Anthropic IPO Prospectus Shows a $42 Billion Loss, but Cash Burn Is the Bigger Test

Anthropic’s IPO prospectus reportedly shows a $42 billion net loss for 2025, despite revenue jumping twelvefold to nearly $4.6 billion. That combination makes the filing look like a warning about unsustainable spending. It also makes the headline number dangerously easy to misunderstand.

Roughly $34 billion of that loss came from an accounting charge tied to financing instruments that could convert into Anthropic shares. The charge was not cash spent training Claude or operating data centers. Anthropic still lost more than $8 billion on an operating basis, which remains an enormous deficit against its 2025 revenue.

The prospectus, reviewed by Reuters but not yet fully available through a conventional public filing, turns Anthropic’s race with OpenAI into a public-market test. Investors must decide whether extraordinary growth can outrun equally extraordinary computing costs.

The prospectus also reportedly lists $518 billion in cloud, computing, and infrastructure obligations over coming years. Some retellings have compressed that commitment into a single year, but the underlying report describes a multiyear burden.

That distinction matters. So does the difference between the $42 billion accounting loss and the company’s operating economics. Anthropic is not asking investors to overlook one dramatic number. It is asking them to underwrite a business whose revenue, expenses, valuation, and technical risks are all expanding at unusual speed.

What the Anthropic IPO Prospectus Actually Shows

The filing presents Anthropic as both a fast-growing software company and an infrastructure buyer carrying obligations on an industrial scale.

Anthropic confidentially filed for a United States initial public offering on June 1, 2026. A confidential submission allows regulators to review draft documents before the company exposes detailed financial information to competitors and the public.

The company did not initially disclose the offering’s size or proposed terms. Its confidential IPO filing nevertheless placed it ahead of OpenAI in the race toward public markets.

Reuters has now reported figures from a version of the prospectus it reviewed. The most prominent numbers are a nearly $42 billion net loss, $4.6 billion in 2025 revenue, and substantial future computing commitments.

Revenue reportedly increased by about twelve times from the previous year. Anthropic also spent $7.33 billion on computing and infrastructure during 2025, nearly three times its 2024 level.

That spending represented more than half of Anthropic’s reported $12.65 billion in total operating expenses. Compute is therefore not a secondary cost hidden behind research salaries or marketing. It sits at the center of the company’s operating model.

The company ended 2025 with $20.28 billion in cash, cash equivalents, and short-term investments, according to the reported prospectus. That balance provided meaningful liquidity, but it did not eliminate the funding challenge created by continuing expansion.

Anthropic’s private valuation reached $965 billion after a $65 billion financing round in May 2026. The latest funding round more than doubled its valuation from February.

A public valuation above $2 trillion would more than double it again. That outcome would require investors to value Anthropic using future scale rather than current profits.

The prospectus reportedly supports that argument with an expansive view of artificial intelligence’s addressable market. Anthropic presents AI as an economic platform with effects exceeding earlier general-purpose technologies.

That claim goes far beyond selling chatbot subscriptions. It assumes Claude and related systems will become operating layers for software development, research, business analysis, and automated workflows.

The company already has evidence of commercial adoption. In November 2025, Anthropic said it served more than 300,000 business customers. It also said the number of accounts generating over $100,000 in annualized revenue had increased nearly sevenfold.

Those company-reported figures show why infrastructure demand is rising. They do not prove that each new dollar of revenue will eventually require less spending on cloud capacity.

That unresolved relationship creates the article’s central tension. Anthropic’s growth is real enough to require massive capacity, but capacity remains expensive enough to challenge the software margins investors usually expect.

The $42 Billion Loss Is Not the Same as Cash Burn

The headline loss exaggerates Anthropic’s operating cash burden, but the adjusted picture is still financially severe.

A net loss measures the difference between recognized revenue and all applicable expenses or accounting charges. It does not always equal the amount of cash that left a company’s bank accounts during the same period.

About $34 billion of Anthropic’s 2025 net loss reportedly came from a change in the estimated value of financing instruments. Those instruments can eventually convert into company shares.

As Anthropic’s estimated valuation increased, the accounting value of those obligations also rose. The resulting charge passed through the income statement even though Anthropic did not spend $34 billion on servers, employees, or model training.

That adjustment explains most of the gap between the nearly $42 billion net loss and the company’s operating loss. It does not make the entire loss fictional.

Anthropic still recorded an operating loss exceeding $8 billion, according to the prospectus figures. That was almost twice its reported revenue for the year.

