SAN FRANCISCO, November 3, 2026 — Anthropic’s amended IPO filing reports nearly $17 billion in third-quarter revenue and about $895.4 billion in aggregate purchase and lease commitments, most of them tied to the computing infrastructure needed to train and run its artificial-intelligence models. The updated filing gives investors a fuller view of the scale of the Claude developer’s business, the accounting volatility in its reported losses and the extraordinary cost of supporting future growth.
Revenue for the three months ended September 30 reached approximately $16.96 billion, up from about $1.38 billion in the same quarter a year earlier. For the first nine months of 2026, revenue rose to $33.24 billion from $2.55 billion in the comparable period of 2025. The figures show that demand for Anthropic’s services has expanded rapidly, but the filing also makes clear that revenue growth is being accompanied by large spending commitments and substantial accounting volatility.
Anthropic recorded a net loss of about $23.29 billion in the third quarter, compared with a loss of roughly $12.14 billion a year earlier. Its net loss for the first nine months reached $104.04 billion, up from $25.11 billion in the corresponding period of 2025. The headline losses should not be read as a direct measure of cash spent operating the business: a large portion of the reported nine-month loss was linked to accounting changes in the value of financing-related instruments, according to the filing and reporting on it.
Also read: Banks Arrange $63 Billion
The company’s third-quarter operating result was also weaker than the previous quarter. Anthropic reported an operating loss of about $567 million for the three months ended September 30, after operating income of roughly $974 million in the second quarter. The change reflects the cost of scaling the business, including computing capacity and other expenses, rather than the same accounting effects that drove much of the reported net loss. Investors will need to separate operating performance, cash flow and non-cash accounting adjustments when assessing the company’s financial position.
The most striking figure in the amended filing is the approximately $895.4 billion in aggregate commitments under purchase agreements and leases as of and after September 30. Anthropic said substantially all of the commitments relate to compute and infrastructure. About $447.5 billion is expected to be incurred before the end of 2030, while approximately $69.8 billion is due in the 12 months following September 30, 2026.
Those obligations are not the same as a bill due immediately, and they should not be described as cash already spent. The total includes long-term agreements, purchase commitments and leases with different timing. The filing says much of the remaining amount is expected to fall due between 2031 and 2045. Even so, the scale shows how much of Anthropic’s future business depends on access to processors, data-centre capacity, power, networking and other infrastructure over many years.
Frontier AI models require large amounts of computing power both during training and when customers use them. Training runs can demand concentrated capacity for extended periods, while inference—the process of generating responses or completing tasks—creates recurring demand that varies with customer usage. As models become more capable and are used in software agents that perform multi-step work, the amount of computing consumed per task can also rise. Providers therefore need to secure infrastructure ahead of demand, even though the revenue generated by future workloads is uncertain.
Anthropic’s filing describes a shift towards a hybrid approach that combines fixed-term cloud agreements with direct leasing of chips and dedicated data-centre capacity. The company says this can give it more flexibility by separating commitments for facilities from commitments for the accelerators installed inside them. Such arrangements may help it secure space early and defer some hardware decisions, but they also leave the company exposed to long-term obligations if demand, technology or pricing changes.
The scale of the commitments has implications for the wider AI market. Large developers are competing for access to advanced accelerators, specialised networking and power-ready sites. Data-centre operators need to finance buildings and grid connections years before some facilities reach full utilisation. Chip suppliers and cloud providers can benefit from this demand, but the same commitments create risks if customers delay projects, shift workloads to more efficient models or fail to generate enough revenue to support the planned expansion.
Anthropic said it held about $131.85 billion in cash, cash equivalents and short-term investments as of September 30. That balance is substantial, but it should be compared with the timing and scale of the company’s obligations rather than treated as proof that every future commitment is fully funded. The filing does not mean the company must pay the entire $895.4 billion immediately. It does mean that investors will be watching the pace of new commitments, the ability to raise capital and the relationship between future revenue and infrastructure costs.
