TRUMP, XI AND TAIWAN 

“I’ve just emerged from nearly three hours of briefing and discussions with the President of the Republic and the intelligence and security services. In my 50 years, I have rarely felt such a sense of vertigo. Vertigo at the gravity of the situations in the Middle East and Ukraine, and at the predatory madness driving the world toward chaos.”

That was Guillaume Lacroix, a French politician, speaking late last week. He is not alone in his alarm. Dark warnings about the risk of war have been emanating from European capitals for weeks. The alarm bells ringing in Silicon Valley about AI have been heard around the world. An energy crunch is building. The bond markets are signalling a possible debt crisis.

This is a fraught moment in world affairs. In theory, that makes it fortunate that the leaders of the two most powerful nations will meet later this week. There is a huge amount for Donald Trump and Xi Jinping to discuss: wars, energy, AI, trade, Taiwan, the White House ballroom.

But the deep suspicion between the US and China — and the limitations of the two leaders involved — make it highly unlikely that there will be real progress on any of the big issues.

One of the main reasons that the US and China continue to circle each other warily — without engaging on issues of common interest — is their continuing struggle over trade.

When Beijing first hit back against the tariffs Trump announced on April 2 2025, the US was dismissive. Scott Bessent, the US Treasury secretary — always fond of gambling metaphors — claimed that Beijing was playing poker “with a pair of twos”. In reality, the restrictions that China imposed on the exports of rare earths and critical minerals looked more like a straight flush. The production lines of key US industries were struggling within weeks.

China and the US signed a trade truce in South Korea in October 2025, which may be renewed in Washington. But the exchange of fire over trade increased mistrust on both sides. Both Beijing and Washington are now on guard against any form of co-operation that could be turned against them in the future.

With mutual trust almost non-existent, there is little chance of the US and China working together effectively on the common threats that transcend national borders — such as rogue AI or climate change.

America’s leading AI companies are pleading for a joint US-China approach to regulation. Bessent says the two nations have just agreed to an AI dialogue. But discussions will be limited by the inherent rivalry between Washington and Beijing. Trump maintains “We’re leading China in AI” and says he wants to “keep it that way because whoever wins AI, wins.”

For its part, the Chinese side is suspicious that calls emanating from the US for a joint approach to AI are an effort to lock in America’s narrow lead in a defining technology. Recent articles in the Chinese official media have focused on the threat that AI might pose to the continuing rule of the Communist Party — rather than to humanity as a whole.

Trump likes to dismiss any idea he finds inconvenient as a “hoax”. That is the word he used about threats posed by AI. It is also the word he has long employed about climate change. The US president has instead prioritised using America’s ample oil and gas to pursue “American energy dominance” — and boasted about his withdrawal from the Paris climate accord and “dozens of other Radical Left international climate accords.” That policy has allowed China to cement its own dominance of the industries that underpin renewable energy — such as batteries, solar panels, wind turbines and EVs.

Both countries are threatened by the energy supply crunch caused by war in the Middle East. China has dabbled in Middle East peacemaking in the past and is a major oil importer from the region.

But even if the US wanted Beijing’s help in ending the conflict — which is unlikely — Xi has no reason to stick his hand into the bin fire that Trump has started in the Gulf. The scope for US-Chinese co-operation over the Ukraine war is even more limited. Beijing continues to play a crucial role in keeping the Russian economy afloat.

Rather than playing a role as a peacemaker, there remains a risk that China will initiate a third major global conflict by invading or blockading Taiwan. The US is struggling in Iran and American weapons stockpiles have run down. Unlike President Joe Biden, Trump has downplayed the idea that the US would fight for Taiwan.

All that might increase the temptation for Xi to make his move soon. On the other hand, the Chinese leader will have noted how the US and Russia have struggled to win wars against weaker nations, which may incline him to concentrate on increasing political pressure on Taiwan instead.

Both Trump and Xi love pomp and ceremony. There will be plenty of that in Washington — but there is likely to be little serious engagement on the issues that matter to the world. This will not be a modern equivalent of the Reagan-Gorbachev Reykjavik summit of 1986, where the two leaders discussed radical nuclear disarmament and glimpsed an end to the cold war.

Trump and Xi represent rival nations. But, unfortunately, the two leaders have many of the same flaws — narcissism, nationalism, narrowness of vision. The world needs far-sighted leaders. That is not what is on offer.

gideon.rachman@ft.com

In the long, rich history of capital markets, investors have been asked to guess the value of some pretty weird stuff. Deep-sea mining, quantum computers, Martian colonies, breeders of lab monkeys — all have found homes and share prices on public stock exchanges. Now there’s a new challenge: valuing the big AI labs, with their unique potential to change the world or destroy it.

One of these is Anthropic, the AI outfit co-founded by Dario Amodei. The maker of the Claude chatbot was this month poised to file for an initial public offering that could value it at $2tn. Until, that is, fears of an AI apocalypse crashed their way into the mainstream. OpenAI, Anthropic’s chief rival, has also been planning an IPO, though founder Sam Altman has now pressed pause on that until next year, saying floating earlier would be “ill-advised”.

