Anthropic’s IPO Filing Reveals the Staggering Economics of the AI Boom

Anthropic generated $16.3 billion in revenue during the first six months of 2026. Over the same period, the company behind Claude reported a net loss of $80.7 billion. And behind those extraordinary figures lies another uncomfortable detail: a relatively small number of customers still account for a meaningful share of its business.

Those numbers appeared in Anthropic’s public IPO prospectus, filed with the US Securities and Exchange Commission on October 26. The company intends to list on the Nasdaq under the ticker ANT, although its preliminary filing leaves the offering price and several other terms unresolved.

At first glance, the financial statements look almost absurd. A company that earned $4.6 billion throughout 2025 has already generated more than three times that amount in half a year, while apparently losing nearly five dollars for every dollar of revenue.

But the apparent contradiction largely disappears once the financial statements are read properly. Most of Anthropic’s reported net loss comes from accounting adjustments associated with financing instruments, not from money spent operating Claude. Its actual operating loss was dramatically smaller, and its cash flow tells a different story again.

The Anthropic IPO filing reveals something more consequential than a spectacular loss: generative AI is becoming a serious commercial business, but one whose economics depend on computing infrastructure, unusually rapid growth and financial commitments of an extraordinary scale.

Table of Contents

• The three numbers that define Anthropic’s IPO filing
• Claude’s revenue growth is almost difficult to believe
• What kind of revenue is Anthropic actually generating?
• The customer concentration problem
• Why Anthropic’s $80.7 billion loss is misleading
• The operating business is improving much faster than the headline suggests
• The real economic challenge: paying for intelligence
• Cash flow reveals another side of the business
• Anthropic’s complicated dependence on Big Tech
• Can Anthropic become sustainably profitable?
• What changes when Anthropic becomes a public company?

The three numbers that define Anthropic’s IPO filing

For years, the financial performance of leading generative AI developers has been reconstructed from funding announcements, private investor presentations, leaked projections and occasional statements from executives.

Anthropic’s public Form S-1 offers something substantially more useful: financial statements, accounting policies, risk disclosures and management commentary that can be examined together.

Three figures provide the starting point:

• $16.286 billion: revenue generated during the six months ended June 30, 2026.
• $80.748 billion: the company’s reported net loss over the same period.
• $72.388 billion: the negative fair-value adjustment associated with convertible notes, derivative liabilities and other financing instruments.

The third number is essential to understanding the second. Without it, readers could easily conclude that Anthropic consumed more than $80 billion operating its AI models in six months. That is not what the filing says.

Customer concentration adds another layer of risk. In 2025, two individual customers each represented approximately 12% of revenue. More recent concentration figures and the company’s growing dependence on enterprise distribution channels need to be assessed separately rather than assuming that the same two customers still account for exactly one-quarter of current sales.

The financial statements reveal both a business expanding at remarkable speed and a company whose commercial structure remains unusually concentrated.

Claude’s revenue growth is almost difficult to believe

Anthropic’s revenue progression illustrates how quickly the market for commercial AI services has developed.

These figures come from Anthropic’s consolidated statements of operations. The interim periods are unaudited.

Comparing the first half of 2026 with the equivalent period in 2025 produces revenue growth of approximately 1,300%, or fourteen times the previous year’s level.

That is the meaningful year-over-year comparison. Simply comparing six months of 2026 with the whole of 2025 understates the importance of using equivalent reporting periods.

The quarterly figures are even more revealing. Anthropic generated approximately $4.73 billion in the first quarter of 2026 and $11.56 billion in the second.

In other words, revenue more than doubled in a single quarter.

For an established software company, such acceleration would be exceptional. For a business selling access to large language models, it suggests that enterprise AI demand is moving well beyond experimentation.

Businesses are paying for Claude to perform work, not merely testing whether chatbots can produce convincing answers.

But there is an important distinction between revenue growth and durable revenue growth. The former is visible in the financial statements. The latter will require several more reporting periods to establish.

What kind of revenue is Anthropic actually generating?

