What happens if OpenAI or Anthropic fail?

LONDON, March 11 (Reuters Breakingviews) - The artificial intelligence boom would make for a nasty bust. Alphabet (GOOGL.O), opens new tab, Amazon.com (AMZN.O), opens new tab, Meta Platforms (META.O), opens new tab, and Microsoft (MSFT.O), opens new tab plan to spend about $650 billion this year largely on chip-packed data centres powering chatbots. Much of that capacity will serve cash-burning AI labs such as OpenAI and Anthropic, the developers of ChatGPT and Claude. Breakneck growth and bottomless ​funding sustain them. If either falters, the consequences could be ugly.

The scale is staggering. Big Tech’s spending commitments are roughly equivalent to 2% of U.S. GDP last year. Morgan Stanley foresees $2.9 trillion of global ‌data-centre investment between 2025 and 2028, of which it expects roughly $900 billion will come from private credit and asset-backed lending. The boom now stretches far beyond Silicon Valley into lenders, utilities and infrastructure investors.

The Reuters Iran Briefing newsletter keeps you informed with the latest developments and analysis of the Iran war. Sign up here.

OpenAI’s ambitions underscore the sheer scale of the financial mobilization. CEO Sam Altman has floated infrastructure build-out deals worth $1.4 trillion, requiring 30 gigawatts of power, enough to supply 20 million U.S. households. Headline figures probably overstate the immediate outlay, given the company’s complex web of overlapping deals and multi-year partnerships. Still, this is a massive amount of capital due ​to flow into cloud providers like Oracle (ORCL.N), opens new tab, Microsoft, Amazon, and CoreWeave (CRWV.O), opens new tab, as well as chip and gear-makers including Nvidia (NVDA.O), opens new tab and Broadcom (AVGO.O), opens new tab.

Funding all of this will be an increasingly difficult challenge. HSBC estimates that OpenAI may need $207 ​billion in additional financing by 2030. True, the company just announced a $110 billion funding round. But such sums increasingly bear asterisks.

Take Amazon. It pledged $50 billion of the round, but this ⁠begins with an initial payment of about $15 billion, with further outlays tied to Altman successfully leading his company onto public markets or surpassing a smarter-than-human technological milestone known as “artificial general intelligence", whichever comes first. SoftBank, expected to add another $30 ​billion, is itself under pressure: S&P Global recently assigned, opens new tab the Japanese group a negative credit outlook as it explores a bridge loan of up to $40 billion, according to Bloomberg.

Both OpenAI and Anthropic are considering initial public offerings as soon as this year, as ​the Financial Times and Reuters reported, so the hope may be that public investors can soon substitute for private ones. True, growth is extraordinary: OpenAI topped $25 billion in annualized revenue as of end of February, The Information reported, up 25% from the end of December. Yet competition is scaling quickly, especially among cheap and ubiquitous open-source systems that pressure pricing across the industry.

Public investors would also need to have strong stomachs for red ink. HSBC analysts in late November estimated almost $280 billion of cash burn for OpenAI between now and the end of 2030. In ​a court filing, opens new tab this week, Anthropic’s CFO said the company had spent over $10 billion on training models and serving up responses to user queries, all to generate roughly $5 billion of cumulative revenue. It shows that the arithmetic is unforgiving unless ​computing costs fall sharply or customers pay far more.

To nonetheless achieve a mooted $1 trillion IPO valuation, OpenAI would need to maintain extremely rapid growth. Nvidia and Alphabet trade on average at 5.6 times expected sales in 2030, according to Visible Alpha. Applying a standard 30% ‌IPO discount ⁠to that multiple, OpenAI would need to generate at least $250 billion in annual revenue by 2030. CNBC reports, opens new tab that internal targets are even higher, at $280 billion. But to succeed, OpenAI would have to build a business the size of today’s Microsoft in just four years.

Big Tech, along with the broader AI supply chain, ultimately depends on the success of OpenAI and Anthropic. Microsoft said last month that 45% of its $625 billion in demand backlog is tied to OpenAI, while Altman has inked a $300 billion deal with Oracle. Together, that’s about two-thirds of the $800 billion HSBC estimates OpenAI could spend on chips and data centers, with the rest spread across Nvidia, AMD (AMD.O), opens new tab, Amazon and CoreWeave.

While the scale is smaller, Anthropic is also set, opens new tab to spend ​many billions more. Yet a recent battle with the Pentagon ​adds fresh uncertainty, beyond whether it can even ⁠fund its plans. After CEO Dario Amodei resisted allowing Claude to be used for mass surveillance or autonomous weapons, the Trump administration labelled the company a supply-chain risk, with defense chief Pete Hegseth claiming that no government contractor can work with it. Anthropic has sued to overturn these moves, saying that they violate its rights to free expression and due process. ​For now, cloud partners Amazon, Alphabet and Microsoft say they will continue offering Claude to customers.

Granted, in a worst-case scenario, creditors and investors tend to assume that a ​buyer would emerge for a weakened ⁠OpenAI or Anthropic. Microsoft, Amazon or another strategic bidder would benefit greatly from their technology, talent and user base.

But such a rescue would likely come with a sharp valuation reset and a rethink of how much computing power the industry truly needs. And there’s another, more general problem. Wall Street is beginning to question the sustainability of the AI spending boom itself. Credit-default swaps on Oracle have climbed to their highest levels since 2008. Shares of Microsoft and Amazon slid after both companies reiterated plans ⁠to increase capital ​spending.

Should AI labs or Big Tech’s generative AI monetization fall short of expectations, the capital flows required to build capacity could quickly dry up. ​If data-centre expansion slows, it would plausibly reduce sales of Nvidia chips, leave major investments in power infrastructure underutilized, and lenders uncertain. In that sense, the failure of a single lab would not be a simple corporate stumble. It would drag down a historically massive boom.

(This column has been refiled to restore dropped words and fix hyperlinks throughout.)

Follow Karen Kwok on LinkedIn, opens new tab ​and X, opens new tab.

For more insights like these, click here, opens new tab to try Breakingviews for free.

Editing by Jonathan Guilford; Production by Pranav Kiran

• Suggested Topics:
• Breakingviews

Breakingviews
Reuters Breakingviews is the world’s leading source of agenda-setting financial insight. As the Reuters brand for financial commentary, we dissect the big business and economic stories as they break around the world every day. A global team of about 30 correspondents in New York, London, Hong Kong and other major cities provides expert analysis in real time.

Sign up for a free trial of our full service at https://www.breakingviews.com/trial and follow us on X @Breakingviews and at www.breakingviews.com. All opinions expressed are those of the authors.

Share

• X

• Facebook

• Linkedin

• Email

• Link

Purchase Licensing Rights

Karen Kwok

Thomson Reuters

Karen is a columnist focusing on global technology and venture capital sectors, writing stories about artificial intelligence, fintech, and semiconductor companies. She also covers deals in the Middle East region and global metal mining sector. Prior to Breakingviews, she was a European gas and power reporter at S&P Global Platts in London and covered funds and equities at Morningstar UK. Karen also briefly worked at Bloomberg. Born and raised in Hong Kong, she is fluent in Mandarin and Cantonese.

• Email

• X

• Instagram

• Linkedin