In 2023, being a startup means figuring out some way to build AI into your product, while it’s also been the year of companies raising monster seed rounds to build huge generative models.
All of this is creating big demand for the specialised AI chips needed to run the technology — and they are in short supply. Some startups are struggling to get quick access due to long waiting lists with cloud providers.
In this, some investors have spotted an opportunity.
Some VC firms are beginning to invest in the companies building the supercomputers that house and deploy these chips as a way to guarantee startups in their portfolios access to compute (the industry name for this resource).
An added bonus is that it might also entice other promising AI startups to take their investment.
Like building a hotel
One of the first companies doing this in Europe is London-based NexGen Cloud, which plans to build a $1bn AI supercomputer in Europe in the coming months, based on best-in-class chips from US company NVIDIA. These chips are in short supply, but the startup is an “elite member” of NVIDIA’s partner network, and is in the fortunate position of having put in early orders for a significant number of chips.
CEO Chris Starkey tells Sifted that VC firms have helped finance the project as a way to get supercomputer access for their startups.
“We’ve had a lot of appetite from VCs who have quite a large portfolio of companies that need sheer compute. They need it for years and years and years, and they struggle [to access it],” he says.
Starkey explains that VCs invest in owning the hardware and NexGen Cloud then manages it by running the technology and the infrastructure to use it on the cloud.
“You can think of it like: we’re effectively building a hotel, they [VCs] are buying the hotel. We help them rent out the rooms [to startups] and they can rent the rooms too [for themselves],” he says.
This means that VCs who invest in NexGen Cloud get a cut of the profit from renting out the hardware, and can get their startups faster access to compute than they’d get on the open market.
“It actually gives them [VCs] a competitive edge,” says Starkey. “We’ve seen it in their negotiations where they say, ‘Hey, we can bring you the money, but we can also bring you compute, or we can bring you the compute as investment.’”
Getting into infrastructure
NexGen Cloud says that it’s one of the first companies to run this model in Europe, and that its closest competitors are US-based cloud compute providers like Lambda and CoreWeave, which made a big splash in August when it raised a $2.3bn debt facility from investors including VC giant Coatue. Prominent angels like Nat Friedman, the founder of GitHub, and Daniel Gross, a former machine learning director at Apple, are also on CoreWeave’s cap table.
“Nat Friedman and Dan Gross, who did that deal with CoreWeave — I’m sure they get good demand as a result of that, because it feels like kind of creating a product that demonstrates affinity for their customer,” says Nathan Benaich, founder of AI investment firm Air Street Capital.
Starkey says deals like this show that typical startup investors are seeing the value of investing in infrastructure, as AI has created a huge demand for cloud-rented compute across the tech ecosystem.
“We’re working with one VC firm that has several companies that are at a point where they’re moving into quite excessive cloud payments. One of those is spending $100m a year — it’s a significant amount of money.”