Not every dollar of revenue has the same economic characteristics.

Anthropic earns money through a combination of subscriptions and consumption-based services. The latter include customers paying for access to its models according to usage, such as API requests and the computational work performed.

During the first half of 2026, consumption-based revenue reached approximately $13.06 billion, while subscription revenue totaled around $3.22 billion.

That means roughly 80% of the company’s revenue came from consumption rather than subscriptions.

This matters because consumption-based pricing can scale rapidly as customers increase their use of AI. A company deploying Claude across thousands of employees or integrating it into automated software workflows may generate substantially more revenue than a business purchasing a fixed number of individual subscriptions.

The disadvantage is that usage can also change quickly.

Customers may optimize their applications to use fewer tokens, switch to cheaper models, negotiate better pricing or move workloads to competing providers. Unlike a traditional software subscription, a consumption-based customer does not necessarily guarantee the same revenue next quarter.

Anthropic’s business therefore combines characteristics of enterprise software with those of computing infrastructure.

It sells intelligence as a service, but that intelligence requires substantial computing resources every time it is used.

The company also recognizes revenue from transactions distributed through cloud marketplaces. Under its accounting approach, Anthropic generally records the full customer transaction as revenue when it acts as the principal supplier, while the cloud provider’s share is recorded separately as an expense.

That treatment is important when comparing Anthropic with competitors that may structure or report similar commercial arrangements differently. Two companies can have comparable customer demand while presenting different revenue figures because of their contractual relationships and accounting policies.

Revenue growth is real under the company’s disclosed accounting framework. Whether that growth translates into comparable economic value across different AI providers is a more complicated question.

The customer concentration problem

Anthropic’s growth is impressive, but its customer base introduces risks that are familiar to investors in enterprise software and infrastructure businesses.

In 2025, two customers each represented approximately 12% of revenue. Together, they accounted for roughly one-quarter of the company’s sales.

For a business generating billions of dollars, that degree of concentration is significant.

A large customer can reduce spending because of budget changes, new internal technology, competitive pricing or a strategic decision to use a different AI model.

The risk is particularly important because some major AI customers are sophisticated technology companies capable of developing alternatives, operating multiple model providers or building their own systems.

Customer concentration should also be distinguished from distribution concentration.

Anthropic relies heavily on cloud marketplaces operated by Amazon Web Services and Google Cloud, among others. Approximately 47% of its 2025 revenue was sold through Amazon and Google cloud channels, according to the prospectus information previously reported by Reuters. The proportion reached approximately 48% in the first half of 2026.

These cloud companies are not necessarily the ultimate customers consuming every dollar of that revenue. They are distribution partners, billing intermediaries, infrastructure providers and, in some cases, investors.

That distinction matters. Losing an individual enterprise customer is one risk. Losing favorable access to a major cloud distribution channel is another, potentially much broader one.

Anthropic’s expanding enterprise customer base offers some protection against concentration. The company reported more than 11,900 customers generating at least $100,000 in trailing annual revenue as of September 30, 2026.

That is evidence of considerable commercial reach. It does not eliminate the importance of the largest accounts or guarantee that spending levels will remain stable.

Why Anthropic’s $80.7 billion loss is misleading

The most sensational figure in the prospectus is also the easiest to misunderstand.

Anthropic reported a net loss of $80.748 billion for the first six months of 2026. Taken in isolation, that number suggests a company operating on an almost unimaginable financial deficit.

But the income statement contains several layers, and the distinction between operating results and financing-related accounting effects is crucial.

Anthropic’s operating loss was $693 million.

Below that operating result, the company recorded a $72.388 billion negative adjustment related to changes in the fair value of convertible notes, derivative liabilities and simple agreements for future equity, commonly known as SAFEs.

These instruments can change in accounting value as the company’s estimated valuation changes. When certain liabilities become more valuable to their holders, the issuer may have to recognize an accounting expense.

That expense does not mean the company transferred an equivalent amount of cash to creditors during the reporting period.

