Anthropic multiplies its revenue twelvefold
Anthropic generated $4.59 billion in revenue in 2025, compared with just $386 million a year earlier. Growth reached approximately 1,089%. For a company founded in 2021, the change in scale has been abrupt. Claude has gone from being a promising ChatGPT rival to a product embedded in enterprises, software development and AI-agent workflows.
This acceleration is also beginning to affect sectors that once appeared far removed from artificial intelligence. BrefCrypto has notably followed the $9.1 billion contract linking Riot Platforms to infrastructure intended for Anthropic. Bitcoin miners are now redirecting part of their energy capacity, land and data centers toward AI computing. This is no longer a minor diversification.
Anthropic’s growth accelerated further in 2026. In May, according to Anthropic’s official announcements, the company reported annualized revenue of more than $47 billion. The funding round reached $65 billion at a post-money valuation of $965 billion.
This run rate must be distinguished from the $4.59 billion recorded over the whole of 2025. The former extrapolates recent activity over twelve months; the latter reflects revenue actually recorded during the fiscal year. The comparison nevertheless shows how quickly the business has changed.
This growth is not enough to make Anthropic profitable.
The operating loss rose from $2.98 billion in 2024 to $8.06 billion in 2025. It therefore amounted to approximately 1.75 times the year’s revenue. In other words, even before certain extraordinary accounting effects, Claude still costs far more to develop and operate than it generates.
The net loss was even more striking: $41.97 billion, or almost nine times revenue.
That figure requires some explanation. Approximately $34 billion came from a non-cash accounting charge linked to the revaluation of instruments issued in earlier funding rounds. The more valuable Anthropic became, the more certain obligations convertible into shares had to be revalued in its accounts.
Anthropic therefore did not send an additional $34 billion to Nvidia, Google or Amazon. A large portion of the loss exists on paper.
The $8.06 billion operating loss is much more useful for understanding the company’s underlying economics.
Another figure deserves attention: just two customers accounted for 24% of revenue, with approximately 12% each. Anthropic does not publicly identify them in the available materials. The prospectus also warns that several major customers are not bound by long-term contracts and may reduce their spending or leave.
Of $4.59 billion in revenue, 24% represents approximately $1.1 billion dependent on two customers.
For a company valued at nearly $1 trillion in May, that concentration is far from insignificant. A company can post dizzying growth while still relying on a relatively concentrated customer base.
This also explains why Anthropic is expanding its enterprise integrations. In May, the company partnered with Blackstone, Hellman & Friedman and Goldman Sachs to create a business dedicated to deploying Claude within mid-sized companies. The objective is fairly clear: turn spectacular AI demand into recurring and more diversified revenue.
The battle is therefore no longer being fought solely on model benchmarks.
It is being fought over contracts.
Compute is already consuming $7.33 billion
The figure explaining much of the losses can be summed up in two words: computing power.
Anthropic spent $7.33 billion on compute and infrastructure in 2025, almost three times as much as in 2024. These expenses represented approximately 58% of the company’s total operating costs.
The comparison with revenue is even more revealing.
Anthropic took in $4.59 billion during the year and spent $7.33 billion solely on computing and infrastructure. Compute therefore represented approximately 160% of its 2025 revenue.
Before paying all its researchers, sales staff, employees, legal fees and other expenses, the computing bill had already exceeded the company’s revenue by nearly $2.74 billion.
This sums up the economics of generative AI today in just a few figures.
Extrapolating the ratio disclosed in the prospectus, total operating expenses in 2025 were approximately $12.6 billion. This is a mathematical estimate based on $7.33 billion representing 58% of expenses, not a new figure reported by Anthropic.
What follows is even more striking.
The prospectus shows approximately $518 billion in future commitments related to cloud services, computing and infrastructure. Here again, a misleading interpretation should be avoided: Anthropic is not announcing a $518 billion check to be signed immediately, nor necessarily an expense concentrated in 2027. These are obligations and commitments spread across different contractual periods. Reuters describes them as one of the deal’s main financial risks.
For comparison, these commitments represent approximately 113 times 2025 revenue.
They also represent more than 25 times the $20.28 billion in cash and short-term investments Anthropic held at the end of the fiscal year.
This ratio obviously does not mean Anthropic must have $518 billion in the bank today. The contracts will be funded over several years through future revenue, capital raised and potentially debt markets. It nevertheless shows how heavily the model depends on a strong assumption: demand for Claude must continue to grow very rapidly.
Amazon already occupies an important place in this mechanism. Anthropic has committed to spending more than $100 billion over ten years on Amazon’s cloud technologies, while Amazon itself has increased its investment in the company.
Anthropic is also expanding its capabilities with Google, SpaceX and other providers. This race for computing power is now large enough to reshape other industries. BrefCrypto recently showed that listed Bitcoin miners are increasingly being valued as much for their AI data centers as for their BTC production. At TeraWulf, for example, HPC had already taken a major place in revenue.
Electricity, land, GPUs, high-voltage networks and data centers have become shared resources for Bitcoin and artificial intelligence.
