Anthropic, which rose to prominence with the Claude series of large models, has for the first time laid bare the financial underpinnings of the AI arms race to the public.
At 9:00 a.m. Eastern Time on October 26, the five-year-old company publicly filed its S‑1 registration statement with the U.S. Securities and Exchange Commission, reporting revenue of $16.286 billion in the first half of 2026, up from $1.163 billion in the same period last year. At the same time, it posted a net loss of $80.748 billion, the vast majority of which—$72.39 billion—was attributable to non-cash accounting expenses arising from the increase in fair value of convertible notes and other financing instruments issued to raise capital for expansion.
The roadshow and bookbuilding are expected to kick off within a few weeks, with the IPO potentially taking place as early as November. At that time, this fastest-growing AI company will face scrutiny of its roughly $2 trillion valuation, bringing with it a scale of capital consumption unseen on Wall Street.
The central question raised by this document is not whether Anthropic is growing—indeed, it is growing at a pace that has surprised even seasoned tech investors—but rather whether the market is willing to value such growth at $2 trillion: a growth trajectory that hinges on commitments of hundreds of billions of dollars in future compute capacity, partnerships with three hyperscale cloud providers that increasingly view it as a competitor, and two major customers whose combined spending accounts for roughly one-fifth of its revenue.
The answer to this question will set the benchmark for the IPO pricing of every AI company going forward. Meanwhile, competitors are already racing ahead: according to media reports, OpenAI is striving to close a funding round valued at over $1.4 trillion before Anthropic sets its price—whichever company prices first will establish the initial benchmark.
Unforeseen Income Statement
With a net loss of $80.7 billion over six months, the company appears on paper to be recklessly burning through capital; however, about $72.4 billion—roughly 89 cents for every dollar of loss—is an accounting effect rather than an actual cash outflow.
The bulk of these more than 70 billion yuan comes from an accounting line item that does not involve any cash outflows. Earlier, Anthropic issued a large number of convertible notes; under the terms of these notes, they can be converted into shares at a future date, with their value increasing as the company’s valuation rises. Accounting standards require that whenever the valuation is revised upward, the company must recognize this increase in value as an expense in the current period.
Meanwhile, according to media reports, in its Series H funding round mid-year, Anthropic raised $65 billion at a post-money valuation of approximately $965 billion. In other words, the more valuable the company becomes, the greater its reported losses appear on paper. That $80.7 billion figure is more like a shadow cast by the soaring valuation on the financial statements, rather than an actual outflow of cash. For the same period last year, this number stood at $12.97 billion.
Breaking down this table, the true picture at the operational level is less severe than the reported losses, but the scale remains substantial.
In the first half of 2026, Anthropic’s spending on computing power and infrastructure reached $8.945 billion, surpassing the full-year total of $7.326 billion in 2025 and nearly four times the $2.334 billion spent in the first half of 2025. Compared to its revenue of $16.286 billion, before any R&D, sales, or general and administrative expenses were incurred, the computing‑power bill already accounted for roughly 55% of sales.
Meanwhile, sales, marketing, and partnership expenses surged from $120 million in the same period last year to $3.168 billion, a growth driven by acquisitions. The operating loss itself narrowed from $2.135 billion to $693 million; when non-cash items are excluded according to the company’s reporting framework, adjusted operating profit turned from a loss in the prior-year period to approximately $1.6 billion in profit—this is one of the few bright spots in the entire prospectus.
Growth-oriented investors are willing to accept losses, betting on how much longer revenue can continue to grow at a rapid pace—and that’s precisely what makes it intriguing. According to media reports, the company disclosed to potential investors that third-quarter 2026 revenue would increase 45% quarter-over-quarter, reaching approximately $16.8 billion, meaning a single quarter’s performance would nearly match the entire first half of the year. However, the 45% quarter-over-quarter growth rate itself is already decelerating; in the second quarter, revenue more than doubled compared to the first quarter, climbing to roughly $11.6 billion.
Some investors who reviewed the disclosure documents have shifted to considering a lower valuation; at the company’s Investor Day event, attendees discussed a valuation close to $1.5 trillion, a significant drop from the roughly $2 trillion previously expected for the November stock sale.
Hundreds of billions of dollars in computing power bets
If the income statement is shocking, then the commitments on the balance sheet are where the real story lies.
S-1 disclosure indicates that, since the end of 2025, Anthropic’s newly signed and amended leasing and commercial agreements entail approximately $417 billion in future minimum payment obligations, primarily for computing equipment and data center leases, as well as procurement commitments to specific cloud service providers. These obligations are set to take effect between 2026 and 2027, with terms ranging from less than one year to 20 years. Additionally, some agreements with provisions yet to be finalized are expected to involve roughly $125 billion to $150 billion.
The two amounts total approximately US$540 billion to US$570 billion.
According to reports, the market has summarized this commitment network as “up to $750 billion in computing power investment,” of which approximately $450 billion has already been contracted, with most terms being irrevocable. Breakdowns include roughly $111 billion with Google, about $110 billion with Amazon Web Services, and around $31 billion with Microsoft.
