Anthropic has launched the prospectus for what may be the biggest-ever initial public offering – and the fate of the entire US artificial intelligence complex may rest on it.
The float of the leading frontier model entity, which is expected to raise about $US100 billion, is critical to the continuing ability of a sector that has a voracious and open-ended appetite for capital.
It has to succeed and be seen as successful if the access to the trillions of dollars of new funding the industry will seek over coming years is to remain open.
If it were to fail, or to be regarded as less than very successful, the highly interconnected and interdependent sector that relies on that continuing access to equity and debt – from those at the pointy end like Anthropic and OpenAI to hyperscalers like Amazon, Google, Microsoft and Meta to neoclouds like CoreWeave and Nebius – would be at risk.
And for all but the hyperscalers, the risk would be existential.
The scale of the access to cash required was underscored by the Anthropic’s prospectus. The company — which is about to raise $US100 billion or so, already has $US112.4 billion of cash and recently increased the size of its revolving credit facility from $US3 billion to $US15 billion — warns in it that it may need to raise even more funding sooner than it had previously anticipated.
Anthropic’s IPO has to succeed and be seen as successful if the access to the trillions of dollars of new funding the industry will be seeking over the next few years is to remain open.
To maintain access to funding, AI companies have to be able to boast ever-increasing valuations.
Anthropic was widely expected to raise its $US100 billion at a company valuation of about $US2 trillion, although recent reports suggest its bankers have brought that back to a range of $US1.5 trillion to $US1.75 trillion because of the recent pullback in the prices of listed technology stocks.
About a year ago, Anthropic raised $US13 billion at a valuation of $US183 billion. In February, it raised another $US30 billion at a $US380 billion valuation. In May, it got another $US65 billion at a $US965 billion. Overall, since its first capital raising in 2021, it has raised more than $US140 billion at ever-increasing valuations of its business.
That escalation in perceived value cannot slow or – horror! – stop if the boom in AI investment and development is to continue. As the leading frontier lab, Anthropic’s performance is critical to the rest of the sector. It has to be able to demonstrate a dazzling sales trajectory and the quasi-realistic hope of an eventual earnings explosion.
It does have explosive growth, with revenue in the first half of this year of $US16.3 billion, up 1300 per cent from the $1.16 billion generated in the same half last year. Its third-quarter revenue growth was, however, a ‘mere’ 45 per cent, so the rate of growth (on a vastly larger base) has been slowing.
Anthropic had a net loss of $US80.7 billion, but most of that ($US72.4 billion) was a non-cash charge relating to the increased value of convertible notes it has on issue. At an operating level, the loss was $US693 million.
Its filing shows it spent almost $US9 billion in the half year on computing power and infrastructure ($US1.1 billion in the same period last year) and has commitments of about $US417 billion to buy compute – chips, data centre capacity and cloud services – in the future, although it has foreshadowed spending more than $US750 billion over the next decade.
Anthropic’s founders conceded in the share sale documents that their investment in compute “may seem irresponsibly large at first glance,” but described it as “bold, but sane.”
More bang for bucks
The encouraging aspect of the company’s numbers is that the revenue growth has been outstripping its growth in spending.
Moreover, Anthropic is getting more bang for its bucks.
In the first quarter of last year, it spent a little more than $US2.50 on compute and infrastructure for each dollar of revenue it generated. In the June quarter, it spent less than US50¢.
If it can maintain that trajectory, the daunting scale of its future capital expenditures might start to make economic sense.
The 355-page prospectus was notable, as was the prospectus for Elon Musk’s SpaceX earlier this year, for its dire warnings of the risks of AI, with 94 pages outlining the potential pitfalls.
Anthropic and its chief executive Dario Amodei have been more vocal than others in the industry in warning of the potentially “catastrophic” and even existential risks associated with the development of AI. The filing also outlined, however, that the technology could cure diseases and grow economies and “revolutionise every area of human endeavour.”
The other unusual element of the document was its disclosure of the company’s proposed governance model which is, to put it mildly, novel.
Within the governance structure is a “Long-Term Benefits Trust,” which has the power to nominate (or remove) four of the company’s seven-member board, although it doesn’t have the power to remove either Dario Amodei or his sister and co-founder Daniela. It appoints its own members, one of whom is former US Federal Reserve Board chair Ben Bernanke. Shareholders can elect the remaining directors.
There are seven co-founders of Anthropic, including the Amodeis, who will receive a special class of stock that confers no economic interests, but grants them 50.1 per cent of the voting interests, cementing their control.
Ordinary shareholders will effectively have no say in the way the company is run, which is a breach of corporate governance norms, but fits with the vision the founders had when they left OpenAI to develop their own AI venture that the technology should be developed responsibly and for the long-term benefit of humanity.
That mission wouldn’t necessarily sit easily within a more conventional profit-maximising listed company.
OpenAI’s Sam Altman and his backers will be watching the fate of the Anthropic IPO nervously. OpenAI was also scheduled to float this year, also targeting a valuation of about $US1.5 trillion, but appears to have deferred its listing into 2027. If Anthropic’s float stutters, OpenAI’s access to funding – its viability – would be threatened.
In fact, Anthropic will provide a test for the entire industry. There have been signs recently, in sliding share prices and spiking bond yields, that the equity and debt markets’ willingness and ability to meet the AI sector’s demands for capital is being stressed.
These are massive lumps of capital that are being sought. SpaceX raised more than $US86 billion of equity in its IPO, then promptly hit the debt market and is now raising another $US40 billion via a deal backed by Broadcom chips. Meta has announced plans for a $US95 billion equity raising.
Most of the hyperscalers’ AI spending is outstripping their cashflows, forcing them to tap debt markets to keep up with the capital expenditures.
At some point – and the process might have started, given the battering the neoclouds have been getting on Wall Street in recent weeks – the sector needs to be rationalised and consolidated to enable the markets to back a smaller and stronger cohort of AI aspirants.
If Anthropic drains the market of available capital, even a successful float might have some unintended but very unpleasant consequences for those AI players that carry too much debt as interest rates are rising and their access to equity is more limited than that of AI heavyweights like Anthropic.
The Market Recap newsletter is a wrap of the day’s trading. Get it each weekday afternoon.
From our partners
About the Author
