앤트로픽의 수익성이 동네 김밥집과 비교되는 이유 - kmjournal.net

The promise around generative AI has always sounded familiar. Lose money early, scale fast, and eventually unlock high margins. That’s the classic software story. Build once, sell many times, and costs don’t rise as quickly as revenue. But AI is starting to look like a different kind of business. Lately, one blunt comparison has been making the rounds about Anthropic. Its profitability, some say, is worse than a casual Korean eatery like a kimbap shop. It sounds like a joke, but it points to something serious about how AI actually makes money. In AI, More Revenue Often Means More Cost Most tech startups are built on the idea of operating leverage. As revenue grows, fixed costs get spread out, and margins improve. That logic doesn’t always hold in AI. In fact, generative AI is starting to resemble industries where costs scale alongside revenue. Think consulting firms or system integration companies. The more projects they take on, the more people they need. Revenue goes up, but so do expenses. AI services behave in a similar way. More users mean more queries. More queries mean more compute. And more compute means higher costs. So instead of a model where selling more leads to bigger profits, AI often becomes a model where selling more simply increases the bill. The Real Issue Behind Anthropic’s Profit Debate The key metric here is gross margin. That’s what’s left after subtracting the cost of delivering the product. For Anthropic, that cost is largely driven by inference. In simple terms, it’s the compute power required every time a user interacts with the model. And that cost has turned out to be heavier than expected. Anthropic lowered its projected gross margin for 2025 from 50 percent to 40 percent. The main reason is rising inference costs. That’s where the “kimbap shop” comparison comes in. In food service, ingredient costs are often expected to stay below 30 percent of revenue. By that standard, a 40 percent gross margin doesn’t look like the kind of high-margin business people associate with software. The gap between expectation and reality is becoming harder to ignore. It Gets Tougher Below the Gross Margin Line Gross margin is only part of the story. Once you move down to operating profit, the pressure increases. AI companies still have to cover salaries, research, and model training. Those costs don’t disappear as the business grows. So even if a company manages to stabilize its inference costs, it still carries a heavy burden from ongoing development and talent. Anthropic, despite its strong position in the market, is still deep in an investment phase. The business looks big from the outside, but the financial structure hasn’t fully matured. If Anthropic Is Struggling, Others Have It Harder Anthropic’s situation isn’t an outlier. If anything, it’s considered one of the better cases in the industry. The company’s gross margin reportedly improved from negative territory in 2024 to around 40 percent in 2025. That’s a significant shift. Long-term projections suggest it could reach 70 percent by 2027 and achieve positive cash flow by 2028. Still, those are future targets, not current reality. Even top-tier AI companies are talking about what profitability could look like, not what it is today. Even Successful AI Apps Face the Same Pressure This challenge isn’t limited to model developers. AI applications are feeling it too. Take Anysphere, the company behind Cursor, one of the most successful AI coding tools right now. Despite strong growth, its gross margin is reportedly around negative 30 percent. The reason is simple. It relies heavily on external AI models, and those costs add up fast. Reports suggest the company generated about 500 million dollars in revenue but paid around 650 million dollars to Anthropic. That kind of cost structure makes it difficult to turn growth into profit. The Irony of Today’s AI Market There’s a growing sentiment in the market that the most profitable part of the AI boom isn’t AI products themselves. It’s education. Courses, tutorials, and content about how to make money with AI are spreading rapidly. Compared to building and running AI services, they come with far lower costs and fewer risks. It sounds cynical, but it reflects a real imbalance. The hype is high, but the underlying business economics are still catching up. The Real Question: Can AI Scale Profitably? There’s no doubt that generative AI is a major technological shift. But the conversation is changing. It’s no longer just about how fast companies can grow. It’s about whether that growth can translate into sustainable margins. Anthropic’s case highlights the core issue. Is AI a business where scale leads to profit, or one where scale simply increases cost? The market is starting to pay closer attention to the numbers behind the narrative. Inference costs, model fees, training expenses, and talent costs are no longer secondary concerns. Growth and profitability are not the same thing. And for AI, that distinc