How was the $900 billion Anthropic forged?

Author: Su Yang, Tencent Technology

Citing people familiar with the matter, Bloomberg disclosed that the AI startup Anthropic is in early-stage talks with investors, aiming to raise at least $30 billion in new funding, with a valuation exceeding $900 billion.

Sources said the fastest timeline for this round of financing is expected to be completed by the end of May 2026, but the transaction has not yet been finalized and no term sheets have been signed.

If the fundraising succeeds, Anthropic would not only leave OpenAI (valued at $852 billion in March) far behind, but also challenge the market capitalizations of tech giants such as Apple and Microsoft. It’s worth noting that the investors who backed the company early have largely chosen to sit this round out.

$30 billion in annual revenue and a 40% gross margin

Why can a company see its valuation surge 15-fold within 14 months? The answer seems straightforward: growth rate.

Based on data from public reports, Anthropic’s annualized revenue rose from $1 billion as of December 2024 to $30 billion by the end of March 2026. This means that over consecutive years, it maintained a growth rate of more than 10 times.

Such a growth curve may have no precedent in the history of enterprise software.

In the Fortune Global 500 list, eight of the top ten companies are Anthropic customers. There are more than 1,000 enterprise accounts; they spend more than $1 million a year on Claude. In particular, its developer-focused coding product, Claude Code—launched in May 2025—reached $2.5 billion in annualized revenue by February 2026, and enterprise subscriptions more than tripled within the first six weeks of the year.

At a valuation of $900 billion and annualized revenue of $30 billion, the price-to-sales ratio is about 30x. This multiple sounds extreme, but supporters are betting on the future. They believe that a company growing at 10x annually cannot be valued in the usual way. Its pricing logic assumes that by 2028 it will still be able to maintain a similar compound growth rate; at that point, looking back, today’s valuation would be reasonable.

Regarding Anthropic’s revenue, its competitor OpenAI has raised its own doubts. OpenAI argues that the $30 billion annualized revenue Anthropic reports uses a gross revenue accounting method—when customers use its models through platforms such as Amazon Cloud and Google Cloud, all end-user spend is recorded as revenue, and then the payments made to cloud platforms are listed as expenses.

OpenAI estimates that after deducting these intermediary costs, Anthropic’s true annual revenue is closer to $22 billion. The $8 billion gap is purely a methodological choice, but at the time of the IPO it will become a key focus for the market and regulators.

What matters even more than the revenue accounting is costs.

According to materials, Anthropic plans to spend about $19 billion in 2026 on training and inference computing. That figure is nearly equal to its full-year revenue. More troubling is that because inference costs are 23% higher than expected, its gross margin has been compressed to roughly 40%—a level far below most mature enterprise software companies.

Anthropic is not yet profitable and is expected to turn profitable only in 2028. For a company whose valuation is approaching $1 trillion, this combination of financial metrics is indeed unusual.

The arms race for computing power driven by valuation

Why does Anthropic need to raise so much money?

On paper, it’s for development and expansion—but in reality, of this $30 billion, the vast majority is to pay for the computing infrastructure it has already committed to but has not yet built. This looks like a completely different model from traditional tech financing.

In the past, startups raised funds to refine products and expand markets, and then used growth to gradually match their valuation. But in the AI era, startups need to raise at very high valuations first, using the money to lock in massive computing power for the future, and then hope that this computing power will drive leaps in model capabilities—leading to revenue growth—and ultimately to prove that the expensive valuation is justified.

It’s like the chicken-and-egg debate.

Now, the valuation-driven promise of computing power requires even higher valuations in the next round to pay for it, and the cycle keeps accelerating. Anthropic is the most extreme example of this model.

Once this cycle starts, it’s hard to stop. It can push a company up to the cloud, but it can also drag it into the abyss in an instant.

At the start of 2026, Anthropic’s valuation soars to $380 billion

A few days after Anthropic CEO Dario Amodei completed the previous $30 billion fundraising round, he told Fortune that if progress in AI is delayed by 12 months, Anthropic would go bankrupt.

For a company valued at $900 billion, the distance between “extraordinary success” and “operational bankruptcy” may only be a few bad quarters.