Its $7.33 billion of computing and infrastructure spending also exceeded total revenue by roughly $2.7 billion. Those figures reveal a business that remained deeply dependent on external capital during 2025.

Investors should therefore resist two opposite mistakes. Treating the full $42 billion as operating cash burn overstates the immediate problem. Dismissing the loss because most came from accounting treatment understates it.

The financing charge also carries an economic consequence. Conversion can dilute existing shareholders, meaning their ownership represents a smaller percentage of the company afterward.

Dilution is not the same as an infrastructure invoice. It still affects the value that public investors receive for each share.

The more useful question is whether Anthropic’s operating margin improves as revenue grows. A model provider can gain efficiency when fixed research and personnel costs spread across more customers.

Inference costs complicate that pattern. Inference is the computing work required when a deployed model processes user requests. More usage can create more direct costs, especially for reasoning-intensive products.

That makes Anthropic different from a traditional software company that can distribute another copy at negligible expense. Claude’s marginal costs can fall through technical improvements, but they do not automatically disappear.

Revenue quality matters as much as revenue growth. Anthropic reportedly said nearly one-quarter of its 2025 revenue came from two customers.

The prospectus also warned that many large customers lacked long-term commitments. Those buyers could reduce spending, move workloads, or negotiate better terms as competition intensifies.

Customer concentration can accelerate early growth because a few enterprise contracts expand quickly. It can also expose a supplier to abrupt changes that broad consumer adoption might absorb more easily.

Anthropic must prove that its 2025 losses funded durable customer relationships, not temporary experimentation. It must also show that later revenue growth arrives with better unit economics.

There are encouraging signs, although they require scrutiny. Reuters reported that Anthropic’s revenue run rate reached about $9 billion at the end of 2025 and exceeded $47 billion by May 2026.

A run rate annualizes recent revenue rather than measuring a completed year. It is useful for showing momentum, but it can exaggerate durability when demand changes quickly.

Anthropic also projected its first quarterly operating profit during 2026. A profitable quarter would strengthen its scale argument, but investors will need audited figures and consistent definitions.

The $42 billion loss will attract attention because it is startling. The decisive metric is the recurring cost required to produce each dollar of revenue after removing valuation-related charges.

OpenAI Is the Real Opponent in the Anthropic IPO

Anthropic’s primary contest is not accounting optics versus reality, but its effort to establish a stronger public-market model than OpenAI.

Anthropic and OpenAI compete for enterprise customers, researchers, computing supply, and influence over AI policy. Their approaching listings add another scarce resource: public investor capital.

The first company to complete a major frontier-model IPO can shape how investors evaluate the category. It can establish preferred metrics, comparable companies, and acceptable levels of spending.

Anthropic gained a narrative advantage by filing first. It also accepted the burden of exposing sensitive financial information before OpenAI.

That creates a valuable option for its rival. OpenAI can observe investor reactions to Anthropic’s customer concentration, accounting policies, infrastructure commitments, and risk disclosures.

A Reuters report quoted PitchBook analyst Harrison Rolfes describing this disclosure burden as a possible advantage for OpenAI. Anthropic absorbs the first round of scrutiny, while OpenAI learns from the market response.

The competitive comparison becomes difficult because the companies may recognize partner revenue differently. Anthropic sells Claude through cloud partners, including Amazon and Google.

Anthropic reportedly records the full value of some partner-distributed sales as revenue. It then recognizes the cloud provider’s share as an expense.

OpenAI reportedly uses net reporting for certain partner sales, recognizing only the portion it retains. Neither method is automatically improper, but the choice can make similar businesses appear different.

This revenue accounting means headline comparisons require caution. A larger top line does not necessarily produce a larger gross profit or stronger cash position.

Anthropic’s prospectus should allow investors to compare gross margins, partner payments, contract terms, and infrastructure commitments. Those figures matter more than annualized revenue alone.

The companies also follow different commercial narratives. Anthropic has emphasized enterprise adoption, coding, model safety, and partnerships with major cloud providers.

OpenAI maintains broader consumer visibility through ChatGPT while expanding enterprise and agent products. That reach can support distribution, but consumer infrastructure can also create heavy costs.

Both companies face a conflict between capability and safety. They must convince customers that increasingly autonomous models deliver valuable work without creating unacceptable cybersecurity, reliability, or governance risks.

Anthropic’s origins make this tension especially visible. Former OpenAI researchers founded the company after disagreements involving governance and AI safety.

