Watch what happens with Anthropic
By the Editor
Once in a while, a business news story can tell us that the deeper currents in our economic life might be about to change direction.
At the moment that story is about Anthropic, one of the top creators of so-called “frontier” AI models, which wants to list its shares on the US stock market.
No date has been set yet, but last month Anthropic published a US regulatory form called an S-1 which lists all the risks that investors in its shares might be taking on (of which, more later).
Since then, the share prices of some companies in the AI-industrial complex have fallen significantly - ostensibly because of the US mid-term elections and in the case of giant chip-maker Nvidia, a sex-and-spying scandal - and the mood in the markets is starting to shift.
After more than two years of argument about whether or not the AI boom is a bubble, mainstream opinion seems to be moving towards the long-held view of sceptics that it is. This shift in sentiment matters because it's well-known that markets are driven by groupthink among people with highly imperfect knowledge and a very narrow set of motivations.
Strikingly, market speculation is already marking down the value which a share offering might put on Anthropic from as much as US$2 trillion only a few weeks ago to US$1.5 trillion or less.
(Update: after this article was published, Anthropic reportedly announced that it is aiming for a share price which would value the company at up to US$1.4 trillion. It looks as if the company and its bankers are still trying to judge what the market will take.)
What goes up can also come down
Against a fast-changing backdrop, outsiders can only guess what might happen with Anthropic’s initial public offering (IPO).
Perhaps Anthropic might still be able to sell its shares at a price high enough to convince investors in general that the AI investment boom has more life in it.
Or the shares might have to be sold at such a low price that a critical mass of investors conclude that the boom is ending. It is also possible that the IPO is postponed to avert that outcome, which would itself send a pessimistic signal.
Anything but a positive outcome for the IPO could make things very messy, because it would imply that hundreds of billions of dollars’ worth of existing investments in AI are not going to pay off. That is a view that AI sceptics have held for a long time and which some much-quoted market analysts are also taking.
But even a relatively successful IPO would not address Anthropic’s fundamental problem, which is the commercial problem of AI in general: sophisticated AI models are incredibly costly to develop and run and show no signs so far of generating the kinds of profits which would justify those high costs.
Anthropic has reportedly pledged nearly US$900 billion over the coming years to buy or lease computing capacity. At the moment the company makes losses, however, and its revenues are growing but still small given the scale of its commitments.
To meet these commitments and satisfy its new stock market investors, Anthropic will have to become very profitable in a matter of years. Unlike Amazon, Google and Microsoft in their early days, however, Anthropic does not seem to have a first-mover advantage which is so durable that competitors cannot hope to catch up.
Anthropic’s Claude models are in competition with its rival OpenAI and with other US and Chinese AI models, some of which are much cheaper for customers. Anthropic, like its competitors, is also facing hostility from communities against the building of more of the data centres on which its models rely. Here is a good summary of the problem from a critical perspective.
Anthropic matters not only because of its great size but also because it is entwined in complicated and often opaque ways with the rest of the AI industrial complex, notably Amazon, Microsoft and Google which provide it with computing capacity and market its products but are also its competitors. (For more on the “cloud giants", read this).
Anthropic admits these problems in its S-1. Its response is basically that it plans to invest even more money to keep ahead of its competitors, a strategy which it concedes may not work. The IPO requires investors to take a huge gamble on the future, in other words.
More could be said about Anthropic, for instance about its attempts to present itself as an entity with a benevolent social mission (its founders set out this view in a letter which starts on the second page of the S-1.)
The key point is this, however: Anthropic is so big and important within the AI industrial complex that whatever happens with its IPO will not only be a signal of what is happening to the AI boom but also a catalyst for whatever happens next.
And right now, the outlook is not looking nearly as rosy as even a few weeks ago.
Why it matters
The growth of commercial AI (including data centres) relies on huge, continuing inflows of capital. A severe loss of investors’ faith in AI's future profitability would dry up those flows and threaten existing investments.
Amazon, Microsoft and Google are big and lucrative enough that they would almost certainly survive a messy end to the AI boom. However, other companies within the AI-industrial complex might get into financial trouble and, therefore, their investors.
As we learned from the 2007-8 global financial crisis, modern finance is so vast, opaque and intertwined that the scale of the risks, and their location within the system, may only become clear once a crisis is well underway.
That makes it unwise to try and predict whether we are looking just at some froth coming off the US stock market, or a painful correction which governments and central banks are nonetheless able to manage, or an almighty crash with world-historic effects like that of 2008.
Since no-one can be sure how bad things might get, the tendency is for commentary (including this article) to hedge its bets. That said, the systemic risks to the global financial system are big enough for relatively sober voices of capitalism like the Bank for International Settlements and the IMF to have warned about them.
The end of the AI boom could cause all sorts of painful disruptions in the wider economy. It would also enlarge the political space for critics of corporate power across civil society to push home a vital point on govermments, with strong public support this time.
Behind all the visioneering and the sci-fi gloss, the strongest driving force of the AI boom has been the desire of US Big Tech to extend its dominance over the digital economy and prolong its run of super-profits, with total indifference to the effects on society and the climate.
Governments across the West, far from trying to check Big Tech’s ambitions, have largely acquiesced in them in the hope that AI can magically solve their own pressing problems. If the boom ends badly, that simply won’t be a credible position any more.
An end to the AI investment boom would not end AI, given that new technologies don’t tend to be uninvented. But it must mark an end to, and rebuttal of, all the cant about free markets, “innovation” and “growth” which has been drummed into us for so many years and which has justified the stance of those governments.
Nor should we willing to transition from neoliberalism to a system based on costly public subsidies and protections for private capital and favouritism towards well-connected big business elites, all dressed up in a rhetoric of “national interest”.
New technologies must be controlled by our societies and put to the uses which societies decide through truly democratic processes, not left as a plaything in the hands of a greedy, reckless, super-wealthy minority of humanity who are currently trying to play God through the stock market.