The debate around whether AI companies’ future models could extinguish the human race has understandably tipped the spotlight away from the question of what investors should pay for their shares. But this remains a good moment to assess how, exactly, the valuation question can best be answered — and whether $2tn for Anthropic is the stuff of financial hallucination.

There are three ways to think about it. And along the way, investors will have to consider a giant unpriceable risk and a giant unpriceable opportunity.

Multiples of madness

In a world where businesses make regular things in predictable ways, there’s a simple way to decide what they’re worth: forecast their earnings a year or two hence — or revenue if they have no earnings — and put that number on a multiple similar to that of listed peers.

The first reason this is tricky for Anthropic is that the company is changing so rapidly, even by Silicon Valley standards. Founded in 2021, the creator of Claude hadn’t even produced a dollar in revenue until 2023. By August, it was making around $65bn on an annualised basis, although such unofficial numbers should be treated with kid gloves. This growth seems to have come as a surprise to Amodei too. Only 18 months ago, Anthropic expected its revenue in 2027 to be just $12bn.

Some content could not load. Check your internet connection or browser settings.

Now, some investors predict a revenue run-rate of $320bn by the end of next year. They are not impartial, of course. But if they are right, then $2tn would represent a valuation of just seven times its 2028 sales. That’s a little less than Microsoft, according to LSEG. SpaceX, meanwhile, trades at 16 times that year’s revenue. Take that as the benchmark — they are both companies with wild aspirations and charismatic leaders — and Anthropic could in future be worth $5tn.

It might seem foolish to use near-term estimates to gauge the value of a company whose future is so hard to map out. But that’s exactly what Wall Street does. In the case of SpaceX, underwriters Goldman Sachs and JPMorgan were among the many who set their generous target prices by taking a punt on profit one or two years from now, and slapping on a peer-group-inspired multiple.

Fortunately, the financial community offers plenty of other ways to crack the valuation nut — some of which suggest Anthropic’s mooted $2tn price tag may be less generous than it appears.

TAM-tastic growth

Much debate about Anthropic’s worth will revolve around its “total addressable market”. The idea of deriving a company’s value from the size of the potential revenue pool is decades old but became common currency in the 1990s when analyst Henry Blodget used it to suggest Amazon was worth $400 a share — almost double where the stock was at the time.

Now, TAM is de rigueur in any venture capitalist’s slide deck or IPO prospectus. Its usefulness is another matter. Uber touted a $12.3tn TAM in its 2019 listing; its annual revenue today is less than $60bn. WeWork’s $3tn addressable market of office properties didn’t save it from bankruptcy. But for a company like Anthropic, TAM is as good a place as anywhere to start.

Investors aren’t groping totally in the dark, thanks to SpaceX. Elon Musk’s company, which makes the Grok chatbot as well as owning social network X, claimed in its May listing documents that “enterprise apps”, which would include its products and services backed by AI, make up a market worth $22.7tn — a little more than the entire annual revenue of the S&P 500.

With such a large theoretical market to play for, a little goes a long way. Even the mighty Microsoft’s revenue is less than 2 per cent of what Musk sees as the industry’s overall opportunity. If Anthropic can muster 3 per cent, it would stand to make nearly $700bn of revenue. Pop that on a multiple of 10 times and discount it back assuming it takes three years to achieve, and out comes a valuation of $4.5tn.

Some content could not load. Check your internet connection or browser settings.

Even though it’s barely four months since SpaceX made those projections, TAM inflation has already kicked in. Anthropic’s filings may tell investors it sees a market worth $30tn, The Wall Street Journal has reported. Analysts at Morgan Stanley, likely to be named as an underwriter on Anthropic’s IPO, estimated last week that generative AI could tap a market twice that size.

Say AI ends up taking 10 per cent of what Morgan Stanley sees as the $60tn market for “knowledge work” and digitisable consumer spending — a feat that incidentally would involve replacing about 100mn workers. If Anthropic can swipe a third of that, it would end up with $2tn in annual revenue. Perform the 10-times multiple trick again, assume it takes five years to get there, and presto: a valuation today of $10tn.

If even that doesn’t slake your thirst for AI exuberance, consider the real bull case, based not on markets AI might cannibalise but on new economic activity it might create. Anthropic’s research arm has posited that such technologies could unlock more than $10tn of US GDP by 2030 in its “extreme” scenario of widespread adoption and highly productive AI.

Simplistic maths suggests that’s equivalent to $100tn of equity value. And if all of that sounds uncomfortably vague and speculative, then welcome to the AI boom.

AI now, pay later

Even if you believe that TAMs and new sources of GDP are real, there is a catch: it’s not clear that it’s the AI labs who will pocket the money. Anthropic and OpenAI are dab hands at making models, but they’re competing with Alphabet, Meta Platforms, Nvidia, Chinese players and all of the other people along the supply chain, from customers to providers of chips and data centres, who will try to claw back some of AI’s economic benefits.

