AI News

Anthropic Announces IPO Price Range, Valuation Halved from Expectations Five Weeks Ago

36Kr · AIUpdated Nov 11 21:13

AI summary

Anthropic announced on November 10 an IPO price range of $70 to $78 per share, corresponding to a valuation of up to approximately $1.02 trillion based on the base share count, halved from the $2 trillion high point discussed in the market about five weeks ago. The company's third-quarter revenue was $16.96 billion, up 1127% year-over-year, with an annualized revenue run rate of about $81 billion as of the end of October; seven co-founders plan to sell a combined approximately 23.65 million shares, cashing out up to about $1.845 billion at the IPO price range. The prospectus also disclosed total contracted committed spending of $900.9 billion, the vast majority of which is related to AI compute and infrastructure.

Why read this

SelectedAI score 82

From IPO pricing and revenue growth to the $900.9 billion compute commitment, the capital structure shows the coexistence of high growth and heavy investment at AI companies.

Topics

Tags

Body · AI translation

On November 10, Anthropic officially announced the IPO price range: $70 to $78 per share, corresponding to a valuation of up to approximately $1.02 trillion based on the basic share count.

About five weeks ago, market discussions of the IPO valuation still put expectations at a high of $2 trillion; now that has been cut directly in half.

On a fully diluted share basis, the upper limit of the IPO valuation is approximately $1.4 trillion, still higher than the company's $965 billion financing valuation in May of this year.

Just one day before the official pricing, Anthropic had also sent investment banks a proposed offering plan to raise $69.3 billion, which was subsequently withdrawn due to incorrect pricing information. The final public offering size was adjusted to a maximum of $64.4 billion.

Even more striking is that Anthropic's business is still growing at high speed. Third-quarter revenue was $16.96 billion, up 1127% year over year; as of the end of October, the annualized revenue run rate had reached $81 billion. Separately, according to TechCrunch, Anthropic's seven co-founders plan to sell 3,378,400 shares each, for a total of approximately 23.65 million shares. Based on the aforementioned offering price range, the seven could cash out a combined maximum of approximately $1.845 billion (about RMB 12.3 billion).

Claude's revenue surged 11-fold in a year, yet Anthropic's expected IPO valuation has shrunk sharply.

Claude's revenue surged 11-fold; why does Wall Street still consider it expensive?

Claude once produced a gratifying revenue growth trajectory. In 2023, Anthropic's full-year revenue was only $41 million. It grew to $386 million in 2024 and reached $4.585 billion in 2025.

By 2026, in the first three quarters alone, the company achieved $33.242 billion in revenue, more than seven times last year's full-year total. Third-quarter revenue alone was $16.956 billion, compared with only $1.382 billion in the same period last year.

Bloomberg compared Anthropic with the pre-IPO revenue growth of major technology companies in history. Even for companies such as Google and Meta, their growth curves before going public back then were far behind Anthropic's today.

This growth is already supported by a large amount of enterprise spending.

As of the end of September this year, customers that contributed at least $100,000 in revenue to Anthropic over the past 12 months had increased from more than 950 in the same period last year to more than 11,900. Among them, more than 1,900 customers reached the million-dollar level, and about 250 reached the ten-million-dollar level.

Claude Code, the Claude API and other enterprise applications are entering software development, internal knowledge management and various business processes. Enterprise customers are no longer just experimenting with chatbots; some companies have already incorporated Claude into their daily production systems.

But a set of data disclosed at an investor meeting in October also prompted the market to begin reassessing Anthropic's growth space.

In the second quarter of this year, Anthropic's revenue grew 144% quarter over quarter, falling to 47% in the third quarter. That figure is still very high, but Reuters in August, citing people familiar with the matter, said Anthropic's forecast for 2028 revenue is close to $200 billion. Calculated on an annualized basis from third-quarter revenue, meeting that target would still require sustained high-speed expansion over the next two years.

At the same time, price competition in model services is intensifying. OpenAI, Google and Chinese open-source model vendors continue to launch new products, and as model capabilities improve, the price per token continues to fall. Enterprises may call more tokens, but they do not necessarily need to increase spending by the same proportion.

Anthropic therefore faces two changes occurring at the same time: more and more enterprise customers, and increasingly cheaper model services.

Investors were previously willing to pay high valuations for sustained ultra-high-speed growth; now they are beginning to reassess future revenue growth rates and profit margins. After all, the company's valuation has already reached the trillion-dollar scale, and any change in long-term growth assumptions corresponds to a huge difference in value.

Greater pressure also comes from the cost side.

