned (USD Mn) | 399 | 10,437 | 9,928 | 34,540 | 41,100 |
Credit Ratings
| Agency | Rating | Outlook | Date | Basis |
|---|---|---|---|---|
Moody's | Aa1 | Stable | Affirmed August 2026 | $106 billion of cash and marketable securities pro forma the June raise; adjusted total debt/EBITDA of 0.2x; expects revenue +77% in FY2027 and +54% in FY2028, and $424 billion of cumulative FCF over the two years |
S&P Global Ratings | AA | Stable | Affirmed August 2026 | $106 billion of cash and securities against $33.5 billion of funded debt; forecasts over $800 billion of FOCF in FY2027–FY2029 and EBITDA of $271 billion, $371 billion and $491 billion; downgrade trigger is adjusted leverage above 1.5x |
Fitch | Not publicly rated | — | — | — |
Notes Details
| Notes | Coupon (%) | Maturity | Amount |
|---|---|---|---|
2026 Notes | 3.20 | 2026 | 1,000 |
2028 Notes (new) | 4.250 | 15 Jun 2028 | 3,500 |
2028 Notes (legacy) | 1.55 | 2028 | 1,250 |
2029 Notes | 4.350 | 15 Jun 2029 | 3,500 |
2030 Notes | 2.85 | 2030 | 1,500 |
2031 Notes (new) | 4.500 | 15 Jun 2031 | 4,000 |
2031 Notes (legacy) | 2.00 | 2031 | 1,250 |
2033 Notes | 4.750 | 15 Jun 2033 | 3,500 |
2036 Notes | 4.950 | 15 Jun 2036 | 4,000 |
2040 Notes | 3.50 | 2040 | 1,000 |
2046 Notes | 5.550 | 15 Jun 2046 | 3,000 |
2050 Notes | 3.50 | 2050 | 2,000 |
2056 Notes | 5.625 | 15 Jun 2056 | 3,500 |
2060 Notes | 3.70 | 2060 | 500 |
Unamortised discount and issuance costs | — | — | -134 |
Net carrying amount | — | — | 33366 |
Commitment Type Details
| Commitment type | Total (USD Bn) | Due within 12 months | Due after 4 years |
|---|---|---|---|
Supply and capacity | 279 | 179 | 1 |
Cloud service agreements | 29 | 11 | 1 |
Data centre leases not commenced | 25 | 1 | 20 |
Equity investments | 25 | 21 | 0 |
Capital expenditures | 8 | 8 | 0 |
Total | 366 | 220 | 22 |
Analyst Conclusions
Management guidance
| Period | Metric | Guidance |
|---|---|---|
| Q3 FY2027 | Revenue | $108.0 billion ±2% (implies +89% YoY) |
| Q3 FY2027 | Gross margin | 74.0% ±50 bp, GAAP and non-GAAP |
| Q3 FY2027 | Operating expenses | $9.2 billion GAAP / $9.0 billion non-GAAP |
| Q4 FY2027 | Gross margin | Trough at 71%–72% |
| FY2027 | Tax rate | 16.0%–18.0% |
| FY2027 | China Data Center compute | Zero assumed |
| FY2028 | Revenue growth | Approximately 70%, implying $673–700 billion |
| FY2028 | Gross margin | 72%–73% |
| FY2028 | Supply | Expected to remain a bottleneck through at least year end |
Consensus for FY2027 is approximately $396 billion of revenue. Zacks consensus implies FY2027 EPS growth of approximately 69.2% and FY2028 of approximately 34.3%.
Bull case
-
The FY2028 guide is a floor, not a forecast. Huang stated plainly that "our demand is much greater than 70%; our supply allows us to confidently deliver 70%." A company that has beaten its own guidance fourteen consecutive quarters has just issued a supply-constrained number. Every incremental unit of HBM or CoWoS capacity that comes online converts directly into revenue above the guide. Kress's projection of $1.3 trillion of FY2028 hyperscaler capex against a $2 trillion cloud backlog frames the demand envelope.
