Anthropic CEO Dario Amodei.

On September 29 Beijing time, foreign media disclosed Anthropic's IPO prospectus materials, positioning Anthropic to potentially become the first U.S. frontier AI lab to enter public markets.

From a $965 billion valuation during its May funding round to a listing expectation that may now exceed $2 trillion, investors are paying increasingly higher prices for Claude's future growth. The prospectus reveals that this expectation has its rationale, but also comes at a tremendous cost.

Revenue surged from $386 million to $4.59 billion, a 1,088% increase. Operating losses widened from $2.98 billion to $8.06 billion. Compute and infrastructure spending reached $7.33 billion, accounting for 58% of total operating expenses. However, cash and short-term investments on the books stood at $20.28 billion. Of even greater concern is the $518 billion in cloud, compute, and infrastructure obligations. According to foreign media reports, this commitment spans multiple years. Anthropic is locking in massive compute capacity ahead of demand that has not yet fully materialized, and going public will allow more public investors to participate in this grand bet: how fast can Claude's revenue grow to support the costs the company has already committed to?

Over 10x Growth and $42 Billion Net Loss

What supports Anthropic's push toward a high valuation is, first and foremost, a steep revenue growth curve. In 2024, the company's full-year revenue was only $386 million; by 2025, revenue had reached $4.59 billion, a year-over-year increase of 1,088%, expanding to nearly 12 times the previous year. For a company that only launched its first Claude model in 2023, it has quickly moved past the early stage of product commercialization.

But compared to OpenAI, Anthropic's revenue scale in 2025 still shows a clear gap. According to foreign media reports, OpenAI's revenue that year was approximately $13 billion, about 2.8 times that of Anthropic.

Entering 2026, Anthropic's growth further accelerated. According to foreign media reports, as of the end of July, its annualized revenue run rate had exceeded $65 billion. Anthropic expects revenue to reach approximately $190 billion to $200 billion by 2028. At a $2 trillion valuation, the price-to-sales ratio against 2025 revenue is approximately 436 times; using 2028 projected revenue, this multiple drops to about 10 times.

From 2024 to 2025, Anthropic's revenue grew more than 10 times.

The enormous gap between these two figures almost encapsulates the core bet of this IPO: over the next three years, the company's revenue needs to expand to more than 40 times its 2025 level to justify this valuation logic.

The investment supporting growth is "heavy." In 2025, Anthropic's spending on compute and infrastructure reached $7.33 billion, nearly triple the previous year, accounting for 58% of total operating expenses and equivalent to 1.6 times full-year revenue. Even setting aside other expenses, that year's revenue was insufficient to cover this single expenditure item—"for every $1 of revenue, $1.60 was spent on compute."

Anthropic's AI compute spending exceeded $7.3 billion in fiscal year 2025.

However, looking at the trend, revenue is catching up to compute spending more quickly. Based on disclosed data, in 2024, compute and infrastructure spending was approximately 6.6 times revenue, dropping to 1.6 times in 2025. The absolute amount of operating losses still widened from $2.98 billion to $8.06 billion, but the ratio relative to revenue has declined significantly. Scale growth is improving cost coverage, though it has not yet filled the loss gap.

The approximately $42 billion net loss may be the most eye-catching and easily misunderstood figure. Anthropic's net loss in 2025 was nearly $42 billion, of which approximately $34 billion came from non-cash accounting charges arising from changes in the valuation of financing instruments: these instruments, which may convert into Anthropic shares in the future, need to be remeasured on the books as valuations rise, with corresponding changes recorded as losses, without generating equivalent current cash outflows. The $8.06 billion operating loss is another financial metric that cannot simply be equated with cash burn; deducting one accounting charge from net loss is not sufficient to fully reconstruct cash flow.

Similar accounting effects also appeared at OpenAI. The Financial Times reported in June that OpenAI's net loss in 2025 was approximately $39 billion, including approximately $30 billion in one-time accounting charges related to investor rights under its old structure. The two companies have different specific financing arrangements, but both demonstrate that looking only at tens of billions of dollars in net losses can easily overestimate the actual cash burned.

But this does not mean funding pressure has eased.

The $518 Billion Compute Bill

$518 billion is another figure in the prospectus that is easily misunderstood. The prospectus discloses that Anthropic has approximately $518 billion in cloud, compute, and infrastructure obligations. This figure was once interpreted as the compute budget for the coming year, but that is not actually the case. The $518 billion represents the total value of contracts to be signed over many future years, many of which are multi-year contracts, not spending for a single year.

