Smashing the AI bubble theory, Anthropic makes a profit for the first time

The profit statement of cutting - edge model companies has finally shown a positive figure.

According to The Wall Street Journal, Anthropic is approaching a critical quarter: the company expects its revenue in the second quarter of 2026 to exceed $10.9 billion, more than doubling from the $4.8 billion in the first quarter, and to achieve quarterly operating profit for the first time. Reuters then reported that Anthropic is expected to have an operating profit of about $559 million in the second quarter.

In the past few years, doubts about the AI bubble have never disappeared. Although large models are very popular, they are also extremely costly: training models requires money, providing inference services requires money, and expenses on GPUs, data centers, electricity, and talent... every aspect is a bottomless pit.

The higher the revenue and the greater the usage, the heavier the cost. People can only see continuous investment but hear no returns for a long time.

This inevitably makes many people wonder: Are cutting - edge model companies really a good business, or are they just capital black holes that can only survive on financing?

Now, just before the IPOs of several "pump" companies, Anthropic has given a strong answer: After cutting - edge models truly enter enterprise, programming, and long - task agent scenarios, it is possible for revenue to outpace costs.

In other words, large model companies can transform their technological capabilities into real business results.

01 Why is Anthropic the first to reach the profit line?

The most direct reason why Anthropic achieved quarterly profit so quickly is that its revenue growth is extremely rapid.

According to the figures disclosed by The Wall Street Journal, Anthropic's revenue in the first quarter of this year was $4.8 billion, and its revenue in the second quarter is expected to exceed $10.9 billion. The revenue more than doubled within a single quarter.

This growth rate is rare for any software company. Moreover, Anthropic is not a light - asset SaaS company but a cutting - edge model company. It bears extremely high costs: training models, providing inference services, purchasing computing power, and supporting the high - frequency calls of Claude globally.

Rarely, while having a very steep cost curve, it has achieved an even steeper revenue curve.

The core of the market's doubts about large model companies in the past lies here: it's not that there is no revenue, but that the revenue might be eaten up by costs.

The more users there are and the more calls are made, the heavier the inference bill; the stronger the model, the more expensive the training; the larger the company scale, the greater the investment in data centers, electricity, chips, and engineering teams.

Anthropic has proven that in certain high - value scenarios, the revenue of large model companies can exceed costs.

The key reason why it reached the profit line so quickly is that Claude has targeted the enterprise and programming scenarios.

The growth of Claude mainly comes from enterprise customers, developers, programming tools, long - task agents, and automated workflows. These scenarios are completely different from ordinary consumer subscriptions.

When ordinary subscribed consumers use AI, they often write emails, search for information, chat, polish text, generate images... The user scale can be large, but the price is limited, and it's difficult to control the usage. For a company, a monthly - subscribed user who pays $20 but makes a large number of model calls every day may not necessarily be profitable.

However, enterprise customers are different.

Enterprises are willing to pay higher prices for stability, permission management, data security, system integration, API calls, and workflow automation. Especially in the programming scenario, the ROI (Return on Investment) is very direct. If a model can help engineers write code, run tests, fix bugs, and understand large code libraries, it saves not ordinary office time but the time of high - paid engineers.

Anthropic's layout in this area is very obvious. In the past year, it has clearly been developing Claude towards an "enterprise AI toolbox".

In August 2025, Anthropic integrated Claude Code into the Team and Enterprise plans. Enterprise administrators can purchase advanced subscriptions with Claude Code, combining chat, code generation, and development processes in the same package.

It also simultaneously launched expenditure caps, seat management, usage analysis, policy settings, and compliance APIs to help enterprises control budgets, manage employee permissions, track usage data, and meet audit and compliance requirements.

In the industry scenario, finance is the most obvious example. In May this year, Anthropic released 10 agent templates for financial services, covering high - frequency scenarios such as investment research, valuation, financial statement analysis, KYC, and monthly settlements. In terms of data and tool ecosystems, Claude can connect to market data and research platforms such as FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, and LSEG, as well as enterprise - owned data warehouses, research libraries, and CRMs.

This is the real form of enterprise AI: packaging models, data, permissions, plugins, workflows, and industry templates together, rather than just a single chat box.

