Who Is Funding Anthropic: A Complete Breakdown of Anthropic's Major Investors, Funding Sources, and Backers

“Claude hunts down Chinese users” is by far the most popular AI meme on the Chinese internet right now.

Some people have collected Anthropic’s access restrictions and turned them into a role-playing game. Players act as Anthropic auditors and try to spot Chinese users hiding among Claude users.

Turning access to AI applications into an endless game of cat and mouse is clearly not the normal relationship between a legitimate AI company and its users. Yet this strange situation fits Anthropic’s usual style perfectly.

People share Anthropic accounts through many different channels, and people in the AI community know about ways to get around its access restrictions. At a public event, the company’s executives once said, “The region with the largest traffic volume is Singapore.” Everyone there exchanged knowing smiles.

The founder’s past experiences and rumors about Baidu have added even more mystery to the company.

Now, this company is moving toward the stock market with a huge valuation of 2 trillion US dollars and racing to achieve the largest initial public offering (IPO) in history. Yet unverifiable rumors and playful memes have left some careful analysts and members of the public confused. This is not normal. In other words, the risks are extremely high.

What kind of company is Anthropic, really? Does the fact that Chinese users work so hard to get around its restrictions and keep their accounts active provide strong enough evidence to support its sky-high valuation?

More and more information is being shared that presents Anthropic in a positive light. To understand what the company is really worth and whether that value can last, we need to look past the mystery, like Sherlock Holmes, and find the clues that truly matter.

Let’s take a closer look at how Anthropic’s business actually works.

01

Step into the lobby: Who is making all the noise?

When you search for Anthropic online, the first things you see are all kinds of rumors that make the company seem mysterious. Some people discuss how to get access to Claude. Others share difficult programming problems that CodeX has solved, ask about accounts, or exchange tips on how to use the service.

Together, these stories shape our first impression of Anthropic. But is that the whole picture?

If we look more closely at the people actively discussing the company on public social media platforms, we can roughly divide them into three groups.

The first group is programmers, who spread the word about what the product can do. They share experiences of using CodeX for AI programming, compare the performance of different models, and treat Claude as an essential tool in their daily development work. An AI company that most ordinary people cannot access or know much about has attracted so many technical users. This creates a feeling that the product is rare and hard to get.

The second group is people who buy and resell accounts. The more restrictions Anthropic sets, and the harder it becomes to use its services smoothly, the more curious people become. Those who lack programmers’ technical skills may turn to account-selling businesses, purchasing agents, and third-party services. Clearly, none of this money goes to Anthropic.

Why does Anthropic seem unwilling to make money from the Chinese market? Its strict restrictions go against the usual business goal of earning as much money as possible, leading to all kinds of guesses. These guesses attract the third group: people who enjoy gossip.

These people are not heavy AI users and may never have used Claude themselves. But that does not stop them from joining the discussion. Some connect the founder’s past work at Baidu with Anthropic’s business decisions and suggest that the CEO may have developed a strong fear of China because of past experiences.

Others discuss Anthropic’s company culture. They describe it as having the attitude of a tough, distant tech expert who does not care about Chinese users or even major copyright owners such as Disney and Marvel. They also say its content has few limits and that the company seems to be daring others to challenge it.

At this point, we can see that Anthropic’s image on the Chinese internet is mainly shaped by these three groups: a small group of programmers, people who profit from the shortage of accounts and people’s anxiety, and gossipers who are not particularly interested in the product itself.

So the people who talk most excitedly about Anthropic are not the main customers of its business. Their discussions cannot truly explain what kind of company Anthropic is or how it earns money. We cannot understand the company by standing in the lobby and listening to secondhand rumors. We need to go somewhere closer to the truth: the finance room.

02

Enter the finance room

Who is paying for Anthropic?

Money does not lie. Compared with rumors that are difficult to verify, financial figures give us a much clearer picture of how Anthropic’s business works.

As part of its plans for an IPO, Anthropic has made some public figures available that can help us understand the company. The first important clue in its financial records is its revenue of 4.6 billion US dollars. But who paid all that money?

According to Reuters, Anthropic’s revenue in 2025 was close to 4.6 billion US dollars, about 12 times its revenue in the previous year. This remarkable growth is what supports the extremely high valuation given to the company by investors. So who was responsible for the 12-fold increase in revenue?

(Screenshot of Reuters’ report)

More than 80% of Anthropic’s revenue comes from services that charge customers based on how much they use them, mainly through API access and token fees. These customers are usually professional users, such as software developers, or businesses that repeatedly use the models. In the AI industry, this is known as the business-to-business (B2B) market. Less than 20% of revenue comes from subscriptions, such as monthly or yearly plans that individual customers can buy directly through the app.

We now know that businesses and developers are Anthropic’s main source of revenue. What about its costs?

The next important clue is that the two cloud platforms, Amazon and Google, account for about 47% of Anthropic’s total sales.

According to Reuters, cloud platforms accounted for only 11% of Anthropic’s revenue in 2023. By 2025, that share had grown to nearly half. This is an important sign that B2B AI businesses depend heavily on the digital systems their customers already use.

