Photo: Pexels
Anthropic has finally put a price tag on its ambition, and the number is so large it is hard to take in at first glance: about $895 billion. That figure, buried in an amended stock-market filing made public on Monday 2 November 2026, is what the company behind the Claude chatbot has committed to spend, almost all of it on computing power and the data centres that provide it. It arrives in the same document as a quarterly revenue figure that would have seemed fanciful a year ago — $17 billion in three months — and a net loss of $23.3 billion in the same period.
Read together, those three numbers tell the story of the artificial intelligence industry in late 2026 better than any single earnings call. Revenue is growing at a pace almost no large company has ever sustained. The cost of staying in the race is growing faster. And the public market, which is about to be asked to fund the gap, is starting to ask harder questions. Anthropic's listing, expected later in November on the Nasdaq under the ticker ANT, will be the first time ordinary investors can buy shares directly in a standalone AI laboratory of this scale. What they decide will set the price for the rest of the boom.
What Anthropic actually filed on 2 November
Anthropic first filed its public prospectus, known as an S-1, on 26 October. That document covered the first half of 2026 and left several important figures blank, including the total value of its future spending commitments. The amendment filed on 2 November fills in the third quarter, which ended on 30 September, and puts numbers where the blanks were. The share price range is still empty, so the listing has not yet been priced.
The headline figures are these. Revenue in the July-to-September quarter was $16.96 billion, reported as about $17 billion. That is 47 percent higher than the $11.56 billion recorded in the second quarter, and 1,127 percent higher than the $1.38 billion of the same quarter a year earlier. For the first nine months of 2026, revenue reached $33.24 billion, against about $2.5 billion in the same period of 2025. The company also reported net revenue retention of about 900 percent as of 30 September, a measure of how much existing large customers increased their spending, and said it had 1,900 customers spending more than $1 million each.
On the other side of the ledger, the third-quarter net loss was $23.29 billion, taking the nine-month net loss to about $104 billion, compared with a $25.1 billion loss a year earlier. More than $94 billion of the nine-month loss comes from non-cash fair-value charges on convertible notes and related instruments — accounting entries that rise as the company's own valuation rises, rather than cash leaving the bank. The operating result, which strips out those valuation effects but still counts the real costs of running the business, moved from a $974 million profit in the second quarter to a $567 million operating loss in the third quarter.
Two costs explain most of that swing. Computing costs rose by around $5 billion in the quarter, as Anthropic paid more for short-term computing capacity. The filing also includes a non-cash charge of about $2 billion linked to the company's employee donation-matching scheme. Neither figure, on its own, means the business is collapsing. Together, they show how quickly the economics of a frontier AI company can move when it buys computing capacity at short notice.
The $895 billion bill, explained in plain terms
The commitment figure needs careful reading, because it is the number most likely to be misunderstood. Anthropic has not already spent $895.4 billion. It has signed purchase commitments and leases worth about that amount, covering deals already on the books and leases that have not yet started, and it says substantially all of it relates to compute and infrastructure. The breakdown in the amended filing is as follows: about $320 billion in contracts on the books at the end of September, mostly hosting deals with cloud providers; about $100.2 billion in deals signed since 30 September, mainly data-centre leases and cloud hosting; and the remainder in further commitments that run far into the future.
About $69.8 billion of the total comes due in the twelve months to September 2027. About $447.5 billion is due before the end of 2030, and almost all the rest falls between 2031 and 2045. For comparison, a confidential draft reported in September had put the commitments at $518 billion. The new total is therefore not a small revision. It is a much larger claim on the company's future cash flow than investors had been led to expect a few weeks ago.
The filing also shows the direction of travel in costs already incurred. Compute and infrastructure expenses rose by $14.4 billion in the first nine months of 2026, an increase of 329 percent on the same period last year. Anthropic relies on Amazon, Google and SpaceX, as well as smaller specialist cloud providers known as neoclouds, for computing capacity. That dependence matters, because the same companies are also its competitors, its landlords and, in some cases, its investors.