For enterprises, Claude is a productivity system that can be integrated into R & D, customer service, investment research, data analysis, document processing, and internal processes. It is not just an icing - on - the - cake chat partner but a tool that can be used for accounting.

This is why the programming and agent scenarios are so important for Anthropic.

The value of ordinary chat products is relatively scattered. Users may find them useful, but it's difficult to accurately measure how much revenue they actually create. Programming tools are different. Whether the code is written, whether the bugs are fixed, whether the tasks are completed, and whether the engineers' efficiency is improved are all more visible.

The same logic applies to long - task agents. Although they may consume more computing power, the problems they solve are closer to the real enterprise workflows. As long as the tasks are important enough, customers are willing to pay higher prices.

It can be said that Anthropic does not rely on low - price popularization to increase revenue but enters more expensive, more essential, and more easily commercialized scenarios to transform Claude's capabilities into higher - quality revenue.

This also explains why Anthropic's revenue structure is more likely to generate profits earlier than OpenAI's.

OpenAI has the strongest consumer entry point. ChatGPT is one of the most important applications in the AI era, but on the other hand, the consumer entry point means huge usage costs.

A consumer - level entry point means consumer - level calls. The more users there are and the deeper they use, the more terrifying the inference bill becomes. In the short term, a super consumer - level entry point does not necessarily mean a good profit margin.

Anthropic's path is more like that of an enterprise AI supplier. It does not occupy a consumer - level entry point like ChatGPT, but it has captured the enterprise customer and developer scenarios earlier and more intensively. Its revenue is more concentrated, the customer unit price is higher, the reasons for customers to pay are clearer, and it is easier to form stable revenue around APIs, team editions, enterprise editions, programming tools, and agent workflows.

02 The theory of the AI bubble hits a financial counter - evidence for the first time

Anthropic's profit this time is not just about the company making money. More importantly, it has made the theory of the AI bubble hit a hard - to - bypass financial counter - evidence for the first time.

In many cases, the theory of the AI bubble is due to the difficulty of establishing a unit economic model.

Although large models are useful, they are also extremely expensive. Training models requires a huge amount of computing power, and providing inference services requires continuous payment. GPUs, electricity, data centers, and research teams all involve heavy - asset investments. What's more troublesome is that unlike traditional software companies, for AI companies, the more users there are, the higher the marginal cost. The more users and the more frequent the calls, the heavier the inference bill.

If each model upgrade means higher training costs and each user growth means higher inference expenses, then no matter how fast the revenue grows, it will only magnify the scale of losses.

This is why many people worry that large model companies will eventually become capital "pumps" in the capital market: constantly raising funds, constantly buying graphics cards, constantly expanding data centers, but always having difficulty converting the technological popularity into profits.

Cutting - edge model companies, including OpenAI and Anthropic, are all telling the same story: models will become new entry points, agents will reshape workflows, and computing power will become new infrastructure. But no matter how grand the story is, it cannot avoid one question: When will the profit statement show a positive figure?

Now, Anthropic has at least given a phased answer.

It has not proven that all AI companies can make money, nor does it mean that the large model industry has got rid of the money - burning mode. But it has done a very crucial thing: A company in the most cutting - edge model competition can indeed make revenue growth outpace cost growth in the real business market.

The theory of the AI bubble is not completely unfounded. The valuations are too high, the capital expenditures are too large, and many applications have not found stable payment models. These problems are real. Even for Anthropic, this profit does not mean that it will always be profitable. Subsequent computing power contracts, model training, inference requirements, and data center investments may all eat up the profits again.

However, if a cutting - edge model company can achieve operating profit while doubling its revenue, it's hard to simply say that large model companies can only survive on financing and burn money on GPUs and parameters.

The more accurate questions become: Which scenarios can generate profits, which customers are willing to pay, which model companies can control costs, and which companies can transform their technological advantages into financial advantages.

Anthropic's profit this time is good news for all large model companies.

It has at least set a precedent: as long as the scenarios are essential, customers are willing to pay, and products can enter real workflows, large models can also generate profits.

As a result, the commercialization paths of other large model companies will also increasingly incline towards the same route.

