Few numbers rarely seen in the AI industry have appeared. Anthropic, the U.S. AI company behind the Claude chatbot, is expected to post revenue of $10.9 billion in the second quarter of this year, from April to June. That is the amount reported by the Wall Street Journal based on materials sent to investors.
The point to watch is not the revenue figure itself. Just the previous quarter, revenue was $4.8 billion. In other words, it more than doubled in just three months. At this pace, Anthropic would record its first quarterly operating profit since its founding. Expected operating profit: $559 million.
The industry that used to lose money as it made more money, that formula has been broken.
Until now, the AI industry has followed a strange pattern. As users increased and sales grew, companies took on bigger losses. Every chatbot response required massive computing power, and the electricity and equipment costs needed to run that compute were higher than the fees charged to users.
Put simply, it was like a restaurant that lost more money the more customers it served. The ingredients cost more than the price of the food. So bringing in more customers was not necessarily a good thing.
Anthropic’s latest profit outlook directly challenges that formula. The key is on the cost side, not the revenue side. In the first quarter alone, the company spent 71 cents in computing costs to earn every $1 of revenue. In the second quarter, that figure is expected to fall to 56 cents. That means the money left over from every dollar rises from 29 cents to 44 cents.
A 15-cent difference may sound small. But at a scale of more than $10 billion in revenue, 15 cents separates billions of dollars in profit and loss. This is the decisive reason the company moved from loss to profit.
So where did the revenue come from? Not from the kind of chatbot that people usually imagine. The main driver of Anthropic’s explosive growth was Claude Code, a coding tool used by software developers.
It is a tool that lets developers tell AI, “Build this feature,” and the AI actually writes the code. For companies, that means faster work from developers and lower costs. Because the benefit is clear, businesses are willing to pay.
This distinction matters. A free chatbot used by ordinary consumers does not directly generate revenue for the company, no matter how much it is used. By contrast, AI tools used in business operations bring in monthly subscription fees. As more companies attach AI to tasks such as security checks and data analysis, Anthropic’s revenue has become anchored not around “users,” but around “paying customers.”
This means AI has moved beyond the stage of being a novelty to try once, and into the stage of being a work tool that companies pay for and deploy. Anthropic’s profit is the first scorecard showing that transition.
The good news, however, should not be celebrated without caution. There is a real chance this profit could last only one quarter.
The reason became clear in May 20 filing documents submitted to securities regulators by SpaceX, led by Elon Musk. Anthropic agreed to rent computing power needed for AI inference from SpaceX and pay $1.25 billion per month through May 2029. That amounts to about $15 billion a year, more than 20 trillion won.
This deal is a contract to lease SpaceX’s data center, Colossus, in its entirety. It includes more than 300,000 kilowatts of power and over 220,000 high-performance graphics processing units (GPUs). Such facilities are needed to run AI faster and more reliably, but they come with a heavy bill.
Anthropic itself is being cautious. It expects annual profitability only after 2028. That is because it must keep pouring money into larger models and more infrastructure. This second-quarter profit looks more like the result of revenue briefly pulling ahead just before costs begin to balloon in earnest.
There is also something for enterprise customers to watch closely. The key question is whether response speed and processing capacity can remain stable as usage rises. Whether the investment in infrastructure can support that promise will determine if the profit can last.
This is not news to be dismissed as the earnings report of a single American company. The implications for Korean industry and investors are clear.
The most obvious shift is that the battleground in AI is moving. Until now, AI competition often looked like a race over who could build the smarter model. What Anthropic’s profit shows is that the real issue is different: who can control costs while securing paying customers.
For Korean AI companies and those investing in the sector, that is why cost structure and enterprise customer acquisition should be examined before model performance announcements.
The potential opened by the enterprise AI market is also significant. Rather than flashy chatbots aimed at the general public, tools that penetrate real work such as development, security, and analytics are what actually generate revenue. When Korean companies consider adopting AI, it is more practical to judge it not by “because it is new,” but by “which tasks does it reduce costs in?”
Even the way investment excitement is viewed needs to be recalibrated. Anthropic is raising funds at a valuation approaching $900 billion and is reportedly considering an IPO within the year.
But the shadow of enormous computing costs still follows it. Rather than getting swept up in the numbers behind one profitable quarter, the more sober question is whether that profit can continue into the next quarter.
Anthropic’s profit is the first proof that AI can indeed become a money-making business. At the same time, it shows that this business is still walking on thin ice.
The real competition in the AI era will be determined not by who has the smartest technology, but by who has the business model that can survive the longest.