Anthropic could soon attempt one of the most extraordinary public market debuts in corporate history. Investors are modeling an IPO valuation above $2 trillion for the five year old maker of Claude, according to the Financial Times , with some scenarios stretching as high as $3 trillion. Anthropic has not publicly set that valuation, but the fact that serious investors are discussing it at all marks a new phase in the AI capital boom.

A $2 trillion Anthropic would signal that investors expect foundation model companies to become core infrastructure for global business, commanding economic power on the scale of today's largest technology and energy companies. It would affect how corporations think about software budgets, automation, vendor dependence, model selection and the cost of intelligence itself.

The bigger question is whether the financials and economics can support the high price. Frontier AI companies face huge compute bills, aggressive price competition, open source challengers and third party models that can erode their grip on customers. So is Anthropic's possible valuation an early glimpse of a durable new computing order, or a familiar case of investors identifying a technological revolution and paying too much for it?

This Is Not The Typical Software Economics

Two trillion dollars buys a lot. It would buy roughly two whole Walmarts at the current market cap. It would put Anthropic in the financial neighborhood of some of the most formidable corporations ever assembled. And it could soon become the price tag attached to a company founded only five years ago.

Anthropic was valued at about $380 billion in February. A May financing round valued it at $965 billion. Secondary market demand has since driven implied prices higher. A $2 trillion public valuation would represent an extraordinary repricing of a private company in a matter of months.

The explanation can be reduced to one word: growth.

Investors cited by the Financial Times expect Anthropic’s annualized revenue to reach roughly $100 billion to $120 billion by the end of 2026, after starting the year at a fraction of that level. At $100 billion of revenue, a $2 trillion valuation represents about 20 times annual sales. Expensive, certainly, but not entirely absurd.

But there is a reason Anthropic needs access to sums that once sounded more appropriate for governments than software companies. Artificial intelligence has become a very expensive software business.

The previous generation of enterprise software produced one of capitalism’s favorite business models. Build the product once, store it on somebody else’s inexpensive server, and then sell another subscription at attractive incremental margins. Frontier AI changes that equation.

Every sophisticated query consumes computation and power. Better models require immense training clusters and significant, memory and compute intensive inference computing. Serving millions of users means continuously buying access to chips, data centers, networking equipment and electricity.

A modern 100 megawatt AI data center can cost more than $4 billion to build and operate, according to a Reuters Breakingviews analysis . Roughly 70% of that can go toward servers and graphics processors.

Anthropic's capital requirements illustrate the scale. Apollo and Blackstone are backing a $35 billion capacity expansion tied to Anthropic and Broadcom technology. Elsewhere in the AI economy, Nvidia is assembling financing structures intended to support more than $500 billion of compute investment. This starts looking less like traditional tech businesses and more like massive infrastructure.

Software companies historically earned premium multiples partly because additional customers cost relatively little to serve. Frontier model companies face a different calculation. Usage creates revenue, but usage also creates substantial cost too. Better models attract customers, but building those models demand another generation of expensive hardware. The deeper issue is whether soaring revenue can produce equally compelling free cash flow.

For perspective, Saudi Aramco's 2019 IPO raised $25.6 billion initially and valued the oil giant near $1.7 trillion. Alibaba's 2014 offering raised $21.8 billion. Facebook entered the public market in 2012 at a valuation of roughly $104 billion. At its IPO, Facebook already had hundreds of millions of users and a highly scalable advertising machine. Yet the public market valued it at about one twentieth of the $2 trillion investors are now contemplating for Anthropic.

SpaceX reset the scale this year, going public around a $1.7 trillion valuation and raising roughly $75 billion. But now Anthropic could eclipse even that. This means that investors are not pricing Claude as another software application. They are pricing Anthropic as infrastructure for a substantial portion of economic activity.

If AI agents write code, negotiate purchases, analyze contracts, conduct research, answer customer inquiries, operate software and perform portions of white collar jobs, the company supplying their intelligence could collect a toll on an immense volume of work.

Open Models Could Attack The Toll Booth

The greatest threat to frontier model economics may come from intelligence becoming cheaper, more portable and less dependent on a handful of American labs. China has become the most aggressive source of open weight competition. DeepSeek has released models that combine strong performance with strikingly low inference costs, and Alibaba’s Qwen family has gained traction among developers and businesses looking for systems they can download, customize and run on their own infrastructure. Z.ai is pursuing the same strategy with its GLM family, and Chinese open models have become popular enough that U.S. policymakers and technology companies are openly debating how to counter their adoption.

The competition isn’t confined to China. In Europe, France’s Mistral AI has made open weight models a central part of its effort to build a credible European alternative to the dominant U.S. platforms. In the Middle East, Abu Dhabi’s Technology Innovation Institute continues to expand its Falcon family, including models designed for Arabic language applications and multimodal workloads.

Even in the United States, there is a movement to open weight models as a counterbalance to Open AI and Anthropic’s dominance. Meta released Muse Glimmer just this past week, Nvidia is developing its Nemotron family, and Mira Murati’s Thinking Machines Lab released Inkling, a 975 billion parameter model that users can download, run and customize. Latin America is beginning to develop regional alternatives as well, including Latam GPT, a Chile led project built around the languages and cultural context of the region.

That global spread changes the economics facing Anthropic, OpenAI and other closed frontier labs. A company buying AI no longer has to choose among three or four proprietary APIs. It can mix a premium frontier model with open source and open weight alternatives for distinct or less demanding work. Some organizations can fine tune those models using their own data and operate them on private infrastructure.