It is feasible to vigorously develop productivity tools and take the path of enterprise services. Chatbots can bring scale and entry points, but they may not necessarily bring a good profit statement.

OpenAI has actually been making layouts in this direction in the past few months.

Although ChatGPT is still its strongest consumer entry point, OpenAI is no longer satisfied with just being a chat product with a large user base. It is rebuilding its enterprise - side business: from large - customer sales, industry delivery, cooperation with consulting companies, to enterprise internal system integration, the goal is to make AI truly enter the core workflows of companies.

OpenAI itself has also publicly emphasized this change. In April this year, Denise Dresser, OpenAI's Chief Revenue Officer, mentioned in an enterprise AI article that enterprise business has contributed more than 40% of OpenAI's revenue and is expected to reach the same scale as consumer business by the end of 2026.

In May, OpenAI directly established OpenAI Deployment Company to help enterprises truly deploy AI systems. According to OpenAI, it will embed engineers specializing in cutting - edge AI deployment into customer organizations to transform key workflows with enterprises.

This is very similar to Anthropic's enterprise route: model companies cannot just wait for customers to use their products on their own but should actively enter enterprise sites and turn AI into a purchasable solution.

It's just that Anthropic has gone more purely and more quickly on this path.

03 Anthropic seizes the initiative on the eve of IPO

The timing of Anthropic's profit this time is crucial, occurring at the critical moment when several of the most - watched AI companies are heading towards IPOs.

SpaceX has entered the formal IPO process, aiming to list on the NASDAQ as early as mid - June; OpenAI is also accelerating its preparations for listing. Foreign media reported that it may secretly submit documents as early as late May and strive to list as early as September; although Anthropic has not publicly submitted a prospectus, it has hired the Silicon Valley law firm Wilson Sonsini, contacted investment banks, and is seeking a new round of huge financing in anticipation of the IPO.

In terms of valuation, all three companies have reached the trillion - dollar level: SpaceX's target IPO valuation is about $1.5 trillion (The Wall Street Journal) to $1.75 trillion (Reuters), OpenAI's current valuation is about $852 billion (Reuters on May 21), and Anthropic's latest financing valuation is about $900 billion (Financial Times on May 21).

AI companies are collectively heading towards the public market. At this time, Anthropic suddenly presenting a quarterly operating profit is not only good financial news but also an initiative in the IPO narrative.

OpenAI has the strongest consumer entry point, SpaceX has the most solid infrastructure, and Anthropic has presented the thing that the capital market is most familiar with and hardest to ignore: profit.

In the primary market, investors can pay for visions, technologies, founding teams, and even the "next platform". But in the secondary market, the story must ultimately boil down to finances. Secondary - market investors can accept the losses of high - growth companies, but they hope to see that the losses are controllable, the profit path is clear, and the unit economic model has the opportunity to improve after scale expansion.

In the IPO narrative, profit is very powerful. Especially when the theory of the AI bubble still exists and the market still worries that large model companies are just capital black holes, a cutting - edge model company presenting quarterly operating profit first will make its listing story very different.

However, this does not mean that the road ahead is smooth and everything is fine.

One - quarter profit does not equal permanent victory. Anthropic has only proven that a cutting - edge model company can make money in one quarter, not that it can make money stably in the long term.

The era of AI burning money is not over. SpaceX disclosed that Anthropic has agreed to pay $1.25 billion per month for the computing power services related to Colossus and Colossus II, and the agreement will last until May 2029. This is obviously not a small amount.

Its profit this time is due to the rapid growth of Claude in the enterprise - side market, which made the revenue growth outpace the cost growth in one quarter and pushed the company over the profit line. However, as subsequent computing power contracts, model training, data center investments, and enterprise delivery costs continue to increase, whether this profit line can be maintained requires longer - term verification.

After listing, the capital market will not only look at one quarter. It will repeatedly ask: Can the revenue double again? Can the profit be retained? Can the computing power bill be controlled? Will enterprise customers renew their subscriptions?

Anthropic has had a very good start, but the ultimate proof for a cutting - edge model company is not making money for one quarter but still being able to make money in the next round of model upgrades, the next round of computing power expansion, and the next round of enterprise procurement cycles.

This article is from the WeChat