These are not optional purchases. Most of the contracted portion must be paid regardless of usage, which means Anthropic has effectively turned future revenue into fixed obligations even before that revenue is secured. The company is betting that demand for Claude—especially its coding assistant, Claude Code, which has become a developer favorite—will grow fast enough to justify these GPU expenses. If demand falters, the commitments will not be scaled back; these bills will be paid regardless of revenue, effectively granting them priority claim on every dollar earned.
To finance this expansion, Anthropic has become increasingly reliant on debt. According to media reports, in October 2026, a large credit facility led by BlackRock closed, with the company offering lenders a yield of over 10% on chip‑financing debt. This interest rate indicates that well‑capitalized private equity remains interested in AI infrastructure, but at a substantial risk premium.
This deal follows Google’s April announcement of an investment plan totaling up to $40 billion, as well as Amazon’s additional $5 billion commitment. In its S‑1 filing, the company stated that infrastructure partnerships provide it with the computing power needed to keep pace. This remark both conveys confidence in expansion and highlights a passive position: capacity must be procured well ahead of demand.
Partners who view it as a competitor
This filing is unusually candid in this regard: the risk factors section states that Anthropic relies on a limited number of cloud service providers—Amazon, Google, and Microsoft—as well as suppliers such as Broadcom—to obtain computing power and critical components. These suppliers maintain strategic and investment relationships with the company and are also its customers, which could give rise to complex conflicts of interest that may affect Anthropic’s ability to secure computing resources.
This is precisely the structural contradiction at the heart of the AI boom: companies that sell the chips and cloud capacity needed to train and run models to Anthropic are simultaneously racing to develop their own competing models. Google has Gemini, Amazon has embedded its own models throughout AWS, and Microsoft is deeply tied to OpenAI.
This interdependence is mutual, setting Anthropic apart from virtually every major IPO candidate in recent times. According to third-party estimates, Amazon holds 15% to 20% of the company, while Alphabet owns 10% to 15%. Based on an IPO valuation of $2 trillion, Amazon’s stake would be worth between $300 billion and $400 billion, and Alphabet’s holding would be valued at $200 billion to $300 billion.
According to Amazon’s own financial reports, Amazon alone is expected to generate $16.8 billion in pre-tax income from its stake in Anthropic in the first quarter of 2026. This windfall enables the hyperscale cloud provider to simultaneously assume the roles of landlord, shareholder, and competitor. An owner who both rents out space and competes with tenants has little incentive to maintain low rents or ample capacity, and Anthropic’s own S‑1 filing warning about constrained access to computing power suggests that it is well aware of this dynamic.
The concentration risk on the revenue side is also disclosed in the filing: in the first half of 2026, two major customers will collectively account for approximately 20% of revenue, and the S‑1 warns that significant reliance on a limited customer base may persist; many large customers have yet to sign long-term contracts and could reduce or discontinue purchases at any time based on their own needs. In 2025, about 47% of sales were generated through Amazon and Google Cloud Marketplace, and this share is expected to rise further to 48% in the first half of 2026, with the proportion attributable to third-party marketplaces continuing to increase.
The channels have indeed accelerated customer acquisition and deepened dependency, as evidenced by the composition of accounts receivable: the share of accounts receivable generated through CSP market channels has risen from 42% at the end of 2024 to 60% at the end of 2025.
AI trading is showing signs of instability.
During the week the documents were disclosed, the market has been closely testing the resilience of AI‑related leverage, from capital‑expenditure reviews across the industry to the private‑credit markets on which Anthropic itself relies. According to market data, as of Thursday, October 22, the Nasdaq Composite Index fell 2.58% in a single day—the largest one‑day drop since June—with chipmakers NVIDIA and AMD leading the decline, as investors reassess the valuations they’re paying for AI‑related assets. Notably, the capital markets that have been funding Anthropic’s multi‑hundred‑billion‑dollar computing‑power commitments are also the most aggressive sellers when returns fall short of expectations.
Based on a $2 trillion valuation and projected 2025 revenue of $4.585 billion, investors are paying roughly 435 times the company’s trailing‑12‑month revenue. Even if we annualize the third quarter’s approximately $16.8 billion in revenue—a purely linear extrapolation, not a forecast—that would imply about $67.2 billion in revenue, yielding a forward multiple close to 30x. According to media reports, SpaceX, which went public in June 2026, traded at a multiple of roughly 26x revenue. This premium reflects an expectation of a winner‑takes‑all outcome in cutting‑edge AI, with virtually no margin for error.
This argument holds only if Anthropic’s lead in model quality can be shown to be sustainable. However, the risk factors outlined in the company’s IPO filing undermine this narrative. The company acknowledges that advanced models may exhibit unexpected strategic or self-preservation behaviors, including attempts to resist shutdowns, conceal or manipulate information, or generate outputs that could be interpreted as coercion, deception, or manipulation—potentially even resembling ransomware‑like actions. The filing also warns that losses from a catastrophic AI incident could exceed the scope of its insurance coverage. These are not boilerplate disclaimers; they recognize that, within the boundaries of the product’s capabilities, there exist uncontrollable risks that no amount of revenue growth can eliminate.
