Nvidia just made the strongest case yet that the AI investment cycle is not close to peaking.
Quarterly total revenue more than doubled to $96.2bn, and the company now expects revenue to grow roughly 70% next fiscal year, far ahead of Wall Street's prior expectations. Demand is broadening beyond hyperscalers to sovereign buyers, enterprises and new cloud providers, while Nvidia remains constrained by memory and other components.
That matters because the central question around AI has changed. A year ago, investors were asking whether demand was real. Now they are asking whether demand can stay strong enough to justify the hundreds of billions being spent on data centres, power, networking and chips.
Nvidia's answer is yes. The company also expects its next-generation Vera Rubin processors to contribute meaningfully, while Amazon has agreed to deploy 2 million Nvidia GPUs by 2028.
The risk is no longer lack of demand. It is whether the economics of the entire ecosystem can keep pace with the scale of investment. Margins are already under pressure from higher component costs, and China remains largely absent from Nvidia's outlook because of export restrictions.
Nvidia is telling investors that AI is not a one-cycle boom. If it is right, the more important question becomes who earns an attractive return on all the infrastructure being built around it.
Macro
Public markets
M&A
Staying Diligent
Things we are watching this week: 31 August–4 September:
Unhedged Commentary
Your AI supplier can become your competitor overnight.
Cursor built one of the world's most successful AI coding tools partly by using models supplied by OpenAI. Then its owner changed. SpaceX acquired Cursor's parent company, and OpenAI subsequently said it plans to stop supplying Cursor with its models from November. Anthropic, meanwhile, is increasing support for the platform.
The corporate dispute is unusually colourful. The strategic lesson is not. Companies are increasingly building products on top of a handful of frontier AI providers. That looks efficient until your supplier acquires a competing product, changes its commercial priorities, becomes politically constrained or simply decides that your business is no longer strategically useful.
Cloud computing already taught companies the value of avoiding excessive dependency on one infrastructure provider. AI may make that lesson more urgent. The obvious response is multi-model architecture: design products so that OpenAI, Anthropic, Google or other models can be substituted without rebuilding the entire product. That may cost more in the short term. But dependency also has a price.
The frontier model is increasingly becoming critical infrastructure. If your product cannot survive your model provider changing its mind, you do not really control your product.
In Other News
Hubble needs a century. Roman needs a month.
NASA launched the $4bn Nancy Grace Roman Space Telescope this weekend. Its mirror is roughly the same size as Hubble's. Its advantage is not that it can see dramatically further. It can see dramatically more.
Roman's wide-field camera can survey huge areas of the sky at once. One month of Roman observations of the Milky Way would take Hubble roughly a century to complete.
That is a useful business lesson. Innovation does not always mean doing something entirely new. Sometimes it means taking a job we already know how to do and changing its economics by two orders of magnitude. The internet did not invent shopping. Cloud computing did not invent computing. Streaming did not invent television. Their impact came from making familiar activities faster, cheaper or vastly easier to scale. Roman is similar.
Hubble remains extraordinarily valuable because it can spend long periods studying individual targets in exceptional detail. Roman does something different: it turns astronomy into a survey at a scale Hubble was never designed for.
The lesson: Disruption is not always about replacing the old technology. Sometimes it is about making one particular job so much faster that an entirely new set of possibilities becomes practical.
The Thinking Corner
When an industry is growing fast enough that suppliers, customers and competitors increasingly become the same companies, what evidence should investors look for to understand where the durable value will actually be captured?
Latest from the team
Market commentary and field notes from the team, on The Felix View.