Of course, some Wall Street forecasts already anticipate even higher revenues by 2030. Forecasting extraordinary growth is practically a condition of employment at Top Tier City Banks during a technology boom, given their underwriting obligations. The hard part is determining who pays for it, how much they pay, and how much of that money actually stays with the company after infrastructure suppliers take their cut.

Takeaways by Dark Side of the Boom™

• Wall Street has spent days making room for Anthropic. Now comes the awkward business of figuring out what investors have actually bought.

• Behind the IPO excitement sits a mountain of infrastructure obligations and a revenue hurdle that makes today’s technology giants look surprisingly modest.

• The AI boom has no shortage of capital or ambition. What remains considerably less certain is who eventually picks up the bill.

• Nvidia’s retreat from China could reshape the competitive landscape in Anthropic’s favour, but there is a difference between winning the technology race and making the numbers work.

Now Comes the Hard Part

Wall Street has spent the past several sessions clearing the deck for Anthropic, and now the moment has arrived. The AI developer has priced its IPO at $83 a share, putting a $1.5 trillion valuation on a business whose long-term economics remain remarkably difficult to pin down. The market has certainly made room for the deal. Whether it has made room for the financial obligations that come with it is a different question.

Enough ink has already been spilled over Anthropic’s valuation to float an aircraft carrier, but of course I feel obliged to spill a little more. The bulls see the next great technological monopoly; the bears see an infrastructure-spending binge dressed up as a software company; and somewhere between the two sits a rather inconvenient question: what does this business actually need to earn to justify the price?

Forget trying to calculate Anthropic’s intrinsic value to three decimal places. Anyone claiming that level of precision on a business whose economics could change dramatically over the next decade is selling confidence rather than analysis. The more useful exercise is to work backwards from the capital being committed and establish what kind of financial performance would make the investment worthwhile.

That takes us away from the familiar parade of AI adoption curves, productivity miracles and total addressable markets large enough to swallow entire national economies. Those forecasts make excellent investor presentations. But eventually someone has to pay the electricity bill, replace the chips and generate a return for shareholders.

Anthropic’s S-1 filing identifies about $901 billion in off-balance-sheet financial commitments, largely tied to computing capacity and infrastructure. Another $150 billion to $175 billion in commitments has yet to be assigned a definitive value. Round the combined obligations to roughly $1 trillion, and we have a starting point that should focus a few minds.

These commitments stretch over about 15 years, but an estimated $700 billion comes due within the next five. That timing matters because AI infrastructure is not a toll road you build once and collect revenue from for the next half-century. Chips become obsolete, computing requirements evolve, and the hardware that looks cutting-edge today can become yesterday’s expensive furniture remarkably quickly.

For the sake of argument, make a generous assumption: that trillion dollars represents all the capital Anthropic will need to support the business. No additional spending surprises, no unforeseen infrastructure requirements, no second mountain of investment sitting behind the first.

What should investors expect to earn on that capital?

The established technology giants provide a reasonable benchmark. Over the past five years, Alphabet, Apple, Meta, and Microsoft have generated average returns on invested capital of about 19%, 43%, 19%, and 20%, respectively.

Take 20% as the required return for Anthropic. That is hardly an outrageous hurdle for a company valued as one of the most important businesses of the next decade. On $1 trillion of invested capital, that translates into $200 billion in annual after-tax operating profit.

Now the arithmetic starts getting interesting.

Assume Anthropic eventually achieves a 40% operating margin. That would put it ahead of Alphabet at 30%, Apple at 31% and Meta at 37%, while leaving it slightly behind Microsoft’s 44%. In other words, we are already giving Anthropic the benefit of some exceptionally favourable operating economics.

Apply an effective tax rate of 17%, broadly consistent with its mega-cap peers, and work backwards to the revenue required to generate that $200 billion profit.

There it is. Roughly $602 billion in annual revenue just to earn a 20% return on the assumed trillion-dollar capital base.

This is a deliberately simplified calculation, especially because contractual infrastructure commitments aren't automatically equivalent to accounting invested capital. But as a measure of the scale involved, it does the job rather nicely.

For comparison, Alphabet currently generates approximately $471 billion in annual revenue, Apple $478 billion, Meta $241 billion and Microsoft $346 billion.

Anthropic would need to generate considerably more revenue than any one of those companies currently produces. Not more than some promising AI start-up or the next generation of software providers. More than businesses that have spent decades building enormous global customer bases, distribution networks and ecosystems that competitors struggle to penetrate.

Another way to look at it is through the entire software and IT services sector of the S&P 1500. Across 76 companies, aggregate annual revenue is approximately $908 billion.

Anthropic’s illustrative $602 billion requirement would represent roughly two-thirds of that entire sector’s current revenue.

And we are supposed to treat the valuation debate as though the difficult part is deciding whether the IPO opens 15% or 25% above the offer price.

Of course, some Wall Street forecasts already anticipate even higher revenues by 2030. Forecasting extraordinary growth is practically a condition of employment at Top Tier City shops during a technology boom, given the underwriting obligations. The hard part is determining who pays for it, how much they pay, and how much of that money actually stays with the company after infrastructure suppliers take their cut.

Which brings us to the question that should sit at the centre of every long-term Anthropic investment thesis: where does $600 billion of annual revenue come from?

Corporate AI adoption will undoubtedly expand. Businesses will automate processes, replace elements of traditional software, deploy increasingly sophisticated agents and spend serious money integrating AI into their operations. But not every dollar spent on artificial intelligence represents a new dollar of economic activity.

