Source | Bohu Finance (bohuFN)

Author | All too well

It truly is a case of 'the rising tide surges the forward wave, and each new generation surpasses the last,' as Anthropic has taken off this year.

The most striking aspect is its astonishing revenue growth rate.

In January this year, Anthropic's annualized revenue was $9 billion, soaring to $19 billion in March and reaching $45 billion by May. This growth rate is extremely rare in human business history.

It's simply breathtaking!

Due to this crazy growth, Anthropic seems poised to achieve quarterly profitability.

According to The Wall Street Journal, Anthropic expects its revenue to exceed $10.9 billion in the second quarter of 2026, more than doubling from $4.8 billion in the first quarter.

More critically, Reuters revealed that Anthropic is set to turn a profit this quarter, with an expected operating profit of $559 million.

Meanwhile, OpenAI reported an adjusted operating profit margin of -122% in the first quarter.

In simple terms, for every $1 in revenue, OpenAI loses $1.22. AI model training, inference computing power, product expansion, and user subsidies continue to consume vast amounts of cash.

Thus, it's easy to understand why capital markets increasingly favor Anthropic.

It is reported that Anthropic is recommending a new round of financing, with its valuation expected to surpass $900 billion.

If successfully completed, it could even exceed OpenAI's estimated valuation of $852 billion, becoming the world's highest-valued AI startup.

How did all this happen?

01 Choosing the Right Track: Making Money as Easy as Breathing

From a user base perspective, Anthropic doesn't hold a significant advantage. Currently, it has approximately 134 million monthly active users, far fewer than ChatGPT's user base.

However, the situation is entirely reversed when it comes to revenue. Counterpoint data shows that in the first quarter of 2026, Anthropic accounted for 31.4% of global large model market revenue, surpassing OpenAI's 29%.

The gap in average revenue per user is even more pronounced. Each active user contributes an average of $16.2 per month to Anthropic, compared to just $2.2 for OpenAI. Annually, this translates to approximately $211 per user for Anthropic versus only $25 for OpenAI—nearly a ninefold difference.

Why such a significant gap between two leading model companies?

Because the two companies have taken different paths.

OpenAI follows an internet product approach: acquire as many users as possible first, then monetize slowly through subscriptions, ecosystems, and platformization. It pursues scale effects.

However, there's an awkward aspect: in the AI era, every user invocation incurs real electricity and computing power costs. More users can mean faster cash burn, especially with a large number of free users.

Anthropic, on the other hand, has taken a different, more 'expensive' path.

It focuses on the enterprise sector, aiming to integrate as quickly as possible into enterprise workflows, targeting scenarios where users are genuinely willing to pay long-term and can generate high-value returns.

Thus, from its inception, Anthropic has emphasized model safety, controllability, and complex task-handling capabilities more than many companies. Its later bet on ultra-long contexts was essentially to enable the model to process longer documents and more complex projects.

Currently, approximately 85% of Anthropic's revenue comes from enterprise and developer clients.

Its most profitable businesses also revolve around enterprises, such as APIs, Claude Code, enterprise agents, and automated workflows. Over 500 companies now spend more than $1 million annually on Claude.

Among these, Claude Code is the most typical product.

There are numerous examples of how useful, efficient, and impressive Claude Code is.

For instance, Stripe had 1,370 engineers use Claude Code to complete a cross-language code migration in four days that would have taken ten people weeks to accomplish.

Boris, the head of Anthropic Claude Code, stated in an interview that he now generates 100% of his code using Claude Code.

As a result, Claude Code has achieved astonishing growth.

Launched in May 2025, its annualized revenue surpassed $1 billion by November of that year; by February 2026, it exceeded $2.5 billion, becoming Anthropic's fastest-growing product in history.

Meanwhile, the number of enterprise clients spending over $1 million annually surpassed 1,000 in a short time, including giants like Uber and Netflix.

This is also Anthropic's ambition: to deeply embed Claude Code into the workflows of tech giants with deep pockets.

This crazy growth is not hard to understand because the return on investment from Claude Code is clear.