Anthropic also reported approximately $8.76 billion in other net non-operating expenses, partly offset by interest income of $1.11 billion.

These items help explain why the reported net loss is vastly larger than the operating deficit.

The distinction can be summarized as follows:

The fair-value adjustment is not an ordinary operating expense. Nor is it necessarily economically meaningless: financing instruments can create dilution, settlement obligations and consequences for existing shareholders.

But describing the entire $80.7 billion as money Anthropic spent developing AI would be incorrect.

The company did not lose $80.7 billion because running Claude cost $80.7 billion more than customers paid for it.

Its operating business lost less than $1 billion over the period. The much larger reported net loss reflects the accounting consequences of financing an exceptionally valuable private company.

The operating business is improving much faster than the headline suggests

Removing the distortion created by financing-related charges reveals a remarkable change in Anthropic’s operating performance.

During the first half of 2025, the company generated $1.163 billion in revenue and recorded a $2.135 billion operating loss.

A year later, revenue had reached $16.286 billion while the operating loss had fallen to $693 million.

Its operating margin improved from approximately negative 184% to negative 4%.

The second quarter of 2026 was particularly important. Anthropic reported approximately $974 million in operating income on revenue of $11.555 billion.

That represents an operating margin of roughly 8.4% for the quarter.

For a company whose earlier financial statements showed computing and development expenses vastly exceeding revenue, this is a meaningful milestone.

Anthropic also reported approximately $1.618 billion in adjusted operating income for the first half of 2026. That non-GAAP measure excludes certain expenses, including stock-based compensation, and should not be confused with statutory operating profit.

The improvement suggests that revenue is growing faster than some major categories of operating expense.

However, one profitable quarter does not establish a permanently profitable business model. Anthropic’s costs may rise again as it deploys new models, expands infrastructure capacity or competes more aggressively on pricing.

The company is not operating a mature software business with predictable margins. It is developing an industry in which the cost of supplying the product can change rapidly alongside its capabilities.

The real economic challenge: paying for intelligence

The most revealing operating expense in Anthropic’s prospectus is compute and infrastructure.

During the first half of 2026, Anthropic recorded approximately $8.945 billion in computing and infrastructure expenses, compared with $2.334 billion during the same period in 2025.

That is an enormous increase in absolute terms. Yet relative to revenue, the picture has improved dramatically.

Compute and infrastructure expenses represented approximately 201% of revenue in the first half of 2025. In the first half of 2026, that proportion had fallen to roughly 55%.

In simple terms, Anthropic went from spending about two dollars on compute and infrastructure for every dollar of revenue to spending around 55 cents.

This does not mean that each individual Claude request now costs 55% of its selling price. The expense category includes different infrastructure activities, and the company’s financial statements do not provide a complete model-by-model breakdown of training and inference costs.

Training involves the enormous computational work required to develop a model. Inference is the computation performed when users and applications interact with an already trained model.

Both matter economically, but they behave differently.

Training expenditure can be concentrated around the development of new model generations, while inference costs tend to grow with usage. Improvements in hardware, model architecture and software optimization can reduce the cost of serving a given workload.

At the same time, more sophisticated reasoning models and increasingly autonomous agents can consume far more computation per task.

Anthropic therefore faces two opposing forces: efficiency improvements that reduce the cost of intelligence and rising customer expectations that encourage more intensive computation.

Its prospectus shows that the company has recently been winning that economic race. It does not establish that the advantage will persist indefinitely.

Cash flow reveals another side of the business

Accounting profit and cash generation are related, but they are not interchangeable.

That distinction is especially important for Anthropic because its net loss is heavily affected by noncash financing adjustments.

During the first half of 2026, Anthropic reported approximately $3.929 billion in positive operating cash flow.

This is a major improvement from the $1.513 billion in operating cash outflow recorded during the comparable period of 2025.

Positive operating cash flow indicates that, after the adjustments and working-capital movements required by cash-flow accounting, the company’s operating activities generated cash during the period.

It does not mean Anthropic’s overall cash balance increased by the same amount.

