A quirky group of startups that host and lease the chips that power AI tools have gone on a fundraising spree over the past year.
By Ian MartinForbes team
W founding Vultr hen David Aninowsky started building his own data centers in 2014, customers only cared about cloud computing and the processors that supported it. Graphics processing units (GPUs) belonged in gamer forums, not corporate boardrooms. Today, these chips are essential to the development of artificial intelligence models, and their leasing has propelled Vultr to a valuation of $3.5 billion.
The Florida-based startup raised $333 million from chip giant AMD and hedge fund LuminArx Capital earlier this year. month. Vultr has built more than 30 data centers around the world over the past decade, offering the advanced tier of hosting typically sold by AWS, Google Cloud or Microsoft Azure, and GPU time rental has become its primary driver of growth. “AI is the fastest growing segment of the infrastructure market,” said JJ Kardwell, CEO of Vultr. Forbes.
It’s not just Vultr. Investors have poured some $20 billion over the past year into 25 companies that lease access to GPUs, according to a study. Forbes analysis of documents filed by companies and Pitchbook data. This includes $8 billion in equity and more than $12 billion in debt in the form of loans from Wall Street giants like BlackRock, Carlyle and Pimco.
This is a rapidly emerging sector, but no one yet really knows what to call it: “Cloud GPU”? “GPU factories”? Dylan Patel, an analyst at SemiAnalysis, opted for something else. “New giants will emerge from neoclouds” he said Forbes.
The biggest winner so far has been Coreweave. The New Jersey-based startup had built an armada of GPUs to mine cryptocurrency, but after the market crashed in 2018, it decided to lease them to startups working on AI projects. In the last year alone, it raised $1.75 billion in equity and $8.1 billion in debt, at a valuation of $23 billion.
The group of startups catching up with Coreweave makes for a strange crew. These include crypto refugees like Crusoe Energy, which began mining bitcoin on gas flared from oil rigs, and German miner Northern Data which got a $1.1 billion lifeline from the stablecoin giant Tether to reinvent itself as an AI computing powerhouse.
The field also includes older data center companies like Vultr and France’s OVH that have moved from cloud computing to AI computing. And then there are players like Nebiuswhich emerged from the rubble of Yandex, the “Google of Russia,” with a Finnish data center, $2 billion in cash and a suspended Nasdaq listing. Its shares resumed trading in October and rose 700 million dollars in a private placement from Nvidia and venture fund Accel earlier this month, to propel its new business: GPU rental. A long list of smaller players includes Runpod, which brings together smaller clusters of Nvidia chips, and Fluidstack, which finds buyers for data centers with unused chips.
Investors might be willing to overlook some of the strange stories from this new brood of unicorns because of the profits generated by leasing GPUs, largely by the hour. “At one point, the ROI for purchasing a GPU was six months. It’s been a few years now,” Patel said.
“There is a misconception that Amazon is the Walmart of e-commerce and therefore must be for cloud computing. They really are the Niemann Marcus”
The neocloud boom has not only been driven by the scarcity of AI chips, but also by the massive price drop from giants like AWS and Oracle. Coreweave sells access to AI chips like the Nvidia A100 for $2.21 an hour. AWS charges $5.12 when GPUs are rented by the hour, according to a pricing analysis from neocloud rival Paperspace. Although the Amazon-owned cloud giant offers significant discounts for long-term bookings. “There is a misconception that because Amazon is the Walmart of e-commerce, it must be the Walmart of cloud computing,” said Vultr’s Kardwell. “They really are the Niemann Marcus.”
Neoclouds are able to keep prices lower because they sell bare metal GPUs without the bundled software and services typically peddled by giants like AWS. While businesses value the concierge experience, AI startups are often simply looking for the lowest price. “AI factories are not investing in the software that AWS, Google and Microsoft are doing to make these clouds usable for any kind of computing you need,” said chip analyst Karl Freund.
These more modest, smaller competitors also have an advantage when it comes to getting their hands on the key AI product: the GPUs themselves. Nvidia backed Coreweave and Applied Digital while rival AMD backed Vultr. This pressure on chips has helped make tech giants like Microsoft and Oracle some of the neoclouds’ best customers. Microsoft warned that “capacity constraints” in its data centers were causing slow growth in its AI business in its latest winning calls. And so it turned into Basic weave.
Ironically, neoclouds’ selling point for hyperscalers, who are building airport-sized data centers, may be their ability to scale. Coreweave has doubled its data centers from 14 to 28 over the last year; it has a $10 billion deal with Microsoft over the next five years. Chase Lochmiller, CEO of Crusoe Energy, said he brought a data center powered by renewable energy online in less than a year, while a hyperscaler would take at least three years. And Oracle would have signed a $3.4 billion deal to lease a Crusoe AI data center in Abilene, Texas.
However, challenges remain. Fundraising is one of them. Building and managing data centers is another, especially for smaller teams whose bitcoin mining acumen might not translate to managing finicky GPUs and sensitive customer data. “It’s like building an industrial warehouse rather than a five-star hotel; they are very different things,” warns Lochmiller.
“Let’s be clear, I think a lot of these neo-clouds are going to go bankrupt.”
There are also indications that demand for AI chips is slowing. Some startups have slashed prices by as much as a third since September, while others have started posting “GPUs for rent” ads on the Craigslist-style listings of AI investors Nat Friedman and Daniel Gross. page.
With a new generation of energy-intensive but powerful Nvidia chips entering the market, the largest and most established neo-clouds could well find themselves winners. It’s less certain for small players. For them, the combination of debt-fueled expansion and a potential chip glut exacerbated by expiring multi-year contracts at a time when more efficient chips are becoming commonplace could pose a problem. Patel said: “Let’s be clear, I think a lot of these neoclouds are going to go bankrupt. »
MORE FROM FORBES
