OpenAI has signed about $1tn in deals this year for computing power to run its artificial intelligence models, commitments that dwarf its revenue and raise questions about how it can fund them.
Monday’s deal with chipmaker AMD follows similar agreements with Nvidia, Oracle and CoreWeave, as OpenAI races to find the computing power it thinks it will need to run services such as ChatGPT.
The deals would give OpenAI access to more than 20 gigawatts of computing capacity, roughly equivalent to the power from 20 nuclear reactors, over the next decade.
Each 1GW of AI computing capacity costs about $50bn to deploy in today’s prices, according to estimates by OpenAI executives, making the total cost about $1tn.
The deals have bound some of the world’s biggest tech groups to OpenAI’s ability to become a profitable business that can meet its increasingly steep financial obligations.
“OpenAI is in no position to make any of these commitments,” said Gil Luria, analyst at DA Davidson, who added it could lose about $10bn this year.
“Part of Silicon Valley’s ‘fake it until you make it’ ethos is to get people to have skin in the game. Now a lot of big companies have a lot of skin in the game on OpenAI,” he added.
OpenAI is burning through cash on infrastructure, chips and talent, with nowhere near the capital required to fund these grand plans.
The deals also involve circular arrangements between the world’s most valuable start-up and its partners, as well as complex financing terms that have in most cases yet to be agreed.
Some content could not load. Check your internet connection or browser settings.
OpenAI’s deals with Nvidia and AMD could cost up to $500bn and $300bn respectively, according to Financial Times calculations, although both include incentives that could also help OpenAI pay for the chips it buys.
Oracle’s deal will cost OpenAI another $300bn, while data centre group CoreWeave has disclosed computing deals with OpenAI worth more than $22bn.
OpenAI also launched an initiative with SoftBank, Oracle and others in January known as Stargate that pledged to invest up to $500bn in US infrastructure for OpenAI. It is not clear how the Nvidia and AMD deal will fit into the Stargate plans.
The ChatGPT maker has not disclosed whether it will buy chips directly or through its cloud computing partners, and is expected to lease some Nvidia chips.
OpenAI has secured significant financial incentives from its suppliers in return for its chip purchases.
Nvidia plans to invest $100bn in OpenAI over the next decade, providing cash OpenAI can use to buy Nvidia’s chips for its AI data centres.
AMD will give OpenAI warrants entitling it to buy up to 10 per cent of the company for just a cent a share, depending on their project hitting certain targets, including some linked to AMD’s share price.
AMD shares were worth nearly $204 when markets closed on Monday. If they keep rising, OpenAI could sell its stock to fund its spending on AMD’s chips.
“It’s a pretty innovative structure, which didn’t come lightly,” AMD chief executive Lisa Su said on Monday.
OpenAI chief executive Sam Altman has said becoming profitable was ‘not in my top-10 concerns’ © Kevin Dietsch/Getty Images
Deals with OpenAI have also given an immediate financial boost to the start-up’s partners. Oracle’s market value jumped $244bn after its deal was made public last month. AMD shares jumped almost 24 per cent on Monday, boosting its market value $63bn.
Their circular relationships have added to concerns about a bubble in AI, at a time when investors are worried that spending on AI data centres is propping up US economic growth.
To fund its expansion, OpenAI has raised huge amounts of equity and started to tap debt markets. It secured $4bn in bank debt last year and has raised about $47bn from venture capital deals in the past 12 months — though a significant chunk of that is contingent on a tricky negotiation with Microsoft, its biggest backer.
OpenAI — valued at $500bn this month — is also preparing to raise tens of billions of dollars of debt to fund infrastructure, said people close to the company.
The start-up’s perceived credit risk has prompted concern. Moody’s flagged how much of Oracle’s future data centre business relies on OpenAI and its unproven path to profitability.
OpenAI’s arrangement with Nvidia is likely to help investors get more comfortable making large loans.
The chip giant, which recently surpassed $4tn in market cap, has routinely used its mammoth balance sheet to invest in companies within its supply chain or among its top customers.
The beneficiaries of these deals in turn use the new liquidity to buy more Nvidia chips or borrow money.
Nvidia has invested in CoreWeave, which is also a customer and supplier. CoreWeave has also raised raise more than $12bn of debt secured against its Nvidia chips.
Some content could not load. Check your internet connection or browser settings.
Beyond Nvidia’s support, OpenAI’s partners and investors have bought in to its promises of future growth and a path to profitability.
OpenAI expects to multiply its revenue from its current level of $12bn in the coming years by rolling out new products and doubling the number of paying subscribers for its core product ChatGPT.
Recommended
Chief executive Sam Altman on Monday said becoming profitable was “not in my top-10 concerns”.
“But obviously someday we have to be very profitable, and we’re confident and patient that we will get there . . . Right now we are in a phase of investment and growth and if we can deliver all of this value,” he added.
OpenAI and its growing number of partners are betting AI usage will keep growing exponentially. If growth plateaus, or even slows, the investor enthusiasm that has boosted share prices on the back of these deals could quickly falter.
One Silicon Valley investment veteran said: “The company is in a far more capital-intensive business than Google or Microsoft ever was, and was born with no cost discipline.”
Amazon founder Jeff Bezos and Oracle founder Larry Ellison “only found religion” and drastically cut business costs, “after nearly going bankrupt”, the investor added.
Additional reporting by Cristina Criddle in San Francisco. Data visualisation by Eva Xiao and Haohsiang Ko
