Google’s Cloud Partner for Its AI Chips Is in Talks to Raise $700 Million
The Takeaway
- Fluidstack seeks over $700 million in new financing.
- AI cloud provider Fluidstack is crucial to Google’s TPU chip expansion.
- Fluidstack projects over $400 million in sales for the current year.
Fluidstack, a cloud provider aiding Google’s AI chip push, is in talks to raise more than $700 million in new financing, according to two people with knowledge of the deal.
A one-year-old investment fund started by ex-OpenAI researcher Leopold Aschenbrenner is in talks to lead the financing, one of the people said. Aschenbrenner’s fund, Situational Awareness, has stakes in AI cloud provider CoreWeave and AI developer Anthropic, which is a customer of Fluidstack.
Fluidstack is in the middle of a budding battle between Google and Nvidia over chips to power AI. Fluidstack has primarily rented out Nvidia’s graphics processing units but recently became key to Google’s effort to get Google’s homegrown AI chips, tensor processing units, in the hands of more customers. Until recently, TPUs were only available through Google Cloud.
The investment would help Fluidstack compete in an increasingly crowded field of firms renting out specialized servers to AI developers. Fluidstack’s customers also include Mistral AI.
Fluidstack’s current valuation, and the valuation it’s seeking from the new financing, couldn’t be learned. The amount of capital it’s looking for implies it wants a multibillion-dollar valuation.
Investment bank Morgan Stanley, a longtime adviser to CoreWeave, is working with Fluidstack on the fundraising, according to two people with knowledge of the discussions. Google has also considered investing in Fluidstack, said two people with knowledge of the discussions. Fluidstack’s fundraising talks are ongoing, and the details could change.
Spokespeople for Google and Morgan Stanley declined to comment. Aschenbrenner did not respond to a request for comment. Fluidstack did not comment on the fundraising.
Google has agreed to backstop Fluidstack’s leases to use three data centers currently under development. Fluidstack agreed to host Google’s TPUs in at least one New York–based data center. Anthropic said last month it planned to use up to 1 million TPUs as part of an expanded contract with Google and to invest $50 billion, alongside Fluidstack, in U.S.-based data centers.
Founded eight years ago, London-based Fluidstack has grown quickly over the past year as it has signed big contracts with customers, though it has raised scant equity funding to date.
Fluidstack has relied on borrowed money to fund its server costs. Earlier this year, it secured approval from lenders including Macquarie for more than $10 billion in credit backed by AI chips, including Nvidia’s.
Fluidstack in July projected it would generate more than $400 million in sales this year, The Information reported. It generated more than $65 million in revenue last year, according to a local regulator filing.
That would make Fluidstack a fraction the size of CoreWeave, which projected revenue will double this year to more than $5 billion. Crusoe and Lambda, two other privately held AI cloud providers, have raised a combined $2.9 billion over the past two months to expand their server rental businesses.
Cloud providers focused on AI, including Oracle, have lately faced questions about the profitability of renting out pricey AI chips and about the debt they’re raising to obtain the chips.
Aschenbrenner’s Ascent
Aschenbrenner graduated from Columbia University when he was 19 and previously worked at the Future Fund, a philanthropic fund started by former FTX CEO Sam Bankman-Fried.
At OpenAI, Aschenbrenner worked on techniques for developing AI safely. He was an ally of OpenAI co-founder Ilya Sutskever, who participated in a failed effort to force out OpenAI CEO Sam Altman in late 2023.
In 2024, OpenAI pushed out Aschenbrenner, alleging he had improperly shared information with outsiders. He later said on the “Dwarkesh Podcast” that though he had shared a document with external researchers for feedback, he had removed sensitive information from it. Aschenbrenner said OpenAI took issue with the fact that the document revealed OpenAI was planning to reach “artificial general intelligence” by 2027 or 2028 and had not set “timelines for preparedness.”
“I didn’t think that planning horizon was sensitive,” Aschenbrenner said on the podcast. “It’s the sort of thing Sam says publicly all the time.”
Aschenbrenner also said he also shared a memo with OpenAI board members after a major security incident, which led to him receiving an “official HR warning.”
“The thing I understand, and in some sense it’s reasonable, is that I ruffled some feathers and was probably annoying at times with the security stuff,” he said.
Later last year, he published a manifesto on AGI titled “Situational Awareness,” and started an investment fund under the same name. He secured investments from former GitHub CEO Nat Friedman, investor Daniel Gross, Stripe CEO Patrick Collison and Stripe President John Collison. The fund manages more than $1.5 billion and also invests in public semiconductor and infrastructure companies.