Closed models may retain an advantage on the hardest reasoning, coding and agentic tasks, but businesses do not need frontier intelligence for every invoice, support ticket, document search or internal workflow. Reuters reported that demand is already shifting toward cheaper, customizable open weight systems for many operational tasks.

That matters for businesses. A bank, manufacturer or retailer does not necessarily care whether its invoice extraction system uses the world's smartest model. It cares whether the invoice gets processed accurately for three cents instead of thirty, and is something they can control without being subject to the whims of model availability.

This creates a segmentation problem for companies such as Anthropic. The very best models may command premium prices for difficult coding, science, complex reasoning and autonomous work. Ordinary business activity can migrate toward smaller models, open models or specialized third party systems.

In response to these threats, Anthropic itself is cutting the price of intelligence. Claude Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens. Anthropic says those prices, originally described as introductory, will remain standard rather than rising as previously planned.

While this is great news for customers, it raises a more awkward question for investors. What happens when your product gets dramatically better every year and dramatically cheaper at the same time? Will frontier models be seen as only having a temporary advantage with the rest of the field quickly catching up? In this manner, are the frontier companies just expensive R&D labs for the rest of the world? Will business users hesitate to use new models, waiting for open weight or cheaper alternatives to emerge? And will businesses treat models as interchangeable, leaving little competitive moat?

Is This Dot Com Mania Again?

With these heady valuations and high-visibility IPOs, there are uncomfortable echoes of the past. Money is pouring into infrastructure at a speed almost without precedent. Reuters reported in May that AI related capital spending could reach roughly $800 billion this year, up from hyperscaler capital expenditures of about $260 billion in 2024. Morgan Stanley projected the figure could climb above $1.1 trillion in 2027.

The largest technology companies are expected to spend roughly $750 billion on data centers in 2026. Major hyperscalers are increasingly leaning on debt, private credit, leases and other financing structures to keep the construction machine moving.

The dot com boom and bust cycle at the end of the last millenium offers a useful warning precisely because so much of the underlying optimism proved correct. The internet did remake commerce, media and communication, yet investors still poured capital into companies with fragile economics, unproven demand and valuations built on expectations that outran cash generation. Infrastructure spending surged, business models blurred, and market share often mattered more than profitability. When capital tightened, many of those companies vanished, even as the technology itself kept advancing and eventually produced some of the most valuable businesses in the world.

The dot com boom story was more of a market story than a technology or industry story. Internet and telecommunications stocks pulled enormous amounts of capital toward a relatively narrow group of companies, lifted major indexes and persuaded investors that extraordinary future growth justified extraordinary present valuations. When those expectations broke, the damage spread far beyond failed startups. The Nasdaq lost roughly three quarters of its value between early 2000 and late 2002, destroying trillions of dollars in market wealth. While the technology survived, many investors did not escape the cycle intact.

That history matters more today because AI enthusiasm is increasingly embedded in the broader stock market. Technology stocks now account for more than 39% of the S&P 500’s market capitalization, according to Reuters, above their weight during the 2000 dot com peak. Add AI-exposed giants outside S&P 500’s official Information Technology sector such as Alphabet and Meta, which S&P classifies as Communication Services, and Amazon, which sits in Consumer Discretionary, and the market’s exposure to the AI investment cycle becomes even larger. The top 10 U.S. stocks alone account for roughly one third of the market's value, according to Morgan Stanley data cited by Reuters.

That creates a familiar market vulnerability. If investors continue assigning premium valuations to chipmakers, hyperscalers, data center operators and frontier model companies, rising AI expectations can lift indexes even when much of the market participates less fully. The reverse can happen just as quickly. A slowdown in AI revenue, weaker returns on massive capital expenditures or evidence that cheaper models are compressing margins would not need to destroy artificial intelligence to hurt the stock market. It would only need to force investors to lower the prices they are willing to pay for future AI profits. With so much market value concentrated in the companies funding and supplying the boom, that repricing could drag major indexes lower.

There are warning lights now. Reuters reported that rising AI investment is putting free cash flow under pressure at the major hyperscalers. Their projected capital spending increase through 2027 is running materially ahead of expected growth in operating cash flow.

Another Reuters analysis pointed to concerns around circular financing inside the AI economy, where technology suppliers, infrastructure providers and model companies can become customers, investors and financiers of one another. Follow the money long enough and sometimes it starts arriving back at the same address.

What Businesses Should Take From A $2 Trillion Anthropic IPO Valuation

Business leaders should pay attention not to the headline-grabbing valuation, but focus more on what investors are betting on. If the valuation comes anywhere close, capital markets will be making an enormous wager that AI moves from technology add-on to infrastructure necessity.

The opportunities and dangers of that are significant. Companies should experiment with many model alternatives and keep vendor commitments flexible. They should track the economics of each use case rather than celebrating AI usage as a metric by itself. Ask what a model replaces, accelerates or makes possible.

Most of all, watch unit economics. A simple prototype that saves an employee ten minutes is interesting, but a production system that removes $40 million of annual expense is much more significant for the business. And critically, protect corporate budgets and market exposure against the volatility that could follow if AI expectations reset.

The remarkable part of this discussion of Anthropic’s potential valuation is that the market is already seriously entertaining the price. Five years after Anthropic's founding, investors are discussing a valuation once reserved for the largest oil companies, technology monopolies and industrial empires on Earth. That tells us something profound about AI enthusiasm. What it cannot tell us is whether Anthropic can earn its way into that valuation, or whether the market is once again pricing a technological revolution faster than the economics can support.