This sense of tense balance may explain why sensitive early investors largely did not participate in this round.

Early investors collectively watch from the sidelines

According to Forbes, some of Anthropic’s early backers—institutions that entered at a $4.1 billion valuation in 2023 or at a $61.5 billion valuation in March 2025—almost all had no intention of participating in this round.

The reason is simple: bankers privately expect that if Anthropic goes public as early as October 2026, its public-market valuation could land between $400 billion and $500 billion. This means that if someone bought into the last private round at a $900 billion valuation, then theoretically, before shares are unlocked and can be traded, that investment would already be losing on paper.

This kind of upside-down situation—late-stage private valuations far higher than the expected IPO valuation—is extremely rare in the history of tech financing.

It’s like a signal: either the company is severely overvalued in the private market, or the public market will assign a radically different price. Either way, uncertainty looms.

And the impending decisive event is the IPO itself.

We also previously mentioned the key people behind Anthropic’s IPO and financing—Kris h n a Rao, the company’s finance chief.

The Information reported that at the time, Anthropic’s lifeline for computing was basically tied to Google. Rao believed this was not acceptable and pushed for a new strategy: the computing infrastructure providers must be diversified.

According to The Information, citing people familiar with the matter, Rao had in-depth discussions with one of the investors, Byron Ditter of Bessemer Venture Partners, about this strategy. Afterward, Ditter commented that it was Rao who made the company realize that working with more partners would allow it to develop faster.

Looking back, Anthropic moved faster than OpenAI. They have already signed deep agreements with three cloud computing giants: Amazon, Google, and Microsoft. At the chip level as well, they simultaneously brought in NVIDIA GPUs, Google’s in-house TPU, and Amazon’s chips—forming a diversified supplier network.

But signing agreements is not enough; the core is ensuring that suppliers can truly deliver computing resources. At the end of 2025, Rao led two major deals: one was a lavish $30 billion using Microsoft’s cloud servers to run NVIDIA chips; the other was reserving up to 1 million Google TPUs.

By early April 2026, Anthropic went a step further, reaching new agreements with Broadcom and Google to lock in several gigawatt-scale data center power supply capabilities. These actions are no longer simply about “buying” computing power—they are about “pre-ordering” future infrastructure at scale.

Since Rao joined, the total amount of multiple fundraising rounds he helped the company complete has reached $60 billion. By this January, the company’s valuation had risen to $380 billion.

It can be said that under Rao’s strong push, Anthropic’s computing infrastructure and funding firepower have reached unprecedented scale.

Is there a bubble? The answer will be revealed in six months

At the current pace, if this round of financing goes through as planned, Anthropic is expected to seek an IPO between October 2026 and the first half of 2027. Reports say Goldman Sachs, JPMorgan Chase, and Morgan Stanley are discussing this.

At that time, what the market focuses on will no longer be whether “Anthropic can continue to grow,” but will instead become a referendum on the valuation logic for the entire AI industry: over the past three years, has the way private markets priced AI been correct?

Commitments to capital expenditure by ultra-large enterprises, multi-year compute reservation contracts, 40% gross margins, the dispute between total revenue and net income accounting, and the accelerating “valuation—compute—re-valuation” loop—these complex issues that can often be blurred in private markets will be put under the public market’s microscope at the time of the IPO.

If the public market is willing to give Anthropic a valuation of $1 trillion or even higher, then the entry price at a $900 billion valuation will look like a generous early layout. But if the market only gives it $500 billion, then the situation for the final batch of private investors will be very awkward.

And a third possibility—perhaps one with even more far-reaching impact—is that Anthropic’s IPO will serve as a key data point to validate or disprove the structural assumptions behind AI finance.

Everyone will remember Michael Burry, the prototype of the protagonist in The Big Short. He recently again called for “dot-com stocks” and “chip stocks” bubble in his paid column. Once Anthropic’s IPO discredits the assumptions underlying AI finance, it will be the moment the bubble bursts.

So whether for Anthropic itself, or for the entire AI industry that has gotten used to valuations surging upward nonstop over the past three years, stress testing is only just beginning—and it will soon be revealed by a single stock price chart curve, delivering the most honest and ruthless pricing.