None of this makes the $895 billion figure a debt in the ordinary sense. Some of it is leases for buildings and machines that do not yet exist. Some of it can, in principle, be renegotiated if demand falls. But it cannot be wished away either. A commitment of this size shapes every other decision the company makes: which models to train, how much capacity to reserve, and how much revenue it must generate in 2027, 2028 and beyond simply to stand still.
Why the loss numbers mislead — and why they do not
It is tempting to look at a $104 billion nine-month loss and conclude that Anthropic is burning money at an unsustainable rate. That reading misses the accounting. When a private company's valuation rises sharply, the convertible notes it issued to early investors become more valuable, and accounting rules require the company to record that increase as a loss, even though no cash has changed hands. Anthropic's filing makes clear that most of its headline loss is of this kind. The operating loss of $567 million in the third quarter is a far better guide to the underlying business, and it is small relative to $17 billion in quarterly revenue.
At the same time, it would be wrong to dismiss the loss entirely. The operating swing from profit to loss in a single quarter is real, and it was driven by real spending on computing. Growth, too, is slowing in percentage terms even as the absolute numbers grow: 47 percent quarter-on-quarter growth in the third quarter, against 144 percent in the second quarter. That pattern is normal for a company getting larger — the base is bigger, so the same amount of new revenue is a smaller percentage — but it is exactly the pattern investors watch when they judge whether a growth story is maturing or stalling.
The honest summary is this: Anthropic is not losing $23 billion in cash every quarter, but it is not yet a business that reliably converts revenue into operating profit while paying the full price of the computing it needs. Whether it can do so depends less on the next quarter's revenue than on whether the price of computing falls, whether customers keep expanding their use of Claude, and whether the company can sign longer, cheaper capacity deals instead of buying short-term capacity at a premium.
The control question investors cannot avoid
Anthropic's corporate structure is as unusual as its numbers. The company is a public benefit corporation, legally required to weigh its mission of developing AI safely alongside returns to shareholders. Its co-founders, chief executive Dario Amodei and his sister Daniela Amodei, who is president, will hold 50.1 percent of shareholder voting power through a single special share, known as a Class F share. Investors who buy shares on the market will therefore have little say over the company's direction.
That structure is deliberate. Anthropic was founded in 2021 by former OpenAI researchers with AI safety as its central focus, and the founders have argued that control must stay with people committed to that mission rather than passing to the highest bidder. In October, the company also set aside 5 percent of its value for a Public Benefit Fund. The amended filing adds a new safety warning that was not in the October version: Anthropic says its internal estimates indicate that it may imminently meet its own Responsible Scaling Policy definition of dramatically accelerated progress attributable to AI. In plain language, the company believes its next models may represent a step change, not an incremental improvement.
The roughly 400-page filing devotes more than 90 pages to risks, including the risk that AI systems behave in unexpected and harmful ways. That is not boilerplate. It follows a year of reports about AI systems giving dangerous advice, and it sits alongside the company's own safety research. For investors, the question is not whether Anthropic takes safety seriously — the structure suggests it does — but whether a company controlled by its founders, committed to a public benefit mission, and warning about its own rate of progress will make the same commercial choices as a conventional profit-maximising firm. The IPO prospectus, unusually, invites that question rather than hiding it.
Why this IPO is the market's test of the AI boom
Anthropic is not listing in calm conditions. On Friday 30 October, the Nasdaq fell 2.9 percent in its worst day since June, as investors punished big technology companies for spending heavily without clear evidence of returns. The sell-off followed earnings that showed enormous capital spending: Amazon, Alphabet, Microsoft and Meta together are on course to spend hundreds of billions of dollars on AI infrastructure in 2026, and several reported weaker free cash flow as a result.
Since then, the funding pressure has only become more visible. OpenAI cut its fundraising target from $30 billion to about $20 billion at a $1.25 trillion valuation, after what reports described as a tepid response from investors, and is now expected to close its first round below $20 billion. Meta has raised $95 billion. Alphabet has launched its own fundraising plans. Oracle said on Monday that it was going ahead with a $20 billion plan despite doubts about its financial exposure. Anthropic, according to several media outlets, plans to raise around $100 billion in its initial public offering at a valuation of between $1.5 trillion and $2 trillion. If it does, it would surpass the record set in June by SpaceX, which raised $86 billion.