Some of that expenditure will displace existing software subscriptions, consulting fees, outsourced services and labour costs. Some will be captured by cloud providers, hardware manufacturers and the companies controlling distribution. And some may disappear into the familiar technology-industry tradition of competing away attractive margins just as the customer base starts getting interesting.

Investors also need to distinguish between revenue generated from customers with genuine economic demand and money circulating through an AI ecosystem in which suppliers invest in customers, customers purchase infrastructure from suppliers, and everyone reports another impressive quarterly growth figure.

The revenue can be perfectly real while the ultimate economic return remains considerably less spectacular.

None of this means Anthropic cannot become a financial success. It means that a company priced at $1.5 trillion is already being asked to clear a hurdle that most successful businesses never encounter in their entire existence.

And that is before we introduce Washington, Beijing and the increasingly complicated relationship between Anthropic and the company supplying much of the industry’s computing horsepower.

Nvidia, Anthropic and the China Divide

The Nvidia-Anthropic relationship is one of the more intriguing contradictions in the AI boom. They are financial partners, technological dependants and, until recently, advocates of competing approaches to one of the industry’s most consequential policy debates: whether China should have access to America’s most advanced computing chips.

Jensen Huang, Nvidia’s former chief executive, spent years arguing that restricting semiconductor exports to China was a strategic mistake. He argued that American technological leadership benefits from having the world use American technology. Shut China out of the market, and Washington risks accelerating the development of competing Chinese systems while handing away a substantial commercial opportunity.

That argument had some perfectly sensible economic reasoning behind it. There was also a rather large commercial incentive. China represents an enormous market for advanced computing, and one short-selling research estimate suggested that more than a fifth of Nvidia’s computing revenue was supported by Chinese demand despite existing restrictions.

It is considerably easier to defend free trade when a meaningful portion of your revenue depends on it.

Anthropic chief executive Dario Amodei has long occupied the opposite side of the debate. Alongside co-founder Jack Clark, he was reportedly advocating tighter controls on China’s access to advanced chips as far back as 2021.

The strategic argument is straightforward: advanced AI computing capacity is increasingly viewed as a national-security asset, and allowing geopolitical competitors unrestricted access could undermine America’s technological advantage.

The commercial argument is equally straightforward, even if it gets less attention. Restrict Chinese developers’ access to the most advanced Nvidia hardware, and American model builders gain a potentially significant competitive advantage.

Funny how national-security policy and corporate self-interest occasionally travel in the same direction.

The situation becomes more interesting when you remember that Nvidia is financially backing Anthropic. The chipmaker invested earlier this year and committed to purchasing another $10 billion of Anthropic shares at the IPO price.

So Nvidia has been financing a company whose leadership supports restrictions on one of Nvidia’s most commercially important export markets. That is not necessarily irrational. Nvidia’s future growth depends on far more than Chinese sales, and ownership stakes in leading AI developers could prove enormously valuable. But it illustrates how quickly the commercial relationships inside this industry are becoming intertwined.

Recent developments suggest the balance is shifting decisively towards Anthropic’s preferred outcome.

After Huang’s departure, Nvidia announced it would stop selling advanced chips to China and establish a committee to oversee national-security and trade compliance. Meanwhile, Washington is moving to close the loopholes that allowed Chinese companies to obtain restricted computing capacity indirectly by renting access from data centres in south-east Asia and Brazil.

That changes the commercial equation in several ways.

First, it reduces Chinese AI developers' ability to compete on equivalent computing resources, potentially slowing their technological progress or forcing them onto less efficient domestic alternatives.

Second, it changes where Nvidia’s advanced hardware can be sold. Computing capacity that might previously have served Chinese demand could increasingly flow into infrastructure supporting American AI developers, including Anthropic.

Third, it reinforces the strategic importance of the relationship between Nvidia and the largest US model developers. When access to computing capacity becomes a geopolitical question, securing supply is no longer simply a procurement exercise. It becomes a competitive advantage.

That is a favourable development for Anthropic, although there is a catch.

Restricting China’s access to advanced hardware may strengthen the American AI industry’s competitive position, but it does not magically improve the economics of building and operating enormous computing facilities. Nvidia still has to earn an acceptable return on the chips it sells, infrastructure operators still have to recover their investments, and Anthropic still needs customers willing to pay enough for AI services to cover the whole arrangement.

Geopolitical protection can improve a company’s relative position without guaranteeing that the underlying industry earns attractive absolute returns.

There is also an uncomfortable circularity to consider. Nvidia invests in Anthropic, Anthropic commits vast sums to computing infrastructure, and the infrastructure providers purchase Nvidia hardware. That arrangement can accelerate technological development and drive genuine commercial growth, but it also demands careful attention to where the ultimate customer revenue comes from.

The more capital flows around this ecosystem, the easier it becomes to mistake financial activity for sustainable economic demand.

That brings us straight back to the IPO.

At $1.5 trillion, Anthropic is not being valued as an experimental technology business that might eventually stumble onto a profitable model. It is being priced as an emerging industrial giant, with infrastructure obligations approaching a trillion dollars and investors expecting returns comparable to the most successful technology franchises in history.

The export-control landscape may now be tilting in Anthropic’s favour. Nvidia’s strategic repositioning could strengthen its access to computing power while making life considerably harder for Chinese competitors. Those are meaningful advantages, particularly in an industry where access to the latest hardware can determine who stays competitive.

But they do not change the original arithmetic.

Wall Street cleared the deck to make room for Anthropic’s $1.5 trillion arrival. Getting investors into the deal was the easy trade. Finding $600 billion in annual customer revenue without sacrificing the margins needed to justify the valuation is where the real price discovery begins.