Nvidia (NVDA).

Nvidia designs GPUs: chips that run thousands of calculations in parallel, exactly what training and serving AI models requires. Manufacturing is outsourced, mostly to TSMC in Taiwan.

On top of chips it sells complete server systems and data-centre networking, so customers buy compute factories rather than components.

Data centres dominate the top line. Buyers are cloud providers and AI labs, so a handful of customers account for a large share of sales.

Gaming, professional visualisation and automotive are the smaller legacy segments.

Before the AI boom, Nvidia cards were popular with Ethereum miners. That demand largely disappeared when Ethereum moved to proof of stake in 2022; bitcoin mining runs on dedicated ASICs.

Today the overlap is mostly infrastructure: the same data centres, the same energy debate and partly the same investors.

Customer concentration, export rules towards China and the question of whether cloud capital spending keeps growing at this pace. Google, Amazon and Microsoft are also building in-house AI chips.

The valuation leaves little room for disappointment: small changes in expectations move the share price sharply.

Nvidia's lead is not only in the silicon but in the software around it. CUDA, the libraries and the networking stack from the Mellanox deal make switching expensive: a model pipeline built on Nvidia does not move in a single quarter.

Competition comes from two directions. AMD's MI series is a credible alternative on paper, and the cloud providers build their own accelerators — Google TPUs, Amazon Trainium, Microsoft Maia. Those chips do not need to beat Nvidia; they only need to improve the buyer's negotiating position.

Each quarter, look at data-centre revenue separately from gaming, at gross margin (a proxy for pricing power) and at inventory plus purchase commitments. That last number leads demand: capacity is bought before it ships.

Also track the capex guidance of the four largest cloud customers. In practice that guidance is Nvidia's order book, one to two quarters ahead.

Chip designers earn from every expansion of data-centre capacity, regardless of which model wins. That is the strength of this link — and the reason valuations run ahead of actual deployment.

The sector is cyclical: inventories, lead times and export rules often matter more than the quarterly print. A one-quarter delay at a large customer shows up immediately.

The price shown for Nvidia is indicative and delayed; use your broker's data before placing an order. The explanation above is meant to help you read the news, not to time a trade.

Compare a company with its peers in the same link of the chain before drawing conclusions, and check whether a move follows results, guidance or simply the broader market.