The amended filing forms part of Anthropic’s preparation for a possible public listing. The company has said it intends to list on the Nasdaq Global Select Market under the proposed ticker ANT, but the updated filing does not set a final offer price or specify the amount it plans to raise. The timing of a roadshow or first trading day should not be treated as confirmed unless the company or regulatory documents establish it.
Anthropic’s revenue growth reflects rising demand for Claude from businesses and other users. The company has built services for coding, research, enterprise workflows and other tasks where customers pay according to usage or subscription terms. But a fast-growing revenue base can still be vulnerable to pricing pressure, competition and changes in customer behaviour. A model that becomes more efficient may reduce the computing cost of a task, but it may also allow customers to run more tasks at a lower price. The effect on margins depends on how prices, usage and infrastructure costs evolve together.
The company also faces customer-concentration and supply-chain questions common to the frontier AI sector. Large cloud providers can be investors, infrastructure suppliers, distribution channels and competitors at the same time. That arrangement gives AI developers access to capital and computing resources, but it can also create dependencies on a limited number of partners. The amended filing describes Anthropic’s use of multiple providers and its effort to secure capacity through long-term arrangements; it does not eliminate the possibility of delays, higher costs or changes in contract terms.
Compute capacity is not only a financial issue. Data centres require dependable electricity, cooling, fibre connections and equipment that can be installed and maintained at scale. New facilities can face delays from grid access, permitting, construction bottlenecks and local opposition. In some locations, operators have considered on-site generation or other arrangements to secure power, but those choices bring their own cost, environmental and regulatory questions. Anthropic’s commitments make the availability and price of infrastructure a material factor in its future performance.
There is also a timing problem. A long-term agreement may help guarantee access to hardware, but chip generations change quickly. An accelerator that is competitive when a contract is signed may be less efficient than a later design before the facility reaches its planned utilisation. Developers can try to mitigate this risk by diversifying suppliers, adopting flexible infrastructure and improving model efficiency, but long commitments reduce the ability to change course without cost.
For investors, the important distinction is between growth that produces sustainable cash generation and growth that requires ever-larger financing. Anthropic’s revenue increase is substantial, but the filing also shows that the company expects infrastructure to remain a major use of capital. The next financial reports will help establish whether higher usage and enterprise demand can outpace the costs of securing and operating computing capacity.
The accounting treatment of financing instruments is another issue to follow. A large fair-value adjustment can make net income or loss swing sharply even when the movement does not correspond to a similar cash payment in the same period. That does not make the reported net loss irrelevant: it reflects the accounting consequences of financing arrangements and changes in valuation. But readers should avoid equating the entire $104 billion nine-month loss with money spent on salaries, model training or data-centre bills.
Anthropic’s proposed listing would give public-market investors a chance to examine those trade-offs in greater detail. The filing will be scrutinised for revenue concentration, customer retention, infrastructure costs, contractual obligations, governance and the risks of developing and deploying advanced models. A prospectus provides the company’s own disclosures under securities rules; it is not a guarantee that forecasts will be met or that the business will achieve profitability.
For the AI industry, the figures underline the difference between model demand and the economics of delivering that demand at scale. Strong customer interest can support rapid sales growth, but each additional workload must be served with available computing capacity at a cost that leaves room for a sustainable business. Infrastructure contracts can reduce the risk of being shut out of supply, yet they can also commit a company to spending long after market conditions have changed.
The next milestones are the SEC’s review of the amended registration statement, any subsequent revisions to the financial disclosures, and the company’s decisions on offering size, pricing and timing. Investors will also look for evidence that revenue growth is translating into more durable operating performance and that infrastructure spending is being matched to demand. Until those details emerge, the filing offers a clearer snapshot of Anthropic’s scale, but not a final answer on the economics of its planned public debut.
Also read: Australia’s 1GW AI Data Centre Drops 2028 Start Target After Regulatory Delay
SOURCE
U.S. Securities and Exchange Commission; Associated Press; Financial Times; Axios.