That context is what makes the Anthropic listing different from a normal technology IPO. It is not only a test of one company. It is the first opportunity for retail investors to buy shares directly in a standalone AI laboratory, rather than gaining exposure through a diversified technology giant. If the listing is priced strongly and holds up, it will be read as evidence that the market still believes the spending will pay off. If it struggles, it will be read — fairly or not — as evidence that the private-market valuations of the past two years were too high. Either way, the reaction will affect OpenAI's planned 2027 listing, the funding plans of every other AI company, and the share prices of the chipmakers and cloud providers that supply them.
The other AI story this week: open source at Wuzhen
On the same day Anthropic's amended filing was being digested in Washington and New York, a very different vision of AI's future was on display in China. The 2026 World Internet Conference Wuzhen Summit opened on Tuesday 3 November in Wuzhen, in Zhejiang province, under the theme of open source and openness, collaborative development and shared benefits. The summit runs until 5 November and features 27 sub-forums, including sessions on trusted interoperability among AI agents, robotics and embodied intelligence, and digitally driven green transition.
The centrepiece is the AI Wuzhen Dialogue, which brings together executives from global technology companies, leaders of open-source large-model companies and AI scientists to discuss how to build a global open-source AI ecosystem. An open-source ecosystem experience hall opened on 1 November at the Wuzhen World Internet Science and Technology Museum, showing technical achievements in AI open-source development and progress on China's national pilot base for AI application testing. The Light of Internet Expo brings together more than 670 enterprises from 54 countries and regions, including Intel, Qualcomm, Microsoft, SAP, Alibaba, Huawei and Tencent.
The contrast with Anthropic could hardly be sharper. Anthropic's model is closed, capital-intensive and controlled by its founders: spend hundreds of billions on computing, train the most capable models, and sell access to them. The open-source model promoted at Wuzhen is cheaper to adopt, easier to adapt, and harder for any single company to control. Since early 2025, a growing number of open-source large models have emerged, offering lower costs and greater efficiency, and the Wuzhen organisers argue that open source is an important pathway toward inclusive AI development. Anthropic's filing does not settle that argument. It sharpens it, by showing exactly how expensive the closed model has become.
Our analysis: three things the filing really tells us
First, commitments are becoming the story, not revenue. For two years, AI coverage has focused on how fast revenue is growing. Anthropic's filing shifts the focus to the other side of the balance sheet. A company that has committed $895 billion to computing has, in effect, taken out a mortgage on the next twenty years of AI demand. If demand keeps growing at anything like the current pace, that mortgage looks like a sensible investment in the factory of the future. If demand slows, or if cheaper models reduce the amount of computing each task requires, it looks like an overhang that will weigh on every future decision. Investors are no longer being asked to bet only on Anthropic's models. They are being asked to bet on twenty years of AI demand.
Second, the deceleration in growth matters more than the headline loss. A 1,127 percent annual increase grabs attention, but no company sustains four-digit growth. The more useful number is the quarter-on-quarter slowdown from 144 percent to 47 percent, combined with the swing to an operating loss driven by short-term computing purchases. That combination suggests Anthropic is hitting the practical limits of buying capacity quickly, rather than a collapse in demand. The company's future depends on solving a supply problem — securing enough computing at a predictable price — as much as on building better models. That is a less exciting story than a breakthrough model, but it is the one that will determine whether the business makes money.
Third, the IPO will price two different ideas at once. One is Anthropic itself: its revenue, its costs, its governance and its safety commitments. The other is the AI boom as a whole: whether the hundreds of billions being spent on data centres, chips and power will generate returns that justify the spending, or whether the industry is building capacity ahead of demand. Anthropic's founders have tried to separate those two ideas by keeping control and writing safety into the company's legal structure. The market may not respect that separation. If the listing falters, it will be blamed on the boom. If it succeeds, it will be credited to the boom. Either way, the company that wanted to be judged on its mission will, for a few weeks at least, be judged on its price.