-
The demand base has genuinely broadened. ACIE grew 138% year on year in Q2 FY2027 to $40.3 billion, faster than Hyperscale at 101%, and now represents 45% of Data Center revenue. AI clouds, sovereign programmes, industrial and enterprise deployments are structurally different buyers with different funding sources and different cyclicality. The Japan national AI infrastructure programme, the Korean gigawatt build with SK Telecom and NAVER, and 35 European supercomputers are politically funded and price-insensitive. The concentration risk that has defined the bear case for three years is measurably, if slowly, improving.
-
NVIDIA has bought the answer to its own biggest threat. ASIC substitution in inference was the most credible structural bear argument. NVIDIA paid $20 billion for Groq's LPU architecture and compiler, shipped Groq 3 LPX in full production within eight months, and integrated it into Vera Rubin racks through Dynamo for a claimed 35x throughput per megawatt. It then paid $12.9 billion for the distribution layer of the open-model ecosystem. Neither move was defensive positioning; both are shipping product. At 19.1x forward earnings, the market is not paying for either.
Bear case
-
The financing structure has turned a hardware company into a credit story. NVIDIA has invested nearly $50 billion in frontier labs, guaranteed up to $105 billion of lease and power obligations for an OpenAI affiliate, carries $108.5 billion of maximum gross guarantee exposure, and is mobilising over $500 billion of third-party capital toward entities whose principal use of that capital is buying NVIDIA product. The Q2 FY2027 10-Q added indebtedness as a standalone risk factor for the first time. Credit markets have already reacted: five-year CDS doubled from approximately 40 bp to 82 bp on guarantee reports alone. If AI monetisation disappoints, NVIDIA does not merely lose revenue — it absorbs losses on equity positions, honours residual-value guarantees, and writes down $31.6 billion of inventory built against $279 billion of non-cancellable supply commitments. The FY2023 analogue was a $1.35 billion Arm charge and $2.2 billion of provisions; the FY2029 analogue would be an order of magnitude larger.
-
Margin compression is already guided, and the cause is structural. Gross margin will fall from 75.0% to 71–72% by Q4 FY2027 on memory pricing. The memory shortage is not cyclical noise; it is the direct consequence of an industry-wide AI build-out that NVIDIA created and is contractually committed to accelerating (a reported $267 billion of memory spend through 2029). Applying 72.5% to $690 billion of FY2028 revenue instead of 75% costs roughly $17 billion of gross profit. Consensus EPS growth of 34.3% for FY2028 against 70% revenue growth already reflects this — the market has priced the margin problem, which means the upside from the revenue guide is smaller than it looks.
-
The customer base is the competitor base, and it is executing. Every one of NVIDIA's five largest customers is shipping alternative silicon at frontier scale: Anthropic on up to one million Google TPUs, Microsoft on Maia for GPT-5.2-class workloads, AWS on Trainium, Meta on MTIA and up to 6 GW of AMD Instinct, OpenAI on 6 GW of AMD. Broadcom's $73 billion AI backlog and its stated path to over $100 billion of 2027 AI revenue is, in substance, a forward order book for chips NVIDIA will not sell. ASICs are on track for 27.8% of accelerator server shipments in 2026. NVIDIA's revenue share has already fallen from approximately 87% in 2024 to 75–81% in 2026. Share loss has been invisible because the market tripled; it will not stay invisible if the market stops tripling.