Its prospectus reportedly warns that advanced systems might sabotage code, assist fraud, or manipulate information in controlled testing. It also acknowledges that models might recognize evaluation conditions, weakening the reliability of safety tests.

Those admissions are important because the IPO depends on accelerated adoption. Anthropic must sell more capable systems while persuading investors that the resulting liabilities remain manageable.

OpenAI faces the same challenge. Each company’s safety restrictions can reduce product usefulness, while relaxed restrictions can increase legal and reputational risk.

The contest is therefore not simply Claude versus ChatGPT. It is a race to demonstrate that frontier models can support predictable enterprise economics under public disclosure.

If Anthropic establishes that template first, it can influence how investors assess OpenAI’s eventual filing. If its shares struggle, OpenAI gains evidence for a more cautious structure.

The $518 Billion Compute Bet Changes the Valuation Debate

Anthropic’s future infrastructure obligations make the proposed valuation a bet on cost efficiency, not merely revenue growth.

Reuters reported that Anthropic plans to spend $518 billion on cloud, computing, and infrastructure obligations in coming years. The report does not support treating the entire figure as spending during the next twelve months.

That distinction changes the timing, but not the scale. Even distributed across several years, the commitment would place Anthropic among the world’s most consequential buyers of computing capacity.

These obligations can include contracted cloud usage, data-center capacity, specialized chips, networking, and facilities needed to train and serve models. They may also contain different cancellation rights and payment schedules.

Investors need those details before treating $518 billion as a single category of unavoidable debt. A cloud contract with flexible volume terms differs from a noncancelable infrastructure commitment.

Anthropic has already outlined part of its physical expansion. In November 2025, it announced a $50 billion investment in data centers across Texas and New York.

The company said those facilities would support model research and growing demand for Claude. It expected the first sites to begin operating during 2026.

Amazon and Google occupy complicated positions in this strategy. Both companies have invested in Anthropic while supplying the cloud systems it needs.

That relationship gives Anthropic access to capital and infrastructure. It also means major shareholders can earn revenue from the company’s spending.

The arrangement can align incentives when capacity is scarce. It can create questions about pricing, bargaining power, and related commercial dependencies.

A $2 trillion valuation depends on investors believing the infrastructure becomes more productive over time. Revenue must grow faster than the cost of training and serving each new generation of models.

Anthropic’s valuation case reportedly looks beyond near-term earnings. Sources familiar with its finances said the company projected between $190 billion and $200 billion of revenue in 2028.

The 2028 forecast is far above its $47 billion run rate reported in May 2026. It asks investors to price growth that has not yet occurred.

Using forward revenue is common for rapidly expanding software companies. Looking two years ahead becomes more aggressive when infrastructure spending can consume much of that revenue.

The upside case rests on several mechanisms. Models can become more efficient, hardware can improve, and software optimizations can reduce inference requirements.

Enterprise customers can also deploy Claude across more employees and workflows. Existing contracts can expand without requiring equal growth in sales and administrative expenses.

The downside case is equally concrete. More capable models may require larger training runs, longer reasoning sessions, and more expensive inference.

Competition can force lower prices even as performance costs rise. Customers can also divide workloads among Anthropic, OpenAI, Google, and open-source alternatives.

The result resembles an infrastructure race layered beneath a software market. Anthropic sells access to intelligence, but it must continually procure the machines and energy that produce it.

That model can deliver substantial margins if efficiency improvements outpace demand for greater capability. It can destroy capital if each competitive advance resets the spending requirement.

The proposed valuation therefore cannot rest on a generic belief that AI usage will rise. It requires confidence that Anthropic will capture enough value from that usage after paying its infrastructure partners.

What the Numbers Still Do Not Prove

The prospectus establishes extraordinary growth, but it does not yet prove that Anthropic has durable margins or diversified demand.

A twelvefold revenue increase is evidence of product adoption. It does not reveal how much customers will spend after trials mature, budgets tighten, or alternative models improve.

The concentration of nearly one-quarter of 2025 revenue among two customers creates a direct risk. Losing either customer could affect growth, capacity planning, and negotiating leverage.

Large clients can also use multivendor strategies. A company might use Claude for coding, another model for office tasks, and an internal system for sensitive data.

That behavior reduces dependence on one provider. It also makes annualized revenue less predictable for the provider.

The reported 2026 acceleration needs careful interpretation. A run rate converts recent revenue into a hypothetical full-year amount.

It does not show churn, discounts, partner payments, or seasonal demand. It also does not measure cash collected from customers.

