US · Educational sample
NVIDIA (NVDA)
The pick-and-shovel of AI — extraordinary economics meeting extraordinary expectations.
Business
NVIDIA designs GPUs and systems that train and serve large models, monetizing both silicon and a software stack around CUDA. Data-center mix has become the profit engine; gaming and auto are options on a smaller base. Gross margins reflect scarcity pricing when demand outruns foundry supply.
Moat
CUDA is the real fortress: years of libraries, talent, and customer code paths that competitors cannot clone overnight. Systems integration and networking deepen the lock-in beyond a single chip. Network effects among developers keep the switching cost rising with every deployment.
Valuation
Owner earnings look pristine in the boom — but the price often capitalizes a multi-year AI capex supercycle as if it were a perpetual annuity. Value investing asks what residual cash remains if hyperscaler spend merely normalizes, not collapses.
Price bands
Buy zone only if a stress-case data-center revenue path still clears your required return. Wait while euphoria prices uninterrupted H100/Blackwell scarcity forever.
What to ask
Prefer owning the shovel seller when the shovel is on sale — not when every investor already owns the narrative.
Risk
Customer concentration, custom ASICs from cloud giants, and a boom-bust capex cycle can reverse margins quickly. Export controls and foundry capacity are exogenous shocks that no spreadsheet fully prices.
Educational research only. Not investment advice. Sample memos are static illustrations of the ValueDesk lens — run a live memo on any ticker in the desk.