Anthropic’s growth depends not only on how good Claude is, but also on whether the company can sell its models to more customers through major cloud platforms. Imagine that a business already uses AWS or Google Cloud. When it wants to add AI to its systems, buying AI services through a cloud platform it already knows can make it easier to fit the new technology into its existing systems and daily work. That makes the familiar platform a natural first choice.

This creates both advantages and disadvantages for AI model companies.

On the one hand, they can use the sales networks of these two tech giants to reach many business customers. But these sales channels are not free. Anthropic must pay fees to distribute its services through the cloud platforms, as well as pay for computing power.

In 2025, Anthropic’s operating loss exceeded 8 billion US dollars. Its spending on computing and infrastructure reached 7.3 billion dollars, far more than its revenue. This is a familiar business story: the companies selling the shovels often make the first profits during a gold rush.

Another issue is that the tech companies operating these major cloud platforms are often developing their own large AI models or supporting companies they have invested in as they develop AI. For example, Amazon invested in OpenAI, while Google developed Gemini itself. These companies have reasons to limit Anthropic’s competitive position.

One small but important detail: although Claude is often described on the Chinese internet as very expensive, the US market actually accounts for nearly two-thirds of Anthropic’s sales. Some of those users may be Chinese people accessing the service by getting around its restrictions, but their share is likely very small. What really determines Anthropic’s business value is its B2B customers, who use its models repeatedly, and the cloud platforms that help bring Claude into more business settings.

The next question is how loyal Anthropic’s customers really are. Will they continue paying for its services over the long term?

03

A meeting at the top

What makes Anthropic hard to compete with?

How can Anthropic keep its customers and protect its revenue? Simply saying that its models are better than those of OpenAI and Google does not tell us much. The technology industry has many examples of companies with excellent technology and products that still fail as businesses.

What about building strong relationships with cloud platforms and turning their existing customers into AI customers? This advantage can also disappear quickly. Many businesses do not want to rely on a single model, and cloud platforms can help them avoid doing so by offering a wider choice of models.

Another option is to serve a few major customers, build the model into their daily work, and make it expensive or difficult for them to switch to another provider. But this approach is poorly suited to overseas IT markets, where workers are scarce and expensive. It is difficult to lower the extra cost of serving each customer through large-scale operations, so this model cannot easily justify a high price premium.

Which path should Anthropic take? The answer is all of them, with an extra advantage added to each one.

For example, Anthropic treats technology research and development as the main way to compete, while also placing special emphasis on AI safety. The article argues that the company promotes this message by warning of dangers from AI that are difficult to verify, while presenting itself as able to offer ways to prevent those dangers.

For example, last June, Anthropic published a long article claiming that Claude was already doing a large share of internal coding work and that AI systems were beginning to correct and build themselves. The company argued that everyone should slow down AI research and development because otherwise the consequences could be unimaginable. Four months have passed since then, and no such unimaginable consequences have been found. Yet this kind of “AI doomsday” message can attract many followers in Western societies with Christian cultural backgrounds. If people believe AI is very dangerous but cannot stop using it, they may turn to Anthropic because they see it as a safer place to turn.

Anthropic’s partnerships with cloud platforms are not just simple buying and selling arrangements. The company also connects its interests with those of cloud providers through long-term purchasing agreements.

According to Reuters, over the next seven to ten years, Anthropic has committed to paying at least 111.1 billion US dollars to Google, 110 billion US dollars to Amazon, and 31.4 billion US dollars to Microsoft for long-term infrastructure services, regardless of how much it actually uses those services.

These commitments show an aggressive plan to expand computing capacity, even though the amount appears larger than the company’s current business needs. Anthropic is not only betting on business growth. It is also building long-term relationships with cloud providers to reduce conflict with them as it competes in the AI model market.

Securing computing power in advance will increase Anthropic’s financial burden and could make its losses even larger. But it also creates a stronger partnership between Anthropic and the cloud providers. These providers are themselves exposed to the risk of an AI bubble. If the market grows, both sides can benefit. If it fails, both sides may suffer.

This creates a tightly connected cycle. Anthropic needs enough computing power to train its models and compete over the long term. To get that computing power and reach more customers, it must work with major cloud platforms and accept huge costs. Whether it can keep paying those costs depends on whether its product teams can turn expensive computing resources into tools that customers are willing to pay for.

Of course, this cycle can also work in the opposite direction. If competitors catch up with Anthropic’s models, customers switch to other products, and computing costs fail to produce enough revenue, the company’s huge investments could turn from an advantage into a financial burden it cannot escape.

04

Outside the building

After examining all these details, we seem to have uncovered the company’s secret. Its revenue is growing quickly, but it has also made huge commitments to invest in infrastructure. It faces enormous future costs, and there is still much uncertainty about its competitive position.

Anthropic’s business and the way it operates are not difficult to understand. So why is the market full of confusing rumors and stories? Perhaps stories are what give an ordinary business a mysterious image. Tech enthusiasts find hope in these stories, gossipers find excitement, and early investors use them to sell their shares and cash out.

Anthropic itself is not mysterious. Someone has simply added another layer of mystery to a familiar business story and put it up for auction once again. And the place where this new story is being sold is outside the building.