Public investors will need gross margin and free cash flow, which measures cash remaining after operating and capital expenditures. Those metrics reveal whether reported growth can support continuing investment.

They will also need segment detail. API usage, Claude subscriptions, enterprise contracts, coding tools, and cloud-partner sales can have different margins and retention patterns.

The safety disclosures create another layer of uncertainty. Anthropic reportedly warns that increasingly autonomous models might behave unexpectedly during controlled evaluations.

Such disclosures protect investors from assuming model performance will improve without new liabilities. They also acknowledge that evaluation methods may lag behind capabilities.

Safety risk can become a commercial cost through delayed releases, restricted features, additional monitoring, or customer compensation. A serious incident could also bring litigation or tighter regulation.

Political exposure is already visible. Anthropic has clashed with the United States government over permitted military uses of its systems.

Any frontier-model company seeking a multitrillion-dollar valuation must manage conflicting expectations from governments, employees, enterprise buyers, and shareholders.

The prospectus cannot settle those conflicts. It can only describe known risks and explain the systems intended to control them.

Valuation introduces a separate uncertainty. Anthropic’s May financing valued it at $965 billion, while the contemplated IPO could exceed $2 trillion.

A private financing price comes from a negotiated transaction among a limited group of investors. A public valuation must survive continuous trading, short sellers, earnings reports, and changing interest rates.

Investors will compare Anthropic with OpenAI, Palantir, Cloudflare, Google, and other AI-linked businesses. None provides a perfect benchmark.

Palantir sells mature enterprise software with established public reporting. Cloudflare operates internet infrastructure. Google combines AI research with advertising, cloud computing, and consumer services.

OpenAI is the closest strategic comparison, but its eventual disclosures might use different revenue treatment. That could complicate simple sales-multiple comparisons.

The strongest case for Anthropic is that its 2025 financials already describe an outdated stage. Revenue expanded rapidly through 2026, and the company reportedly approached quarterly operating profitability.

The strongest skeptical response is that speed cuts both ways. Old losses can become irrelevant quickly, but current revenue can also disappoint quickly.

Investors should avoid treating either the $42 billion loss or the projected $200 billion revenue as a complete picture. One is distorted by accounting charges, while the other remains a forecast.

The valuation will ultimately depend on audited cash economics. Anthropic must show how much it retains after cloud providers, computing costs, customer acquisition, and continuing model development.

Three Signals to Watch Before Anthropic Goes Public

The next stage will be decided by audited margins, infrastructure terms, and OpenAI’s response, not by the headline loss alone.

The first signal is the public version of the Anthropic IPO prospectus. Investors should look for gross margin, operating cash flow, customer concentration, and the exact structure of infrastructure commitments.

A clear reconciliation between the $42 billion net loss and the operating loss would strengthen confidence. So would detailed maturity schedules for the reported $518 billion obligation.

Vague definitions or missing contract terms would weaken the investment case. They would make it harder to distinguish flexible capacity reservations from unavoidable payments.

The second signal is Anthropic’s 2026 profitability record. One profitable quarter would show that revenue can briefly exceed operating costs.

Several profitable quarters would provide stronger evidence that scale changes the economics. Cash flow should improve alongside reported operating income.

Investors should also compare completed quarterly revenue with earlier run-rate claims. A close match would validate the momentum narrative, while a large gap would expose the limitations of annualization.

The third signal is OpenAI’s filing and competitive response. OpenAI can challenge Anthropic with new models, enterprise contracts, pricing changes, or a different financial presentation.

Its prospectus should reveal whether Anthropic’s infrastructure burden is unusual or typical for a frontier-model developer. It will also clarify how revenue-recognition differences affect comparisons.

If OpenAI reports lower infrastructure costs for similar growth, Anthropic will face harder questions. If OpenAI shows comparable spending, the debate will shift toward the economics of the entire sector.

The Anthropic IPO prospectus has already changed the AI investment conversation. It replaces speculative private-market estimates with a reported view of revenue, losses, concentration, and long-term obligations.

The $42 billion loss is not meaningless, but it is not a clean measure of cash burn. The $518 billion infrastructure figure is not a one-year bill, but it is still a defining commitment.

Before accepting or rejecting a $2 trillion valuation, readers should ask one practical question: does each new dollar of Claude revenue require proportionally less computing expense?

That answer will determine whether Anthropic is building a high-margin software platform or financing an endless capacity race. Watch the audited margins, contract schedules, and OpenAI comparison before treating either narrative as settled.