The big risk is that the architects of AI themselves end up getting crumbs. That could happen if their models start to look fungible, meaning that for a user, one is much like the next, and the price is therefore set by the cheapest. Palantir boss Alex Karp talks about “commodity cognition”, a fancy way of saying that he thinks the value of AI models themselves will slide ever closer to zero.

Karp reckons that value should accrue to whoever owns the data or makes the application in which the AI model runs — and, of course, to companies like Palantir that show companies how to deploy it. In that world, enterprises will be as ambivalent about who provides the AI “tokens” as they are about which company pipes water or electricity to their office. The engine makers of AI will be able to charge enough to cover their costs, but not much more.

Some content could not load. Check your internet connection or browser settings.

Competition among different AI model-smiths is indeed fierce. China is undercutting the US with products like Moonshot AI’s Kimi and DeepSeek, offering a slightly lower quality for a much lower price. Meta and Google parent Alphabet have huge advertising businesses to subsidise their AI exploits, meaning they can afford to give away models cheaply, or even for free.

Moreover, signs of a price war are already clearly visible. Google and OpenAI have slashed prices for some of their less advanced offerings. Many customers, including some of Anthropic’s investors and underwriters, are deploying “routing” software that diverts tasks to the most cost-effective AI engine. The days of reckless “tokenmaxxing” are coming to an end.

Meanwhile, chipmaker Nvidia — an investor in both OpenAI and Anthropic — has potentially blown both companies a raspberry by acquiring Hugging Face, a major distribution platform for free-to-use “open weight” models. These increasingly stand up against Anthropic and OpenAI’s “closed” models, at least on less advanced tasks.

Not everyone believes in commodity cognition. AI founders tend to argue there will always be a premium on the best models, and that companies will pay top prices to solve what IT folks often refer to as the “hard 5 per cent” of problems.

Besides, the leading labs are racing towards “recursive self-improvement”, at which point AI models will upgrade themselves with little or no human help. If that occurs, believers think it will usher in an “intelligence explosion” that could put a huge moat around the winners and render all lesser AI all but worthless.

In some ways, that’s a risk rather than an opportunity. If self-improving superintelligence arrives, it may indeed unlock that $10tn-plus of extra GDP that Anthropic sees as possible. But it could also turn the entire world on its head in ways that make asset prices, as we know them now, meaningless.

Reality check

In truth, the threat of models going to zero could push the AI labs like Anthropic to focus more on making software-like “apps” instead. Amodei’s company and OpenAI are particularly exposed, since unlike Musk’s, which spans satellite communication, space transit and cloud computing rental, they only really do one thing.

The maker of Claude is already getting wise to the benefits of what tech bros refer to as “moving up the stack”. Its launch of a new plug-in for lawyers in January is what triggered the so-called SaaSpocalypse, a savage sell-off of shares in companies that sell “software as a service”. OpenAI launched a new research and writing tool called “Astra for Law” on Thursday.

Some content could not load. Check your internet connection or browser settings.

When fears of software’s demise hit this year, stocks in companies such as Salesforce and ServiceNow shed a collective $300bn in just a couple of days. That suggests the prize for Anthropic in disrupting the software market is potentially large — maybe enough to justify a $2tn valuation. That amount is, after all, less than one-third of the combined market capitalisation of software and services companies in the S&P 500 index.

Many software firms have triumphantly pronounced the SaaSpocalypse to be a fiction; Salesforce chief Marc Benioff calls it “crazy nonsense”. But it is too early for such complacency. Salesforce’s recent tie-up with Anthropic to create a new product called Claudeforce, fusing Anthropic’s models with Salesforce’s customer service wares, suggests the lines are going to blur.

The last word

Any valuation model for Anthropic or OpenAI would be remiss if it didn’t include that most unusual of risk factors: a potential apocalypse.

This is not a new concept — witness the stark, one-sentence “Statement on AI Extinction Risk” co-signed by Altman, Amodei and hundreds of other researchers and experts in 2023. Theories of AI doom have been percolating in online forums for decades. But Amodei’s call for a collective slowdown in model development this month, and stark warnings from former Anthropic researcher Jacob Coxon, have cranked up the volume on such fears.

What should an investor think about this? There’s no easy answer. Clearly, if AI really does make humans extinct, every investment goes to zero. But lesser destruction — say, the “hundreds of billions of dollars in damage” about which Amodei frets in his recent essay, could leave a big hole in Anthropic’s valuation were it to end up liable for such a mishap, to say nothing of its reputation.

Conversely, there could be huge spoils in store for whoever creates trustworthy AI powerful enough to act as a bodyguard against other labs’ rogue bots. In the days after Amodei’s call to “pace the frontier”, cyber security and virus-scan stocks rose in double digits. If a malicious AI can create hundreds of billions in damage, then the digital equivalent of Moderna’s Covid vaccine ought to be worth many hundreds of billions too.

Counterintuitive as it might seem, the fear of mass destruction could prove helpful, even if it results in all the labs slowing their roll. With godlike AI occupying less of their brain space, Anthropic and OpenAI will have far more incentive to squeeze revenue and profit out of the technologies they already have — and spend more cautiously, or on more focused things. That might preclude a $100tn valuation. It does, though, make $2tn more reasonable.