Forecasting computing power demand has also become a difficult problem

The prospectus updated on November 10 disclosed one of Anthropic's most astonishing sets of figures to date.

$900.9 billion.

This is the total committed expenditure under contracts the company has signed, including procurement and leases, most of which is related to AI computing power and infrastructure. Of this, approximately $451.6 billion is expected to occur before the end of 2030, about $70.7 billion in commitments will come due within 12 months starting from September 30, 2026, and a large amount of the remaining expenditure will extend from 2031 to 2045.

$900.9 billion is equivalent to about 11 times Anthropic's annualized revenue run rate of approximately $81 billion as of the end of October.

These contracts span multiple years and include different maturities and payment terms, so they cannot be directly treated as debt already due. But such a massive procurement arrangement has already deeply bound Anthropic's operations and computing power needs for many years to come.

Enterprises using Claude Code can increase or decrease usage at any time. In the first three quarters of 2026, about 79% of Anthropic's revenue came from usage-based services.

But GPUs, servers, data centers and electricity need to be prepared far in advance. Large computing facilities often take a year or even longer from chip procurement to building computer rooms. Anthropic must forecast future demand before customers actually begin using the service.

In the third quarter of this year, the company already paid the price for underestimating demand.

To meet rapidly growing customer demand, the computing power Anthropic delivered in the third quarter included about $5 billion of short-term contracted computing capacity. The cost of this capacity was significantly higher than the resources provided under long-term contracts.

Adjusted operating profit rose from $2.2 billion to $2.5 billion for the quarter, but the adjusted operating margin fell from 19% to 15%. Under GAAP accounting standards, the company posted an operating loss of $567 million in the third quarter, compared with operating profit of $974 million in the second quarter. In addition to high-priced computing capacity procurement, about $2 billion in non-cash donation matching expenses also affected the quarter's results.

Strong demand is of course a good thing. But when customer usage grows faster than the computing capacity the company prepared in advance, spot purchases may instead compress profits.

In February this year, Anthropic CEO Dario Amodei discussed this issue on the Dwarkesh Podcast.

He gave an extreme hypothetical: if you forecast that revenue could reach $1 trillion in 2027, and therefore buy computing capacity of the same scale in advance, but revenue ends up at only $800 billion, the company may not be able to bear the corresponding costs. He even said that being wrong by just one year in a growth forecast could drag a company into bankruptcy.

Amodei stressed at the time that Anthropic had carefully assessed its procurement scale and would not bet entirely on the most optimistic growth assumptions. But he also acknowledged that computing capacity procurement always has to contend with enormous demand forecast errors.

For Anthropic, if it buys too little computing capacity, customers may turn to competitors because of insufficient service capacity; if it buys too much, contracts signed in advance may create severe financial pressure.

Moreover, competition in frontier models continues. The training, inference and product expansion of each new generation of models require sustained investment.

The $900.9 billion in contractual commitments are precisely the huge sums Anthropic has staked in advance to compete for the future market.

Nvidia, Google and Microsoft are all on this ship

Anthropic's computing capacity expansion has another peculiarity: the tech giants that provide it with funding are often also its suppliers.

In the IPO plan dated November 10, Anthropic simultaneously arranged a $26 billion strategic private placement, with participants including Google, Nvidia, AMD, Broadcom, Micron, Microsoft and TSMC-related entities.

Among them, Google and Nvidia each plan to invest $10 billion.

These companies sit at different links in the AI industry chain. Nvidia sells GPUs, Google and Microsoft provide cloud services, and AMD, Broadcom, Micron and TSMC participate in the supply of chips and related infrastructure. After Anthropic raises funds and continues to expand computing capacity procurement, some of these companies will also earn business revenue from the related spending.

Cloud providers also control important sales channels for Anthropic.

In 2023, revenue generated by Anthropic through cloud platform marketplaces accounted for only 11%. By 2025, that proportion had reached 47%; in the first three quarters of 2026 it was already 44%.

The ultimate paying customers behind these channels include a large number of independent enterprises. However, cloud service providers simultaneously participate in investment, provide computing capacity and help distribute Claude, making the commercial interests of all parties highly interconnected.

This transaction structure is attracting attention from financial institutions.

In October this year, the International Monetary Fund reviewed 14 AI industry transactions, involving about $300 billion in equity investment, $650 billion in debt financing and $1.4 trillion in future revenue commitments.

Another study by the Bank for International Settlements found that among investment transactions between AI companies from 2021 to 2025, about 46.4% by value also involved a commercial supply chain relationship between the investor and the investee.

Chip companies invest in model companies, and model companies use the financing to buy chips and computing capacity; the related transactions can be built on real corporate demand. But when a model company's revenue falls short of expectations, losses may be transmitted outward along the supply chain.