Anthropic does, however, have an argument that many loss-making companies lack: its commercial demand is growing extremely quickly.
The annualized revenue reported at more than $47 billion in May 2026 is already more than ten times the revenue recorded for 2025. If this pace continues, the ratio between revenue and compute could quickly become less concerning.
If it slows, the hundreds of billions of dollars in commitments will become much harder to absorb.
That is the financial bet behind the IPO.
There is another paradox. Anthropic has built its brand around AI safety while participating in a race that demands ever more compute and increasingly powerful models.
The prospectus devotes approximately 80 of its 261 pages to risk factors. These include scenarios in which advanced systems could conceal certain information, manipulate their interlocutors or seek to avoid being shut down. Anthropic writes that advanced AI could pose “catastrophic or existential risks to humanity.”
Perhaps the most surprising detail concerns the allocation of resources. During one week used as an example in the document, only approximately 6% of computing capacity was devoted to safety work.
Anthropic itself explains that these trade-offs are difficult. More compute devoted to evaluation and alignment means less compute available to train, serve or improve commercial products.
Dario Amodei is also advocating slowing some advances at the frontier of AI. BrefCrypto had already analyzed his call to slow the development of the most powerful systems.
This is therefore not an external warning directed at Anthropic.
Anthropic is telling its future shareholders that the technology they are financing could itself create extraordinary risks.
Rarely has a prospectus contained such a strange conflict between commercial ambition and technological warnings.
A $2 trillion IPO changes the scale
Anthropic could now seek a valuation of more than $2 trillion in its initial public offering. Reuters reports that the listing is more likely to take place after the U.S. midterm elections in November, with no final price or definitive number of shares yet publicly announced.
The contrast with May is striking.
On May 28, Anthropic raised $65 billion at a post-money valuation of $965 billion. Surpassing $2 trillion a few months later would mean more than doubling that value again. Anthropic would immediately join the extremely limited group of companies valued at several trillion dollars.
And yet it is emerging from a year with $4.59 billion in revenue.
At a $2 trillion valuation, the market would value Anthropic at approximately 436 times its 2025 revenue. That ratio would obviously be far less dramatic when using the 2026 run rate of more than $47 billion, which would bring the multiple down to around 43 times. The two figures explain exactly why this IPO will be difficult to analyze using traditional tools.
Looking only at 2025 makes the valuation appear almost absurd.
Looking only at 2026 growth risks assuming that it will continue indefinitely.
Wall Street will have to find a position somewhere between the two.
Anthropic also has an especially unusual governance structure. Its seven co-founders, including Dario and Daniela Amodei, are to control a new entity called Founder LLC. It will control a single Class F share representing 50.1% of the voting power on several major company decisions. Class A shares intended for ordinary investors will carry one vote per share, but their influence will remain mechanically limited.
Anthropic itself warns that this structure could lead to decisions contrary to the short-, medium- or long-term financial interests of Class A holders.
That is intentional.
The company remains a Public Benefit Corporation. The founders want to prevent immediate stock-market pressure from forcing them to launch a model or product they consider dangerous simply because it could increase revenue.
The question is fascinating: what is a share in a $2 trillion company really worth when ordinary shareholders agree from the outset that maximizing their returns is not necessarily the absolute priority?
This structure will probably be one of the IPO’s main areas of debate.
It also directly concerns the crypto ecosystem. BrefCrypto recently explained that crypto investors are already speculating on companies such as Anthropic before their IPOs through pre-IPO contracts. These products do not actually provide shares, but they show that demand exists well before listing.
A $2 trillion offering could also absorb a huge amount of capital. The SpaceX IPO had already shown how a giant listing could draw available liquidity away from other risky assets temporarily. Anthropic would enter a market in which Bitcoin, technology stocks and major AI names often compete for the same growth-seeking investors.
Competition for capital is therefore becoming as important as competition for GPUs.
Anthropic’s IPO will not simply be that of a company commercializing Claude. For the first time, public markets will put a daily price on one of the most expensive economic assumptions of our era: that hundreds of billions of dollars spent on compute today can generate even greater revenue tomorrow.
The figures published so far make it possible neither to conclude that this bet has paid off nor that it is doomed.
They show something more concrete.
Anthropic has managed to multiply its revenue twelvefold in one year, then push its 2026 commercial run rate much higher. At the same time, it recorded an $8.06 billion operating loss in 2025, remains heavily dependent on a handful of customers and has entered an infrastructure race estimated at several hundred billion dollars.
The $42 billion net loss is impressive. It is also partly an accounting figure.
The figure that probably deserves more attention is $518 billion.
That is the figure indicating the scale of the bet. If Claude continues to grow at the pace observed since 2025, Anthropic could become one of the world’s most important technology companies. If revenue slows while compute obligations remain in place, the company will discover just how costly artificial intelligence can be before it becomes profitable.
At more than $2 trillion, future shareholders will therefore not be paying only for Claude.
They will be paying to believe that this growth can catch up with the bill.