The company also recorded approximately $17.1 billion in cash used for investing activities during the first half of 2026. Investing cash flows can include purchases of investments as well as expenditures on long-lived assets, so the entire amount should not automatically be described as spending on AI data centers.

Likewise, the $8.945 billion compute and infrastructure expense in the income statement is not identical to cash paid for infrastructure during the period.

These distinctions are essential when assessing capital-intensive businesses.

At June 30, Anthropic held approximately $112.4 billion in cash and cash equivalents and another $19.3 billion in short-term investments.

That represents a substantial liquidity position, although it must be considered alongside the company’s financing structure and future infrastructure obligations.

Anthropic has also entered into computing agreements involving hundreds of billions of dollars in potential spending over multiple years. Those commitments are not equivalent to immediate cash expenses, and agreements with fixed minimum payments must be distinguished from optional capacity purchases or spending contingent on delivery.

The central financial question is whether Anthropic can generate enough cash from customers to support its future computing requirements without repeatedly depending on new external financing.

Anthropic’s complicated dependence on Big Tech

Anthropic’s financial statements also reveal a structural feature of the generative AI industry: some of the companies financing the AI revolution are simultaneously its infrastructure suppliers, distribution partners and competitors.

Amazon and Google have been central to Anthropic’s expansion. Their cloud platforms provide access to computing capacity and help distribute Claude to enterprise customers.

Microsoft has also become part of Anthropic’s cloud infrastructure ecosystem.

These relationships provide enormous advantages. Anthropic can reach enterprise customers through established cloud marketplaces and secure access to computing resources that would be difficult to reproduce independently.

But the arrangements create dependencies.

Cloud providers have their own AI strategies. Google develops Gemini, Microsoft has extensive AI investments and products, and Amazon operates its own AI services and infrastructure.

Anthropic therefore relies on partners whose commercial interests do not always align perfectly with its own.

Its prospectus acknowledges the risks created by concentrated infrastructure supply, the availability of computing resources and the complexity of these relationships.

There is also a financial consequence: cloud partners may participate in both sides of the industry’s investment cycle, supplying infrastructure while investing in or distributing the companies that consume it.

This does not automatically make the resulting revenue artificial. Customers can purchase valuable services through these arrangements, and the associated contracts create real commercial obligations.

But it complicates any attempt to assess the industry’s economics by looking at revenue growth alone.

Some of the money flowing into AI developers eventually returns to the same technology ecosystem through computing purchases, cloud services and infrastructure agreements.

Understanding who ultimately pays, who earns a margin and who carries the investment risk is essential.

Can Anthropic become sustainably profitable?

The IPO prospectus provides evidence that Anthropic’s operating economics are improving. It does not settle the question of long-term profitability.

Several encouraging developments are visible.

Revenue is growing extraordinarily quickly. Enterprise adoption appears increasingly substantial. Compute expenses have fallen sharply relative to revenue. The company generated positive operating cash flow during the first half of 2026 and reported an operating profit in the second quarter.

Those are meaningful achievements for a business that was still generating less than $400 million in annual revenue in 2024.

But several unresolved questions remain.

Will customers continue increasing their AI spending at anything approaching the current pace? Will competition force Anthropic to lower prices faster than it can reduce inference costs? Can it maintain a technological advantage without continually increasing research and training expenditure?

And what happens when the next generation of models requires a much larger infrastructure investment?

There is also the question of revenue durability.

Traditional enterprise software businesses often benefit from subscriptions, deeply embedded workflows and switching costs. Anthropic is trying to develop similar advantages, but AI customers may have more opportunities to switch models, use multiple providers or build internal alternatives.

The company’s commercial future will depend partly on whether Claude becomes sufficiently integrated into business operations that replacing it becomes costly or inconvenient.

Its growth already demonstrates that companies are willing to pay substantial sums for AI capabilities. The remaining challenge is proving that those payments can support the full cost of continuously improving and delivering the technology.

What changes when Anthropic becomes a public company?