For example, if Anthropic cuts future computing capacity procurement, cloud providers may lose revenue, data center operators may need to adjust expansion plans, and companies that rely on related contracts for financing will also face debt repayment pressure.

After Anthropic's valuation expectations were sharply cut, the 2000 internet bubble is once again being invoked for comparison.

Back then, a large number of internet companies had astonishing user growth but were never able to establish a sustainable profit model. After capital markets tightened financing, a batch of companies quickly collapsed.

The AI industry today already has sizable paid demand, and Anthropic's enterprise customers and revenue growth are direct evidence. But having real demand for a technology does not guarantee that all participants can make money, still less that the scale of capital investment across the industry is reasonable.

In July this year, the Bank for International Settlements published "The AI Investment Race," studying the investment behavior of large tech companies as they compete for leadership in the AI market.

Its model found that under a conservative baseline scenario, corporate investment may exceed the economically efficient level by about 50%; in some scenarios with lower demand elasticity, the scale of investment may even reach three times the efficient level.

This is an economic model's simulation of the investment race, not an actual measurement that the industry has already wasted 50% of its capital.

The study explains a real-world dilemma: if several companies all believe the future AI market will be dominated by a handful of firms, then even if they suspect that computing power investment is overheating, it is still hard for them to voluntarily cut back on procurement. The party that slows investment first may fall behind in model capabilities and service costs.

Every company worries that it is buying too little, yet the industry as a whole may end up buying too much.

Figure | The AI investment race and overinvestment (Source: BIS Working Paper 1367)

The financing market is already showing some signs of stress.

At CoreWeave, an AI computing power leasing company, the credit default swap spread at one point exceeded 1,100 basis points in October this year, and some bond yields reached double digits. For companies that rely on debt financing to buy GPUs and build data centers, rising financing costs will directly erode project returns.

GPUs themselves are also subject to rapid technological iteration. By the time a data center is completed, the next generation of chips may already offer better performance and lower cost per unit of computation. If the utilization rate and rental income of older equipment decline, the debt incurred by companies to purchase that equipment still has to be repaid.

This type of risk is asymmetric between traditional large technology companies and emerging computing power companies. Microsoft, Google and Amazon have mature profitable businesses and can use existing cash flow to support capital expenditure. Some independent computing power companies, by contrast, are highly dependent on external financing and long-term customer contracts.

Anthropic has not yet experienced a financing rupture, and its enterprise customers are still growing. By comparison, some AI infrastructure companies that are highly leveraged and rely on future orders to repay debt have already begun to bear higher financing costs.

Claude is already making this much money, so why is the cash return still something to wait for?

Goldman Sachs' latest Anthropic forecast provides a striking contrast.

It expects Anthropic's revenue to reach $52 billion this year, $136 billion in 2027, grow to $442 billion in 2029, and possibly even exceed $600 billion in 2030.

But under the same set of forecasts, Anthropic will still generate about $65.7 billion in free cash flow outflow in 2027, is only expected to achieve operating profitability in 2029, and free cash flow is only projected to turn positive in 2030.

Anthropic's operating efficiency has in fact already improved markedly.

In the first nine months of 2025, computing and infrastructure expenses were equivalent to 172% of revenue in the same period; by the same period in 2026, the ratio had fallen to 57%. Enterprise revenue is growing significantly faster than the related computing power expenses, and economies of scale are emerging.

Moreover, as of the end of September, the company had about $146.6 billion in liquidity resources, including cash, short-term investments and undrawn credit facilities. This IPO and the concurrent private placement also plan to raise a large amount of funds, giving it the conditions to continue expanding in the short term.

In order to create more possibilities for commercial revenue, Anthropic even said during the roadshow that it will soon launch an AI office suite, including spreadsheets, digital whiteboards and data analysis tools. Affected by this news, Microsoft shares at one point fell 3% during intraday trading.

The capital needs of the frontier model industry remain enormous. Training new models, expanding the scale of inference, building data centers and more intense price competition may all continue to consume cash.

Goldman Sachs' forecast is already optimistic enough.

Yet according to the latest prospectus, contractual payment commitments due within just the 12 months starting September 30, 2026 amount to about $70.7 billion.

Looking back at the AI boom of the past two years, the capital market's expectations for technological breakthroughs have already run far ahead of commercial returns. Now, investors are beginning to re-examine growth expectations that were once taken for granted. Even if AI ultimately transforms society the way the internet did, that does not mean every dollar in this wave of frenzied investment will be earned back.

English text is being prepared.

Source

36Kr · AI