What to watch next
The first test is pricing, expected in mid-November. The filing still leaves the price range blank, and the final range will show how much demand the banks managing the listing — led by Goldman Sachs and Morgan Stanley — actually find. The second test is the listing itself, expected later in November, and whether the shares hold their value in the first weeks of trading, when retail investors get their first chance to buy.
The third test is cash. About $70 billion of Anthropic's commitments come due before September 2027. Watching whether the company funds that bill from revenue, from the IPO proceeds, or from new borrowing will tell investors more than any single quarter's revenue figure. The fourth test is competition: whether OpenAI's reduced fundraising, the open-source models showcased at Wuzhen, and cheaper models from China slow Claude's growth or force Anthropic to cut prices. The fifth is safety: whether the company's warning about dramatically accelerated progress is followed by concrete changes to how it tests and releases its next models.
None of these tests will be settled in November. But November is when the market gets its first chance to vote, with real money, on whether a company can be worth $1.5 trillion to $2 trillion while committing $895 billion to computing it has not yet used. That vote will echo far beyond one listing.
Frequently asked questions
What did Anthropic report in its amended IPO filing?
Anthropic reported third-quarter revenue of about $17 billion, up 1,127 percent on a year earlier, a third-quarter net loss of $23.3 billion, and total purchase and lease commitments of about $895.4 billion, almost all for compute and infrastructure. The filing was made on 2 November 2026 and adds the third quarter to the prospectus first filed on 26 October.
Has Anthropic already spent $895 billion?
No. The figure is the total value of commitments and leases signed, including leases that have not yet started. About $69.8 billion comes due in the twelve months to September 2027, about $447.5 billion before the end of 2030, and almost all the rest between 2031 and 2045. It is a claim on future cash flow, not money already paid.
Why is Anthropic reporting such a large net loss if revenue is $17 billion?
Most of the loss is non-cash accounting charges linked to the rising value of convertible notes, which must be recorded as a loss when the company's valuation rises. The operating loss, which better reflects the underlying business, was $567 million in the third quarter, after a $974 million operating profit in the second quarter, driven mainly by higher computing costs and a donation-matching charge.
When is Anthropic's IPO and what valuation is expected?
Anthropic is expected to list on the Nasdaq under the ticker ANT later in November 2026, with pricing expected in mid-November. Media reports put the planned fundraising at around $100 billion at a valuation of between $1.5 trillion and $2 trillion, but the price range had not been set in the amended filing.
Who controls Anthropic after the IPO?
The co-founders, Dario Amodei and Daniela Amodei, will hold 50.1 percent of voting power through a single Class F special share. Anthropic is also a public benefit corporation, legally required to weigh its safety mission alongside shareholder returns, and it has set aside 5 percent of its value for a Public Benefit Fund.
What does the Wuzhen Summit have to do with Anthropic?
The Wuzhen Summit, which opened on 3 November 2026, promotes open-source AI as a cheaper, more widely shared alternative to the closed, capital-intensive model Anthropic represents. The contrast matters because if open-source models keep improving, they could reduce the computing advantage that Anthropic's $895 billion in commitments is meant to buy.
Sources
- AFP via Malay Mail — “Anthropic revenue jumps 1,100pc as AI giant prepares for IPO”, 3 November 2026 — Malay Mail
- AFP via Channel News Asia — “Anthropic says revenue surged 1,100% ahead of IPO”, 2–3 November 2026 — Channel News Asia
- MadRobot — “Anthropic IPO filing: $17bn quarter, $895bn compute bill”, 2 November 2026 — MadRobot
- VKTR — “Anthropic IPO UPDATE: 900% Net Revenue Retention, Founder Control”, updated 2 November 2026 — VKTR
- Crypto Briefing — “Anthropic swings to a $567 million operating loss in amended IPO filing”, 2 November 2026 — Crypto Briefing
- TipRanks — “Anthropic Reports $17 Billion in Third-Quarter Revenue Ahead of IPO”, 2 November 2026 — TipRanks
- CGTN / Xinhua — “2026 WIC Wuzhen Summit opens with focus on open-source AI”, 3 November 2026 — CGTN
- China Daily — “WIC Wuzhen Summit to unveil new vision for open source, openness”, 23 October 2026 — China Daily
For more analysis, visit abdulrahimmughal.com.