How NVIDIA makes money, who it sells to, and where it is betting, as of mid-2026.
NVIDIA designs the full stack of accelerated computing (GPUs, networking, systems, and software), has it manufactured by TSMC, and sells it, increasingly as complete rack-scale systems, to the hyperscalers and AI labs building out AI infrastructure. The chips are the razor; CUDA is the razor's lock-in. It is a fabless, high-margin hardware platform whose defensibility is software.
Nearly 90% of revenue is Data Center. See revenue streams and the monetization model.
A concentrated base: in Q3 FY2026, four direct customers each topped 10% of revenue, together ~61%. The end demand is hyperscalers (Microsoft, Amazon, Google, Meta, Oracle), AI labs, sovereign-AI buyers, plus gamers and automakers. See customer segments.
The moats are CUDA's two-decade software lock-in, full-stack integration, and an annual product cadence rivals cannot match. The cost structure is fabless: TSMC fabrication, CoWoS packaging, and HBM memory are the binding constraints, and gross margin runs in the low-to-mid 70s. The strategic bets are rack-scale selling, networking, software as recurring revenue, sovereign AI, and robotics, while China export controls are the standing risk to the model.
[1] NVIDIA announces financial results for Q4 and fiscal 2026 (NVIDIA, Feb 2026) [2] NVIDIA Q1 FY2027 earnings (CNBC, May 20 2026)
Direct to hyperscalers, OEM/ODM partners, the CUDA developer flywheel, and the TSMC supply side.
The fabless TSMC dependency, CoWoS and HBM constraints, R&D, margins, and capital returns.
Hyperscalers, enterprise, sovereign AI, gamers, automotive, and the customer-concentration risk.
The full-stack accelerated-computing platform and its dominant AI-accelerator share.
CUDA lock-in, full-stack integration, annual cadence, networking, and the competitive threats.
The market platforms NVIDIA reports and how Data Center dominates the mix.
Hardware unit sales, the shift to rack-scale systems, networking attach, software licensing, and margins.
Market cap, FY2026 financials, Jensen Huang, and China export-control exposure.
The annual roadmap cadence, rack-scale selling, networking, software, sovereign AI, robotics, and the OpenAI/Intel stakes.