Anthropic’s proposed Nasdaq listing could mark a significant change in how the financial performance of leading AI developers is evaluated.

Private companies can disclose selected metrics while keeping much of their financial information confidential. Public companies face regular reporting requirements, securities-law obligations and greater scrutiny from investors, analysts and regulators.

Assuming the IPO is completed, Anthropic would have to provide ongoing financial disclosures that make it easier to track revenue growth, profitability, cash generation and changes in operating expenses.

That transparency could prove particularly valuable for the generative AI industry, where headline figures such as annualized revenue run rates and infrastructure commitments are frequently discussed without complete financial statements.

Public reporting would allow investors to examine whether operating margins continue improving, whether customer concentration declines and how infrastructure spending evolves.

It would also create pressure to explain changes in strategy and financial performance more consistently.

However, an IPO would not automatically eliminate the company’s unusual governance structure.

Anthropic is incorporated as a Delaware public benefit corporation, reflecting its stated commitment to AI safety and responsible development.

Its proposed structure includes special voting arrangements intended to preserve substantial founder influence and a Long-Term Benefit Trust with important governance powers.

According to the prospectus, a special class of founder-controlled stock is designed to carry 50.1% of voting power on matters generally submitted to shareholders. The trust also has authority over the election of a majority of the board’s authorized directors.

These arrangements mean that public shareholders may have less influence over corporate decisions than they would at a company with a conventional one-share, one-vote structure.

They also raise an important question about how Anthropic will balance its safety commitments, commercial ambitions and obligations to public investors.

The preliminary prospectus does not yet establish the final IPO price or the company’s eventual public-market valuation. Those figures should not be treated as settled merely because investors and media reports are discussing possible trillion-dollar outcomes.

What Anthropic’s numbers really tell us about the AI boom

Anthropic’s filing is important because it replaces speculation with a more detailed financial picture.

The company is no longer merely a promising AI laboratory supported by venture capital. It has become a multibillion-dollar enterprise technology supplier with substantial customer demand and a rapidly expanding commercial operation.

At the same time, its financial structure remains unlike that of a mature software company.

Its models require immense computing resources. Its infrastructure relationships create dependencies on some of the world’s largest technology companies. Its revenue is growing at a pace that makes conventional forecasting difficult, and its reported net income is heavily distorted by the accounting treatment of financing instruments.

These are not minor complications. They are central to understanding how the generative AI industry works.

For investors, the relevant questions extend well beyond whether Claude is popular or whether Anthropic can maintain its technological position against OpenAI, Google and other competitors.

The more revealing questions concern the economics of each additional dollar of AI revenue: how much computation it requires, how much of the resulting income is retained, how reliably customers continue spending and how much capital must be committed to sustain growth.

For users of Claude and other AI services, the prospectus also offers a reminder that the apparently effortless experience of interacting with a language model rests on an extraordinarily expensive industrial system.

Every response is produced by computing infrastructure that must be financed, operated, upgraded and eventually replaced.

Anthropic’s $80.7 billion net loss will attract the headlines. Its $16.3 billion in six-month revenue deserves equal attention. But the most consequential development may be the rapid improvement in the relationship between revenue and operating costs.

Anthropic has demonstrated that generative AI can generate enormous commercial demand. Its next challenge is proving that the business can finance the enormous infrastructure required to satisfy it, year after year.

The IPO filing doesn’t resolve that question. For the first time, it gives the public a substantially better set of numbers with which to ask it.

Sources

• Anthropic, PBC — Form S-1 Registration Statement, filed October 26, 2026 (primary source)
• SEC EDGAR — Anthropic IPO filing and associated exhibits
• Reuters — Anthropic’s Soaring Sales, Steep Losses Put AI Frenzy to IPO Test, October 26, 2026
• Reuters — Anthropic’s First-Half Loss and Infrastructure Commitments, October 20, 2026
• TechCrunch — Anthropic Officially Files for Its Public Debut, October 26, 2026
• TechCrunch — Anthropic’s IPO Filing Reveals Its Enterprise Customer Strategy, October 26, 2026