Catalysts and monitorables — next 12 months
| Catalyst | Timing | What to watch |
|---|---|---|
| Q3 FY2027 results | ~20 November 2026 | Whether $108.0 billion is beaten (the fifteenth consecutive beat); gross margin against the 74.0% guide; ACIE versus Hyperscale mix |
| Q4 FY2027 results and FY2028 formal guidance | ~February 2027 | Whether the 71–72% margin trough materialises as guided, and whether the preliminary 70% growth figure is confirmed or raised |
| Hugging Face close | H2 CY2026 | Regulatory review; developer and open-source community reaction; whether Huang's open-platform commitments constrain monetisation |
| GTC 2027 | ~March 2027 | Rubin Ultra and Kyber schedule — the July 2026 slip reports are the single most important unresolved execution question |
| Memory contract renewals | Ongoing | HBM4/HBM4E pricing; whether the SK hynix partnership and custom-HBM strategy blunt the cost curve |
| SAMR final penalty decision | Any time | Quantum of any fine; whether remedies extend beyond a penalty |
| AI Overwatch Act | Congressional calendar | Passage would create a 30-day veto over export licences and a two-year Blackwell ban |
| China policy | Ongoing | Whether Beijing relaxes the customs block; whether NVIDIA restarts H200 production (management has said restart takes roughly three months) |
| Guarantee book | Quarterly 10-Q | Movement in the $108.5 billion maximum gross exposure; CDS levels; any new residual-value guarantees |
| Third-party financing platforms | Ongoing | Whether the $500 billion Apollo/BlackRock/Blackstone/Brookfield/Goldman/KKR platforms reach definitive agreements |
| Vera CPU adoption | Ongoing | Standalone attach beyond AWS and SpaceXAI — the cleanest read on whether NVIDIA can expand TAM without expanding risk |
| Hyperscaler capex guidance | Q4 CY2026 earnings | Whether the $1.3 trillion FY2028 figure Kress cited is corroborated by the hyperscalers themselves |
| OpenAI and Anthropic IPOs | 2026–2027 | Would mark NVIDIA's private positions to market and crystallise a large portion of the $47.9 billion privately held book |
Analyst verdict
NVIDIA at $5.55 trillion is not, on the numbers, an expensive stock. Nineteen times forward earnings for a business guiding to 70% revenue growth, generating 75% gross margins and $96.6 billion of annual free cash flow, is a multiple the market normally assigns to companies growing at a fifth of that rate. The market's scepticism is not about demand — Q2 FY2027 revenue doubled year on year and management issued a supply-constrained forecast because it could not honestly issue a demand-based one. The scepticism is about what NVIDIA has had to become to sustain that growth.
Over eighteen months NVIDIA has transformed from a fabless designer with a fortress balance sheet into something closer to a vertically integrated infrastructure conglomerate with an investment portfolio, a guarantee book and $366 billion of forward commitments. It now underwrites the leases, backstops the power, takes the equity, arranges the third-party capital and books the revenue. Kress's pre-emptive defence — "some will call this circular financing; we see it differently" — is the tell that management knows this is the debate.
Both things can be true. The AI build-out is real, agentic workloads are generating revenue for customers, and no competitor has assembled a comparable stack. Equally, NVIDIA has converted a business with essentially no downside beta into one with substantial second-order exposure to its customers' economics, and it has done so while guiding its own gross margin down 400 basis points on input costs it helped inflate.
The position is best understood as a high-quality operating business wrapped around an increasingly leveraged bet on the AI capital cycle. The operating business justifies the multiple. The wrapper is what will determine whether the multiple holds. Watch the guarantee disclosures and the receivables line more closely than the revenue beat.
Prepared from publicly available information as of 6 September 2026. Figures marked as derived, approximate or unconfirmed should be verified against primary filings before use in any model or transaction. Where sources conflict — notably FY2026 Data Center revenue ($193.7 billion per the Q4 FY2026 press release versus $197.3 billion in some secondary coverage), the Groq consideration ($20.0 billion versus $20.6 billion), the Hugging Face price ($12.93 billion confirmed versus $14 billion reported by Bloomberg prior to confirmation), and geographic revenue under the billing-location versus headquarters-location methodologies — both figures and the basis of the discrepancy are noted in the relevant section. Not investment advice.
Unlock Company Financials
Unlock this company profile to view its financial data.
