NVDA · NVIDIA CorporationSemiconductors
Business Model
NVIDIA designs the chips, systems, networking, and software stack that power modern AI factories. It is fabless, with TSMC and memory suppliers doing the manufacturing, but it captures the platform economics through CUDA, NVLink, GPU systems, inference hardware, networking, and data-center software. The company has moved from selling GPUs into selling a full AI infrastructure platform.
NVIDIA is fabless: it designs accelerated-computing silicon and complete systems, while foundries, memory vendors, and packaging partners carry the manufacturing base. It earns far more than a chip vendor because customers buy an integrated stack — GPU compute, NVLink, networking, systems, libraries, and enterprise software — that raises switching costs and expands revenue per AI cluster.
CUDA remains the software foundation, but the economic control points now extend through Blackwell and Vera Rubin systems, NVLink scale-up, Spectrum-X and InfiniBand scale-out, plus the inference and agentic-software layers. The moat is the co-design of the whole AI factory, not any one GPU generation: replacing the accelerator also means replacing software, networking, deployment tooling, and developer workflows.
Q1 FY2027 revenue was overwhelmingly AI infrastructure: Data Center represented 92.2% of company revenue, split almost evenly between Hyperscale customers and AI Cloud, Industrial and Enterprise customers. The legal reporting lens is different — Compute & Networking versus Graphics — so the investor view needs both taxonomies without adding their totals together.
The model combines premium platform pricing with outsourced manufacturing. In Q1 FY2027, Compute & Networking produced a 71.5% segment operating margin and Graphics 41.6%; FY2026 generated $96.7B of free cash flow on only $6.0B of capex. The result is manufacturer-scale revenue with software-platform cash conversion, while the capital and supply-cycle burden remains concentrated at TSMC, HBM suppliers, and packaging partners.
Investment Verdict
The Q1 FY2027 print and $91B Q2 guide turned the debate from 'AI capex peak' to 'how much supply can NVIDIA ship'; the stock is still priced like a premium GPU vendor, while the evidence now looks like a vertically integrated AI infrastructure platform.
px · close · 2026-08-18
FY2027 revenue reaches the $450-500B range, Vera Rubin ramps faster than Blackwell, Groq 3 becomes a real inference revenue category, and gross margin expands into the 76-78% range.
FY2027 revenue lands around $400-450B, hyperscaler capex holds, Vera Rubin ships on schedule, custom ASICs take some incremental share, and networking plus Groq 3 offset the compute-share loss.
Hyperscaler capex cuts sharply, AMD MI450 or custom ASICs take material training share, Vera Rubin slips beyond the Q4 2026 volume window, or China/restriction risk removes optional upside.
Entryown while Q2 FY2027 revenue guide, Vera Rubin first-ship timing, and gross-margin durability remain intact; add on price weakness that does not come with demand or supply evidence breaking.
Exitreduce if Q2 misses the $91B guide, Q3 guidance decelerates below the ramp path, Vera Rubin slips beyond Q4 2026 volume, or AMD/custom ASICs prove training-scale substitution.
Consensus FY28E EPS $12.89 sits closest to our Bull EPS $13.00 — the market already prices the Bull earnings as its base case.