More noteworthy is the second layer of meaning behind these figures. The conventional interpretation holds that Anthropic’s losses prove AI is a capital black hole; a deeper reading, however, suggests that these losses represent a form of reallocation: Anthropic is channeling hundreds of billions of dollars in investor capital into demand for NVIDIA GPUs, Broadcom custom chips, Google TPUs, and the broader data-center ecosystem.
The promise of hundreds of billions of dollars in computing power is not just a risk for Anthropic; it also provides a floor for revenue across the entire AI infrastructure ecosystem. If Anthropic stumbles, the fallout will not be confined to its own shareholders—it will ripple through every company that sells shovels and picks to it. AI infrastructure has turned a single model company into a demand engine for the semiconductor cycle.
Periodic hype or structural shift
Is this an impending bubble burst, or a genuine paradigm shift? The answer is both—and this distinction is crucial.
Valuation multiples are cyclical; they arise when speculative capital chases scarce assets, and they will contract if growth slows or competitors achieve a decisive leap in capabilities. During the dot-com bubble, companies trading at hundreds of times revenue emerged, most of which eventually collapsed, but a handful grew to match those valuations—Amazon leading the way and subsequently dominating its respective industries for decades.
However, the structural shift is real and independent of valuation multiples. Artificial intelligence is not a product cycle; it is a general-purpose technology being embedded at every level of the economy, with the companies that control cutting-edge models sitting at the throat of this transformation. Anthropic’s multi‑hundred‑billion‑dollar commitment to computing power is not an isolated episode of irrational exuberance—it is the price of entry into a market where capacity, rather than code, has become the scarce resource. Hyperscale cloud providers understand this well, which is why they simultaneously serve as Anthropic’s suppliers, shareholders, and competitors. This landscape is unprecedented and will not revert to the mean.
The strongest argument against buying this IPO is straightforward and deserves serious consideration: Anthropic is committing to fixed costs that can only be sustained if revenue continues to grow at a pace no publicly traded company has ever maintained, while its revenue depends on customers who can walk away at any time and partners who have strong incentives to squeeze it.
If revenue growth slows further from its current pace, and if the 45% quarter-over-quarter increase in the third quarter proves to be a peak rather than a temporary pause, the company will face pressure on two fronts: on one side, non-cancelable bills that are about to expire; on the other, ultra-large-scale cloud partners armed with competitive offerings, eager to poach customers that Anthropic can no longer serve. In such a scenario, the $72.4 billion in book‑value adjustments would be the least of its worries, as operating losses—i.e., cash burn—would become a hard constraint.
Meanwhile, the discussion around a $1.5 trillion valuation is an early signal: the party financing this expansion has already begun to lose confidence.
If Anthropic’s quarter-over-quarter revenue growth falls below 30% for two consecutive quarters, or if it renegotiates or abandons a substantial portion of its non-cancellable compute commitments, the structural growth thesis will collapse. Either outcome would indicate that demand is failing to keep pace with its fixed cost base, at which point valuation multiples would revert and converge toward those of infrastructure companies, rather than sustaining the premium typically enjoyed by platform winners.
What will happen next?
In the short term, the pricing of this IPO will set the tone. If it successfully lists at close to $2 trillion, it will validate the premium attached to cutting-edge artificial intelligence and pave the way for OpenAI’s fundraising; if market reaction is weak or there is a downward revision to $1.5 trillion, it will force the entire private‑equity AI ecosystem to reprice, and the IPO window for other loss‑making unicorns may also close as a result.
In the medium term, focus on three key metrics: revenue quarter-over-quarter growth (whether the 45% growth rate can be sustained or continues to decelerate), cash burn excluding accounting expenses (whether the operating loss as a percentage of revenue is narrowing), and customer concentration (whether Anthropic can diversify its customer base beyond its top two clients). These indicators will reveal to investors whether the company is maturing or merely scaling up its promised ambitions.
Governance structure is the fourth observation: the founder, through Founder LLC, holds Class F common shares that lock in 50.1% of voting rights; the Long-Term Interest Trust has the authority to appoint and remove a majority of board members; and what public shareholders acquire is a growth certificate without majority voting rights.
In the long term, the question is whether Anthropic can turn its lead in model quality into a lasting competitive moat before its massive cloud partners close the gap. If it succeeds, today’s losses will be seen as the cost of building a great company; if not, these compute commitments will become the most expensive leases in corporate history.
The real takeaway from this document is that the funding landscape of the AI race has shifted: no longer dominated by venture capital, it now comprises growth‑hungry public investors, debt markets demanding double‑digit returns, and computing‑power contracts that lock in ten years of spending even before revenue materializes. What Anthropic has filed is not just a IPO application; it is an offer to the market to pay up in advance for the vision of “artificial intelligence becoming a core economic infrastructure.”
The figures indicate that this future will be costly. However, the numbers do not confirm that it will inevitably materialize. (First published on the Titanium Media app; author: Silicon Valley Tech-news; editor: Qin Conghui)
For more exciting content, follow TMTPost on WeChat (ID: taimeiti), or download the TMTPost App.