View Subscription PlansExecutive Leadership
Executive Leadership Team
| Name | Title | Tenure | Background |
|---|---|---|---|
Jensen (Jen-Hsun) Huang | Founder, President, CEO and Director | Since founding, April 1993 | Previously microprocessor designer and director at LSI Logic and AMD. BSEE Oregon State University; MSEE Stanford. Age 63 |
Colette M. Kress | EVP and Chief Financial Officer | Since September 2013 | Previously CFO of Cisco's Business Technology and Operations Finance organisation and CFO of Microsoft's Server and Tools business; 13 years at Texas Instruments. BS Business Administration, University of Arizona; MBA Southern Methodist University |
Ajay K. (Jay) Puri | EVP, Worldwide Field Operations | Since December 2005 | Previously SVP Worldwide Sales and Support at Sun Microsystems; marketing roles at Hewlett-Packard. BSEE Cornell; MBA Harvard |
Debora Shoquist | EVP, Operations | Since 2007 (joined 2007 as SVP Operations) | Previously EVP Operations at JDS Uniphase; SVP Operations at Coherent; roles at Quantum and HP. BS Electrical Engineering, Kansas State; BS Biology, Santa Clara University |
Timothy S. Teter | EVP, General Counsel and Secretary | Since 2017 | Previously partner at Cooley LLP; mechanical engineer at Lockheed. BS Mechanical Engineering, UC Davis; JD Stanford |
Donald Robertson | VP and Chief Accounting Officer | Principal Accounting Officer | Signed the FY2026 Form 10-K |
Chris A. Malachowsky | Founder and NVIDIA Fellow | Since founding | Co-founder; not an executive officer |
Jonathan Ross | Joined from Groq (December 2025) | — | Founder/CEO of Groq; originator of Google's TPU programme. Leads LPU/inference silicon integration |
Sunny Madra | Joined from Groq (December 2025) | — | President of Groq |
Directors' Remuneration
| Executive | Salary (USD) | Stock awards (USD) | All other (USD) | Total (USD) |
|---|---|---|---|---|
Jen-Hsun Huang | 1,497,627 | 24,800,511 | 4,045,692 | 36,343,830 |
Ajay K. Puri | 949,000 | 13,800,000 | 31,000 | 14,780,000 |
Colette M. Kress | 898,580 | 12,830,000 | 16,400 | 14,340,000 |
Debora Shoquist | 848,660 | 12,910,000 | 33,570 | 14,290,000 |
Timothy S. Teter | 848,660 | 12,910,000 | 15,240 | 14,280,000 |
Board of Directors
| Director | Principal occupation / background |
|---|---|
Jensen Huang | Founder, President and CEO of NVIDIA; Chair of the Board |
Tench Coxe | Former Managing Director, Sutter Hill Ventures; director since 1993 — the longest-tenured |
John O. Dabiri | Centennial Professor of Aeronautics and Mechanical Engineering, Caltech |
Persis Drell | Professor and former Provost, Stanford University |
Dawn Hudson | Former CMO, National Football League; former CEO, Pepsi-Cola North America |
Harvey C. Jones | Managing Partner, Square Wave Ventures; co-founder and former CEO of Synopsys |
Melissa B. Lora | Former President, Taco Bell International |
Stephen C. Neal | Chairman Emeritus and senior counsel, Cooley LLP; Lead Independent Director |
Aarti Shah | Former SVP and Chief Information and Digital Officer, Eli Lilly and Company |
Mark A. Stevens | Managing Partner, S-Cubed Capital; former partner, Sequoia Capital |
Major Shareholders
| Holder | Shares (approx.) | Stake (%) |
|---|---|---|
The Vanguard Group | 2266625855 (13G, 31 Dec 2025) | 9.32 |
BlackRock, Inc. | 1,805,935,550 | 7.38 |
State Street Corporation | not separately confirmed | ~3.97 |