Ordinary users often engage with AI for chatting, searching, polishing emails, or generating images. While these demands have large user bases, their value is dispersed. Users may find it useful, but enterprises struggle to precisely measure how much money it creates.

Programming scenarios are different. Whether code is completed, bugs are fixed, projects are launched ahead of schedule, or engineer efficiency improves—all are quantifiable. Once value can be quantified, prices rise rapidly.

This is why many of the most expensive scenarios in the AI industry now focus on programming, finance, law, healthcare, finance, and enterprise agents, as these fields are inherently high-labor-cost industries. AI needs only to replace a little labor for its commercial value to be astonishing.

Anthropic has positioned itself ahead in this regard. By entering more expensive, essential, and easily commercializable scenarios, it converts Claude's capabilities into higher-quality revenue. It has also proven to peers that in certain high-value scenarios, large model companies can generate revenue exceeding costs.

02 Anthropic's Unique Character

Beyond Anthropic's commercial achievements, what has sparked much discussion is its unique character.

Anthropic, founded by seven co-founders, remains remarkably stable with none having left to date. In contrast, OpenAI's leadership has been highly turbulent.

In the AI industry, talent is nearly everything.

How crazy AI companies compete for talent is evident from the millions, tens of millions, or even hundreds of millions in annual salaries offered.

LatePost's 'AI Talent War: Batch-Producing the Era's Influencers' also mentions that in the second half of 2024, Zhou Chang, the core technical lead of Alibaba's Qwen, joined ByteDance with a non-compete agreement, earning tens of millions in total compensation and jumping from P9 to a level equivalent to P11.

Exaggerated, right? But ByteDance quickly reaped rewards. Shortly after Zhou joined, he significantly enhanced Doubao's multimodal foundational model capabilities.

This is why the talent war intensifies: top talent can directly transform model capabilities.

While talent poaching is unstoppable, the challenge for AI companies is how to retain talent.

Anthropic has its methods.

According to media statistics, Anthropic's employee retention rate two years after joining reaches 80%, the highest among leading AI companies at the time. Critically, this data is from 2021-2023, when OpenAI was thriving, and Anthropic hadn't yet truly broken out.

This is also a frequently discussed aspect of Anthropic: a company with a great atmosphere.

First, an extremely strong mission-driven culture.

Anthropic itself is the result of a collective departure of the 'safety faction' from OpenAI. They have always believed that AI must not only focus on scale and speed but also emphasize explainability, controllability, and reliability.

To this end, Anthropic designed a 'long-term interest trust' to avoid being entirely capital-driven in the future. Claude's famous 'constitutional AI' is essentially an extension of this philosophy. Instead of relying on massive human reviews, it provides the model with a set of principles, allowing AI to learn what constitutes safer, more human-value-aligned responses.

It has elevated the 'safety' mission to a religious level.

Second, a highly restrained organizational culture.

Anthropic CEO Dario has said he spends over a third of his time maintaining company culture. When hiring, they don't overly superstition credentials or labels but focus on several questions:

Are you willing to prioritize safety? Do you have a small ego and willingness to collaborate? Can you handle complex problems?

Third, a high degree of transparency.

Dario holds all-hands meetings nearly every two weeks, openly sharing company direction, concerns, and decision-making logic. Employees can also openly discuss or even debate. This high level of transparency fosters strong internal collaboration at Anthropic.

This reminds the author of DeepSeek, which also emphasizes flatness, bottom-up approaches, avoids excessive KPI-drivenness, and isn't in a hurry to commercialize. Many core members are even young people from local universities.

Yet, in this relatively relaxed, idealistic environment, they have created truly stunning products.

DeepSeek even actively benchmarked its model against Anthropic:

Internal employees using V4-Pro for Agentic Coding reported a better experience than Claude Sonnet 4.5, with delivery quality approaching Opus 4.6's non-thinking mode but still lagging behind Opus 4.6 with deep thinking enabled. This, to some extent, indicates that Opus 4.6's thinking mode has become an implicit industry ceiling.

This leading capability and unique company character have made Anthropic increasingly attractive to talent.

