Nvidia Corporation
Designs the GPUs that power the vast majority of AI training and inference workloads globally; CUDA is a 15-year software moat.
Business Segments
Revenue segmentation and growth rates. Operating margins shown where disclosed. Source: latest annual filing (10-K) and subsequent quarterly reports.
Right to Win
Moat analysis: competitive advantages, switching costs, network effects.
- CUDA is a 15-year software moat that no competitor has replicated. It is the de facto API for GPU computing — every AI framework (PyTorch, TensorFlow, JAX) compiles to CUDA. Porting models to a different hardware stack requires rewriting the entire inference pipeline.
- Nvidia's cadence (annual new architecture: Hopper → Blackwell → Rubin) means competitors are always targeting last year's product. By the time AMD or Intel ship a competitive part, Nvidia has moved the performance-per-dollar benchmark.
- The DGX and HGX system-level reference designs mean Nvidia does not just sell chips — it defines the data centre architecture. Hyperscalers build their AI clusters around Nvidia's interconnect and networking stack (NVLink, InfiniBand, Spectrum-X), creating infrastructure lock-in.
- Supply chain moat: Nvidia pre-purchases TSMC CoWoS advanced packaging capacity years in advance. Competitors cannot match volume because the packaging capacity does not exist.
- Software beyond CUDA — AI Enterprise suite, NIM inference microservices, Omniverse — layers recurring software revenue on top of the hardware sale, moving Nvidia toward a razor-and-blade model.
Competitive Landscape
MI300X / MI400 GPU line gaining datacenter traction; CPU+GPU bundling.
Gaudi AI accelerators and foundry strategy for AI chip manufacturing.
Custom ASIC for internal workloads — the most mature hyperscaler-owned alternative.
Custom ASIC design wins with hyperscalers (reportedly Google, Meta).
Risks & Regulatory
Condensed from Item 1A of the most recent 10-K. Each category represents a material risk factor disclosed to the SEC and discussed in subsequent earnings calls.
Customer Concentration
Cloud hyperscalers (Microsoft, Amazon, Google, Oracle, Meta) are estimated to account for 45–50% of Data Center revenue. Each is simultaneously Nvidia's largest customer and its most capable potential competitor (per FY2025 10-K, customer concentration disclosure).
Hyperscaler ASIC Substitution
Google's TPU v6, Amazon's Trainium3, and Microsoft's rumoured Athena chip all target inference workloads. If inference overtakes training as the dominant AI compute workload, custom ASICs with lower unit economics could displace general-purpose GPUs for a material share of the market.
Pricing Power Normalisation
Current 78% Data Center segment margins reflect absolute scarcity of AI compute. As supply catches up and competition enters, margins will compress — the question is how far and how fast. The gaming segment at 51% margin may be a floor.
Export Controls
US government export restrictions on advanced semiconductors to China have already forced Nvidia to create down-binned chips (H20) for the China market. An escalation that closes the China market entirely or expands restrictions to other regions would reduce TAM by 17–25% (per FY2025 10-K, Item 1A and geographic revenue disclosure).
AI Capex Cycle Risk
Nvidia's current revenue trajectory is funded by hyperscaler capex that is, in turn, justified by expected AI revenue that has not yet materialised at scale. If enterprise AI adoption disappoints, the capex cycle reverses and Nvidia's order book collapses — a replay of the 2018 crypto mining correction but at 5x the scale.
Upside Cases
Analyst upside scenarios — not base case, not guidance. These represent the investment thesis that bulls cite, drawn from earnings calls, sell-side research, and disclosed strategy.
- Sovereign AI — national governments (Japan, Saudi Arabia, UAE, Singapore, UK) are building domestic AI infrastructure. These are new buyers that do not compete with Nvidia and are price-insensitive because the compute is a strategic asset, not a cost centre.
- Enterprise AI — if every Fortune 500 company runs on-premise DGX clusters for proprietary data, Nvidia's TAM expands beyond hyperscalers to a market 10x larger. The NIM inference microservice strategy targets precisely this: making it trivial for any enterprise to run an LLM on Nvidia hardware.
- Automotive and robotics are speculative but genuine optionality. Nvidia Drive has design wins with Mercedes, JLR, and BYD. Omniverse for digital twins is the early-stage platform play for the industrial metaverse.
Financials Summary
Three-year financial summary. Figures derived from annual 10-K filings. Revenue, Net Income, FCF, Assets, Debt, and Cash in $M.
Financial Trends
4-year trend across key metrics. Each sparkline shows the trajectory from FY2022 (left) to FY2025 (right).
Recent SEC Filings
Recent Coverage
Profile last updated August 20, 2026. Data derived from SEC EDGAR filings and public disclosures. This is a research summary, not investment advice. All figures are from the most recent annual report unless otherwise noted.
← All Companies