Fidelity Management & Research (FMR LLC) | not separately confirmed | ~4.5 |
Geode Capital Management | not separately confirmed | ~2.34 |
Jensen Huang (individual) | 922,922,938 | 3.77 |
T. Rowe Price, Capital Group, Norges Bank, JPMorgan, Morgan Stanley | not separately confirmed | remainder of top 10 |
All directors and executive officers as a group | 1,020,448,530 | 4.17 |
Competitive Landscape
Key Competitors
| Competitor | Where it competes | Scale (latest disclosed) | Positioning |
|---|---|---|---|
AMD | Merchant AI accelerators (Instinct MI350X/MI355X/MI450), server CPUs (EPYC), gaming GPUs (Radeon) | Data Center segment $5.8 billion in Q1 CY2026 (+57%); FY2025 total revenue $34.6 billion; Instinct estimated $7–8 billion in 2025 | The only credible merchant GPU alternative. Structurally constrained by roughly 11% of CoWoS allocation versus NVIDIA's approximately 60%. Wins: Meta up to 6 GW of Instinct (first 1 GW on custom MI450), OpenAI 6 GW. Server CPU share now 33.2% of units and 46.2% of revenue |
Broadcom | Custom AI accelerators (XPUs) for hyperscalers; AI networking (Tomahawk, Jericho) | AI semiconductor revenue $8.4 billion in Q1 FY2026 (+106%), $10.8 billion in Q2 FY2026 (+143%), Q3 guided to $16.0 billion; $73 billion AI backlog; targeting over $100 billion of annual AI revenue in 2027 | The most dangerous competitor because it does not compete on NVIDIA's terms. It arms hyperscalers to build the accelerators NVIDIA would otherwise sell them |
Google (Alphabet) | TPU v7 Ironwood and successors; internal and now external | — | The only silicon proven at frontier scale outside NVIDIA. Anthropic committed to up to one million TPUs |
Amazon (AWS) | Trainium/Inferentia | — | Simultaneously NVIDIA's largest new customer (2 million GPUs, August 2026) and a determined internal alternative |
Microsoft | Maia | — | Running GPT-5.2-class workloads on Maia |
Meta | MTIA | — | Also a very large AMD customer |
Intel | Gaudi 3, Crescent Island GPU, Xeon server CPUs, foundry and EMIB packaging | Data Center segment approximately $5.1 billion per quarter, predominantly CPU; DC CPU approximately $12 billion in 2025 and declining | AI accelerator share near 1%. Crescent Island not sampling until H2 2026. Simultaneously an NVIDIA supplier (EMIB) and investee ($5 billion) |
Huawei | Ascend 910B/910C in China | — | The default beneficiary of NVIDIA's China exclusion. Positioned to take China's AI chip crown in 2026 |
Cerebras | Wafer-scale training and inference | Private | Operational; carries the same acquisition-risk overhang Groq carried |
SambaNova | Reconfigurable dataflow | Private | As above |
Qualcomm, Marvell, Tenstorrent, Etched | Edge AI, custom silicon, ASICs | Various | Peripheral to the current cycle |
Arm | CPU IP licensed to NVIDIA's competitors and to NVIDIA itself | — | Structural dependency and competitor simultaneously |
“—” = not publicly disclosed.
Peer Comparison FY2026
| Metric | NVIDIA (FY2026) | AMD (CY2025) | Broadcom (FY2025) | Intel (CY2025) |
|---|---|---|---|---|
Revenue (USD Bn) | 215.9 | 34.6 | 62.0 | 53.0 |
Revenue growth (%) | 65 | 34 | 22 | -1 |
Gross margin (%) | 71.1 | 50 | 65 | 33 |
Operating margin (%) | 60.4 | 12 | 35 | -8 |
R&D as % of revenue (%) | 8.5 | 22 | 15 | 32 |
AI-specific revenue (USD Bn) | 193.7 | 8.0 | 40.0 | 1.0 |
Recent Developments
--



