Nvidia's forecast stuns Wall Street

Nvidia told investors on 24 May to expect second-quarter revenue of $11.0 billion, 53 per cent above the consensus forecast of $7.15 billion. First-quarter revenue had been $7.19 billion. The shares jumped 24 per cent in after-hours trading, adding nearly $200 billion of market value in one session and pushing the chipmaker towards a $1 trillion valuation. In a market weighed down by bank failures and gloomy macro forecasts, Nvidia on its own set off a speculative rush into generative artificial intelligence.

Large companies do not usually beat forecasts by this much. This was no modest beat-and-raise built on cautious guidance. Demand for its chips had simply exploded.

How the boom started

Six months ago the company's story was about a hangover. Sales of gaming chips had slumped as the pandemic boom in home computing ended and cryptocurrency miners stopped buying, and in 2022 Nvidia had to write down unsold inventory. Its data-centre business was growing, but few investors expected it to carry the company.

Then OpenAI released ChatGPT at the end of November, and every large technology company decided at once that it needed a comparable product. Microsoft followed with a multibillion-dollar investment in OpenAI in January. Training and running those models requires thousands of specialised processors working in parallel, and Nvidia was the only supplier with the chips, the software and the production capacity ready.

Everyone wants GPUs

The race to build large language models has created a sudden, apparently bottomless demand for Nvidia's accelerated computing chips. Microsoft, Google, Meta and Amazon are each spending heavily, with no coordination, and ordering tens of thousands of H100 and A100 GPUs to build AI infrastructure.

Nvidia's proprietary CUDA software makes its hardware very hard to replace, and Jensen Huang's company earns something close to monopoly margins on AI chips as a result. Corporate America changed its spending plans almost overnight: cut general overhead and put every spare dollar into GPU clusters.

The money is coming out of other budgets. Data-centre spending that would have gone on conventional servers is being redirected to AI accelerators, and Intel, whose processors dominate those conventional servers, reported a sharp fall in its data-centre revenue in the first quarter. Nvidia's gain is partly a transfer from the rest of the chip industry rather than new spending on top of it.

The winners extend beyond Nvidia. TSMC stands to gain from every order, as does SK Hynix, the South Korean firm that supplies much of the high-bandwidth memory placed beside each processor. Server assemblers and makers of networking gear share in the spending too. The losers are the buyers: cloud companies and start-ups paying near-monopoly prices for scarce chips, and smaller AI firms that cannot secure supply at any price.

Why the orders could pause

Semiconductors are among the most cyclical industries in the world, and Nvidia's own history shows it. Demand from cryptocurrency miners inflated its sales in 2018 and then vanished, and the pandemic surge in gaming ended in last year's inventory write-down. Customers who fear shortages order more than they need, and when supply catches up they stop ordering until they have used what they bought.

The closest parallel is Cisco in 2000. Its routers were the essential equipment of the internet build-out, and its market value briefly made it the most valuable company in the world. The internet did transform the economy, but telecoms companies had over-built, orders collapsed, and Cisco's shares lost most of their value over the following two years. Being right about the technology did not protect investors from being wrong about the timing of the spending.

There are other risks too. American export controls imposed last October already bar Nvidia from selling its most advanced chips to China, and could be tightened further. Its chips are made by Taiwan Semiconductor Manufacturing Company, which ties its supply to one company on one island.

A narrow market

Nvidia's rerating also leaves the stock market dangerously concentrated. A handful of very large technology companies account for the whole of the S&P 500's gain this year, which makes the rally look far broader than it is. An equally weighted version of the index, which gives every company the same importance, has barely moved.

Investors are pricing in growth without interruption, on the assumption that spending on AI hardware keeps compounding with no cyclical pause and no period in which customers digest what they have bought.

The case for the bulls

The strongest counterargument is that this cycle is built on firmer ground than the last ones. The buyers are the richest companies in the world, paying for chips out of cash flow rather than borrowing, unlike the debt-laden telecoms carriers of 2000. Nvidia's lead in software means rivals such as AMD, and the in-house chips designed by Google and Amazon, will take time to catch up, and demand currently exceeds what Nvidia can supply.

All of that supports this year's revenue. It does not settle what customers will spend once their first wave of AI clusters is built. The signals to watch are the capital spending plans that Microsoft, Alphabet, Amazon and Meta report in late July, Nvidia's own results in late August, and any tightening of export rules.

Nvidia's guidance proves that the AI boom is backed by real orders for hardware. It has also turned the stock market into a large, concentrated bet on one chipmaker's order book, and the first quarter in which the cloud giants slow their purchases will hurt far more than one company's shares.