NVDA — NVIDIA Corporation: business, risks and options character

In plain English

What the company actually sells

NVIDIA designs the chips and the racks that train and run artificial intelligence. It does not build them itself — it designs them, has them manufactured (largely by TSMC), and sells them as complete systems to the companies building AI data centres. The gaming graphics cards it is famous for are now a small side business; the overwhelming majority of its revenue comes from data centres.

Why the share price moves

One question moves this stock more than any other: are the world's largest technology companies going to keep spending tens of billions of dollars a quarter on AI hardware? When they signal more spending, NVIDIA rises. When anyone suggests the spending might pause, or that a cheaper way to get the same result exists, it falls hard. It is a bet on other companies' budgets.

If you already own it

You own the most volatile large position in the index. If it is a big share of your portfolio, the honest first question is not how to earn income on it, but whether the position size is one you can live with through a 30% drawdown — because it has had several.

The technical read

What actually drives results

Four variables explain most of what this company reports. Everything else is commentary.

Hyperscaler capital expenditure

A small number of cloud and AI platform companies account for a very large share of data-centre revenue. Their quarterly capex guidance is, functionally, NVIDIA's revenue guidance — which is why the stock often moves more on a customer's earnings call than on its own.

Product-cadence execution

The company has moved to roughly annual architecture releases with rack-scale systems rather than loose chips. Each transition is a revenue accelerant if it lands on time and a margin and inventory problem if it slips, because customers defer purchases of the outgoing generation.

Supply, not demand

For much of the AI cycle the binding constraint has been advanced packaging and high-bandwidth memory capacity, not orders. That inverts the usual analysis: the useful question has been how much they can ship, not how much can be sold.

Software lock-in (CUDA)

Two decades of CUDA libraries, frameworks and tooling mean the practical cost of switching accelerators is a re-engineering project, not a purchase order. This is the asset that supports the gross margin, more than the silicon itself.

The moat, stated as a mechanism

Not the chip — the system plus the software. Competitors can match raw silicon specifications far more easily than they can match networked rack-scale systems plus a mature software stack that engineers already know. The moat is measured in re-training time and porting effort.

A moat you cannot describe as a mechanism is not a moat — it is a track record. The test is whether you can name the specific reason a well-funded competitor has not simply taken the business.

What would break it

The parts an owner should be able to state without prompting. None of these is a prediction; each is a live risk with an identifiable mechanism.

Customer concentration. When a handful of buyers represent a large share of revenue, any one of them insourcing or slowing changes the picture immediately — and each of the largest ones is actively developing its own accelerators.
Custom silicon at the inference layer. Training is where NVIDIA's advantage is strongest; inference is a cost-per-token contest, and that is precisely where purpose-built chips from hyperscalers compete best.
The digestion risk. Capital-equipment cycles have historically been cyclical, not monotonic. A single quarter in which the largest buyers announce they have enough capacity for a while would reprice the stock violently, regardless of long-run demand.
Export controls. Policy on advanced-chip sales to China has repeatedly changed the addressable market by administrative decision rather than by competition.

The line to read first in the next earnings release

Data-centre revenue growth alongside gross margin, read together. Growth with expanding margin means pricing power intact; growth with compressing margin means the mix is shifting toward lower-margin systems or competitive pressure is arriving.

How it behaves in the options market

This is the section no ordinary stock write-up contains, and it is the one that decides whether an income overlay on these shares is sensible. Two companies with identical fundamentals can be completely different option underlyings.

Implied volatility

Routinely the highest implied volatility of any company this large — a multiple of the index's, and frequently in territory more typical of a mid-cap growth name than of the market's biggest constituent.

Earnings behaviour

Its earnings release is a scheduled market-wide volatility event. Single-session moves of high single digits to low double digits in percentage terms have been ordinary rather than exceptional, in both directions.

Chain liquidity

Among the most heavily traded options in the world: weekly expiries, dense strikes, tight spreads. Liquidity is never the problem here — the price behaviour is.

Structural traps

A 10-for-1 split in June 2024 means any historical strike, premium or chart from before that date must be divided by ten to compare. The shares are cheap enough per-contract that covered calls are accessible on small accounts, which is part of why so many retail holders write them.

Covered-call character

Covered-call characterHandle with care

This is the clearest example in the whole index of the trap the Screener exists to expose. The credit on a NVIDIA call is spectacular, and the upside it caps has repeatedly been more spectacular still. Selling a call here is selling the single best weeks of the position for a flat fee — and if the shares are called away from a low cost basis, the tax bill lands on top. Model net-vs-hold before writing anything.

This grades the option trade, not the company and not the stock. A “Poor fit” means the chain pays too little to justify capping upside — nothing more.

Dividends and buybacks

A token dividend and large buybacks. The dividend is too small to matter to an income plan, so premium is effectively the only yield available on the position — which is exactly what tempts holders into capping it.

The tax wrinkle specific to this name

Anyone holding NVIDIA from the early part of the AI cycle is sitting on an enormous unrealised long-term gain. Assignment does not just cap the upside; it forces a sale and crystallises that gain, on the exchange's schedule rather than yours. Run the assignment tax through the Calculator before choosing a strike — for a low-basis holder it is frequently the largest single line in the trade.

Now check today’s numbers

Everything above is durable. Everything that is not — price, implied volatility, the actual credit on an actual strike — lives in the tools, with NVDA already filled in.

Model NVDA after tax Scan the live NVDA chain

The Calculator is free and needs no account. The Screener reads a live options chain, so it asks you to sign in first — the link will bring you back here.