According to the disclosed contract allocation, $100 billion went to Amazon AWS, $200 billion to Google Cloud, $80 billion to Lambda and Nscale, $50 billion to Fluidstack, $30 billion to Microsoft Azure, with the remaining smaller contracts distributed to more small cloud service providers. Anthropic is also deploying its own software infrastructure at WULF's colocation facilities. These contracts have varying time spans, with some potentially covering three to five years or even longer. The actual annual expenditure, when spread out, is far below the $518 billion total.

But even so, the scale remains staggering. Based on 2025's $7.33 billion in compute spending, if triple-digit year-over-year growth continues in the coming years, annual spending will quickly approach tens of billions of dollars.

The continued rise in compute spending reflects the capital-intensive nature of the frontier AI race. Training and running larger models requires more chips, more servers, more data centers, and more electricity. Anthropic's choice to deeply align with Amazon and Google is driven both by compute supply needs and by the fact that these two companies are also important investors.

Two Clients Support One-Quarter of Revenue

The prospectus also reveals a risk point: customer concentration. Anthropic's two largest direct customers each account for 12% of revenue, totaling 24%. In other words, just two customers contribute nearly one-quarter of revenue. For a company with $4.59 billion in revenue, this level of dependence is already very high.

More notably, Anthropic explicitly warns in its risk factors that many of its largest customers have not locked in long-term contracts and may reduce or stop spending. This means that if major customers cut AI budgets or switch to competitors' models, Anthropic's revenue could face significant volatility.

The other side of customer concentration is pressure from the competitive landscape. Anthropic's main competitors include OpenAI, Google, Meta, and SpaceXAI. Competition with OpenAI is the most direct, with the two vying across multiple dimensions including enterprise customers, talent, and Washington influence. OpenAI also confidentially filed for IPO in June, with an expected listing in early 2027.

In this competition, Anthropic's unique position lies in the fact that its two major strategic partners, Amazon and Google, are simultaneously its infrastructure suppliers and investors. Amazon has invested approximately $8 billion, and Google approximately $2 billion. Both companies provide Anthropic with the cloud resources needed to train and run Claude, while also being deeply bound at the equity level. This relationship is both an advantage and a potential source of conflicts of interest.

Beyond finances, Anthropic also faces safety, copyright, and policy risks. The company's own research found that AI models could potentially break code, assist fraud, or manipulate information in controlled tests. Amodei has called for the industry to slow the pace of releasing new capabilities, while the company continues to launch new models, with the contradiction between safety commitments and competitive pressure persisting.

Copyright disputes also remain unresolved. Regarding the use of training data, Anthropic has been sued by authors and music publishers, and the related litigation adds uncertainty to its data acquisition and use. The company also clashed with the U.S. government over restrictions on AI tool usage and was once listed by the Pentagon as a supply chain risk, though related measures were subsequently blocked by courts.

From $124 Million to $2 Trillion: Thirteen Rounds of Financing in Five Years

From founding to pushing toward an IPO valuation exceeding $2 trillion, Anthropic took just over five years. In May 2021, Anthropic, founded by the Amodei siblings and others, announced the completion of a $124 million Series A funding round. Subsequently, Amazon, Google, and multiple investment institutions entered, driving the company's valuation rapidly higher: $61.5 billion in March 2025, rising to $183 billion in September of the same year, and reaching $965 billion after completing a $65 billion funding round in May 2026. If this IPO achieves a valuation exceeding $2 trillion, it would more than double the most recent funding round.

Supporting this valuation expectation is Anthropic's judgment on AI's transformative potential. According to the prospectus, the company believes AI's impact on the global economy will exceed that of industrialization, electricity, and the internet.

For investors, this vision ultimately needs to be translated into more specific questions: can revenue grow from $4.59 billion in 2025 to the company's projected nearly $200 billion in 2028; can compute investment gradually convert into profits; and how much impact will customer concentration, safety, and regulatory risks have on growth.

According to foreign media reports, Anthropic may go public after the U.S. midterm elections in November, potentially becoming the first frontier AI lab to land on public markets and providing a valuation reference for peers. But the listing window also faces tests: AI and chip stocks have recently come under pressure, and SpaceX's post-listing stock performance has also prompted investors to re-examine the pricing of mega IPOs. Anthropic needs to continually prove that the speed at which revenue catches up to costs is sufficient to support an ever-rising valuation, and that high growth is sustainable with strong momentum. How long public markets are willing to wait for it will also be an important signal this IPO leaves for the entire AI industry.