Author｜Gong Xingchao

Editor｜Chen Xiaoran

Revenue soared nearly 12-fold in a single year, yet operating losses ballooned to over $8 billion.

As Anthropic prepares to go public, it places its rapid growth trajectory, massive investments in computing power, and AI safety risks under the scrutiny of the open market.

Revenue Growth Fails to Cover Operating Losses

On September 28, foreign media obtained a draft IPO prospectus from Anthropic.

Importantly, this draft was not voluntarily released by Anthropic. In June 2026, Anthropic—a public benefit corporation—filed a draft S-1 registration statement confidentially with the U.S. Securities and Exchange Commission (SEC) in anticipation of its initial public offering of common stock.

Under SEC regulations, Anthropic is required to publicly submit a revised S-1 at least 15 days prior to roadshows for institutional investors, at which point complete financial data will be disclosed. The version reviewed by foreign media was not yet publicly available, and Anthropic declined to comment.

Nevertheless, the document provides a clear picture of the AI company’s current operational status.

In 2025, Anthropic reported nearly $4.6 billion in revenue, marking a 12-fold increase from the previous year. However, its annual net loss approached $42 billion.

A company established just five years ago has successfully commercialized large-scale AI models into a billion-dollar business, yet profitability remains unproven.

Breaking down the $42 billion net loss, approximately $34 billion stemmed from non-cash accounting expenses related to the revaluation of convertible instruments tied to prior financing rounds, rather than actual operating costs incurred by Anthropic that year.

Thus, describing the entire net loss as "money burned" would be misleading. Even excluding this factor, operating losses reached $8.06 billion—significantly higher than the $2.98 billion loss recorded the previous year.

The majority of costs were allocated to computing power. In 2025, Anthropic spent $7.33 billion on computing and infrastructure, roughly triple the previous year’s expenditure, accounting for over half of its $12.65 billion in annual operating expenses.

When compared to annual revenue, for every $1 earned, computing and infrastructure spending exceeded $1.50. This metric does not reflect cost per model call but underscores that while revenue grew rapidly, the resource consumption required to sustain this growth accelerated just as quickly.

Greater scrutiny falls on future obligations. Anthropic plans to undertake contract commitments totaling approximately $518 billion over the next few years for cloud services, computing power, and infrastructure.

These are not immediate payments nor directly equivalent to next year’s capital expenditures. However, by signing long-term contracts, Anthropic has assumed substantial payment responsibilities to ensure sustained access to computing resources.

As of the end of 2025, Anthropic held $20.28 billion in cash, cash equivalents, and short-term investments, providing some financial cushion for business expansion.

Faced with multi-year contractual obligations, the market still demands more concrete answers: how payments will be structured, how much spending can be adjusted based on demand, and how much revenue and profit new computing power will ultimately generate.

Anthropic confidentially submitted its S-1 registration document to the U.S. Securities and Exchange Commission in June 2026. The number of shares issued, pricing, and timing remain undetermined, with the IPO contingent on regulatory approval and market conditions.

The potential valuation exceeding $2 trillion reflects current market enthusiasm, not a finalized offering price.

Computing Power Fuels Growth

Founded in 2021 by researchers who left OpenAI, Anthropic has gradually built a product ecosystem for individuals, developers, and enterprises around its Claude model. Individual users subscribe to Claude, developers access the model via APIs, and enterprises integrate tools like Claude Code into software development and daily workflows.

The enterprise market serves as a key growth driver. Once models are deployed in high-frequency scenarios such as programming, knowledge retrieval, and data analysis, usage volume can scale rapidly alongside business expansion.

Anthropic also delivers services to enterprises through partnerships with Amazon AWS, Google Cloud, and Microsoft Azure, enabling clients with existing cloud platforms to procure and deploy Claude more easily.

In May 2026, Anthropic completed a $65 billion Series H funding round, with proceeds earmarked for expanding computing power, advancing safety research, and accelerating product deployment. Post-investment valuation reached $965 billion. By then, its annualized revenue had surpassed $47 billion, exceeding $65 billion by July. Market projections for full-year 2026 annualized revenue exceed $100 billion, with Anthropic anticipating 2028 revenue to reach approximately $190–200 billion.

For this public offering, market discussions suggest a potential valuation exceeding $2 trillion, positioning it as a candidate for the largest global IPO in history.

Behind this valuation surge lies investor anticipation of surging enterprise demand for AI. However, annualized revenue extrapolates income at a specific moment and cannot replace audited full-year revenue, let alone directly infer annual profits.

For Anthropic, sustaining model capabilities as a driver of customer payments matters more than achieving rapid growth at a single point in time.

Customer concentration reveals additional pressure. In 2025, nearly a quarter of Anthropic’s revenue came from just two clients; many large customers lacked long-term locked contracts, risking reduced spending or halted procurement.

While major clients amplify revenue quickly, they also expose performance to individual budget fluctuations.

Competition further compresses Anthropic’s window to improve costs. Enterprises can test models from multiple vendors simultaneously, adjusting procurement based on performance, pricing, stability, and safety requirements.

Anthropic must continuously release more competitive versions of Claude to retain clients. However, new model training and deployment typically demand additional computing power.

Amazon and Google serve as both key investors and cloud infrastructure partners for Anthropic. This relationship facilitates rapid access to capital, computing resources, and sales channels but forces investors to scrutinize supply capacity, procurement costs, and partnership terms.

Whether computing power investments translate into stable customer renewals—rather than merely fueling an even costlier model race—will determine the viability of its business model.

The Cost of Safety

Equally noteworthy as financial figures are the risk disclosures in the prospectus.

According to foreign media, risk factors occupy roughly 80 pages of the ~261-page prospectus. Anthropic acknowledges that advanced AI may pose "catastrophic or existential risks," listing potential scenarios such as models resisting shutdown, hiding or manipulating information, or engaging in ransom-like behavior.

These disclosures do not imply that extreme outcomes have already occurred. Instead, Anthropic cautions investors that as model capabilities advance, certain behaviors may prove difficult to detect proactively.

If models can recognize evaluation contexts and adjust their performance accordingly, existing testing methods’ ability to assess safety becomes limited.

Safety serves as a key differentiator for Anthropic against competitors and a commitment it must continuously fulfill. The challenge lies in balancing safety research—which requires time, talent, and computing power—with revenue growth dependent on ever-improving product capabilities.

Anthropic admits that its growth and even revenue hinge on new models, with continuous releases necessary to maintain cutting-edge competitiveness. While emphasizing cautious advancement, the company must keep pace with industry iteration speeds—a tension difficult to resolve through safety declarations alone.

This makes Anthropic’s IPO distinct from typical software company listings. Investors must evaluate not only product pricing but also how much the company needs to invest to sustain model leadership, whether long-term contracts can support revenue, and whether safety incidents might alter product launches or client procurement plans.

The valuation itself pushes this uncertainty to extremes. Roughly calculating, Anthropic’s historical price-to-sales ratio exceeds 430x based on a $2 trillion valuation and 2025 revenue of ~$4.6 billion; using the $47+ billion annualized revenue disclosed in May 2026, the ratio drops to ~40x.

The vast discrepancy reflects market pricing in substantial future growth. However, whether annualized speeds materialize as full-year revenue and convert into profits requires validation through subsequent financial disclosures.

Anthropic has proven enterprises’ willingness to pay for Claude. After entering public markets, it must demonstrate customer retention, controlled computing costs, and that safety commitments withstand product iteration.

While a $2 trillion valuation bets on frontier AI’s future, it cannot replace quarterly operational results.