For instance, Andrej Karpathy recently announced on X that he is joining Anthropic. He is a highly influential figure in the AI field, having played key roles in both autonomous driving and large model frontier technology development.

His past titles are also impressive: co-founder of OpenAI and former Tesla AI director.

This continuous influx of top talent means Anthropic is forming an increasingly wide moat.

Two things: a group of top talent who believe in safety-first, long-termism, and increasingly deep enterprise workflows.

03 The Cruelty of AI: Even Anthropic Can't Escape

Anthropic now stands in the spotlight.

Revenue is soaring, enterprise clients are flooding in, Claude Code has become one of the AI industry's most profitable products, and profitability is within reach.

However, the AI industry doesn't allow for easy money.

First, the unavoidable issue of computing power.

In May 2026, SpaceX's prospectus disclosed that Anthropic pays $1.25 billion monthly in computing power fees to SpaceX, totaling $15 billion annually and $45 billion over three years.

Oh, and that's just the tip of the iceberg. Anthropic also has a $25 billion investment commitment and 5GW computing capacity with Amazon, along with additional computing power agreements with Broadcom, NVIDIA, and Google.

Anthropic itself admits that planned infrastructure spending may prevent it from maintaining profitability throughout the year.

Because models must continue training; inference scale must keep expanding; as clients multiply, computing power demand will continue to explode. While revenue grows, so does the 'burn rate.'

Second, enterprise clients aren't that stable.

Anthropic's greatest advantage now is its enterprise focus. But many large clients are also competitors.

Take Microsoft, for example. According to reports, Microsoft has internally required some teams to stop using Claude Code and switch to GitHub Copilot CLI. Many Microsoft engineers admit Claude Code's superior experience, but from Microsoft's perspective, it cannot long-term allow its development ecosystem to deeply depend on Anthropic. Because in the AI era, the toolchain itself is strategic.

Third, Anthropic is simply too expensive.

Anthropic's high costs are well-known.

So much so that one reason Microsoft canceled internal Claude Code authorization was the excessive cost due to token-based billing. In other words, even a company with nearly unlimited cloud resources found it too expensive.

There are also reports that Uber burned through its entire annual AI programming tool budget in the first four months of 2026 and was forced to scale back subsequent usage.

The premise of Anthropic's high pricing is that significant capability gaps still exist among frontier models.

But this gap is gradually narrowing.

For instance, Google directly launched a $1 billion AI price war, publicly stating that enterprises could save over $1 billion annually by migrating significant workloads to Gemini 3.5 Flash. OpenAI's Codex is also increasingly similar to Anthropic's Claude Code, but Claude Code often consumes 3 to 4 times more tokens than Codex, making it more expensive to use.

Then there's the unbeatable cost-effectiveness of Chinese models.

On the OpenRouter platform, the usage rate of Chinese models has surged from about 1% in 2024 to over 60% in 2026. Models like DeepSeek, Kimi, and GLM are redefining the industry's cost benchmarks.

If DeepSeek can achieve 80%-90% of the results at a fraction of Claude's price, Anthropic's high customer acquisition costs will collapse. Now, more and more enterprises are adopting multi-model strategies, choosing based on need.

Thus, Anthropic must not only stay ahead but also continue burning money. Anthropic has proven that frontier model companies can make money in a quarter, but more importantly, it must prove it can do so consistently and stably over the long term.

After all, one quarter of profitability doesn't equal permanent victory.

Reference Sources:

1. Overseas Unicorn: Deconstructing Anthropic: The Best AI Company Might Also Be an Organizational Invention

2. LatePost: The AI Talent War: Batch-Producing the Era's Influencers

3. LatePost: V4 Before Launch: DeepSeek's Traits, Organization, and Liang Wenfeng's Unique Goals

4. Letter AI: Shattering the AI Bubble Theory: Anthropic Achieves First Profitability

5. Synced: Microsoft: I Can't Afford to Burn Tokens for Claude Code Either

The cover image and accompanying pictures of the article are owned by their respective copyright holders. If the copyright holder believes that their work is not suitable for public viewing or should not be used without compensation, please contact us promptly, and our platform will make immediate corrections.