字母榜精选

wrote a column ·

Debunking the AI bubble narrative, Anthropic turns profitable for the first time

For the first time, the income statements of frontier AI model companies are showing positive numbers.

According to The Wall Street Journal, Anthropic is entering a pivotal quarter: the company expects second-quarter 2026 revenue to exceed $10.9 billion, more than double the $4.8 billion reported in the first quarter, and it will post its first-ever quarterly operating profit. Reuters subsequently reported that Anthropic’s expected operating profit for Q2 is approximately $559 million.

Over the past few years, skepticism about an AI bubble has never fully disappeared. While large models have certainly captured immense attention, they are also incredibly capital-intensive: training models costs money, inference services cost money, and expenditures on GPUs, data centers, electricity, and talent—each one is a bottomless pit.

The higher the revenue and the greater the usage, the heavier the costs become. All anyone sees is an endless stream of investment, yet there’s been no clear return to show for it.

This inevitably leads many to wonder: Are frontier model companies truly viable businesses, or are they merely capital black holes kept alive by continuous fundraising?

Now, just ahead of the IPOs of several 'cash-burning' companies, Anthropic has delivered a compelling answer:Once frontier models are genuinely integrated into enterprise workflows, programming tools, and long-horizon agent applications, revenue can potentially outpace costs.

In other words,Large model companies can indeed translate technical capabilities into real commercial outcomes.

Why did Anthropic reach profitability first?

The most immediate reason Anthropic achieved quarterly profitability so quickly is that its revenue growth has been exceptionally rapid.

According to figures reported by The Wall Street Journal, Anthropic generated $4.8 billion in revenue in the first quarter of this year, with second-quarter revenue expected to exceed $10.9 billion.Its revenue more than doubled within a single quarter.

Such a growth rate would be rare for any software company—especially considering that Anthropic is not a light-asset SaaS firm, but rather a cutting-edge AI model company burdened with extremely high costs: it must train models, provide inference services, purchase computing power, and support the high-frequency global usage of Claude.

Remarkably, despite facing an exceptionally steep cost curve, it has managed to achieve an even steeper revenue curve.

This has been at the heart of market skepticism toward large-model companies: it’s not that they lack revenue, but rather that their revenue could easily be devoured by costs.

The more users and API calls they have, the heavier the inference bills; the more powerful the models, the more expensive the training; and the larger the company grows, the greater the investments required in data centers, electricity, chips, and engineering teams.

Anthropic has demonstrated thatin certain high-value use cases, large-model companies can generate revenue that exceeds their costs.

The key reason it was able to reach profitability so quickly lies inClaude's accurate bet on enterprise and programming use cases。

Claude’s growth primarily comes from enterprise customers, developers, programming tools, long-task agents, and automated workflows—scenarios fundamentally different from typical consumer subscriptions.

Ordinary subscription users often employ AI for tasks like drafting emails, researching information, chatting, editing text, or generating images. While user scale can be large, pricing is limited, and usage volume is hard to control. For a subscriber paying $20 per month, heavy daily model usage may not actually be profitable for the company.

But enterprise customers are different.

Enterprises are willing to pay premium prices for stability, permission management, data security, system integration, API access, and workflow automation. Programming scenarios, in particular, offer very direct ROI (return on investment). If a model helps engineers write code, run tests, fix bugs, or understand large codebases, it saves not just ordinary office time—but the time of highly paid engineers.

Anthropic has made its strategic focus in this area very clear; over the past year, it has explicitly steered Claude toward becoming an 'enterprise AI toolkit.'

In August 2025, Anthropic integrated Claude Code into its Team and Enterprise plans, allowing enterprise administrators to purchase premium subscriptions that bundle chat, code generation, and development workflows into a single package.

It simultaneously introduced spending caps, seat management, usage analytics, policy controls, and compliance APIs—enabling enterprises to manage budgets, administer employee permissions, track usage data, and meet audit and compliance requirements.

Within industry-specific applications, finance is the clearest example. In May this year, Anthropic released ten agent templates tailored for financial services, covering high-frequency scenarios such as investment research, valuation, financial statement analysis, KYC, and month-end closing. On the data and tooling front, Claude integrates with market data and research platforms including FactSet, S&P Capital IQ, MSCI, PitchBook, Morningstar, and LSEG, as well as with enterprises’ internal data warehouses, research repositories, and CRM systems.

This is what enterprise AI really looks like:It bundles together models, data, permissions, plugins, workflows, and industry-specific templates—rather than offering just a standalone chat interface.

For enterprises, Claude is a productivity system that can be embedded into R&D, customer support, investment research, data analysis, document processing, and internal workflows. It’s not merely a nice-to-have chat companion—it’s a tool whose value can be directly quantified on the balance sheet.

This also explains why programming and agent use cases are so critical for Anthropic.

The perceived value of general-purpose chat products tends to be diffuse. Users might find them useful, but it’s hard to precisely measure how much tangible benefit they actually deliver. Programming tools are different. Whether code gets written, bugs get fixed, tasks get completed, or engineer productivity improves—these outcomes are far more visible and measurable.

Long-task agents follow a similar logic. They may consume more compute power, but they address problems that align more closely with real enterprise workflows. As long as the task is sufficiently important, customers are willing to pay a premium for them.

It can be said that,Anthropic isn’t chasing revenue through low-cost mass adoption. Instead, it’s targeting higher-value, mission-critical, and more readily monetizable scenarios—converting Claude’s capabilities into higher-quality revenue streams.

This also explains why Anthropic’s revenue structure is positioned to achieve profitability sooner than OpenAI’s.

OpenAI owns the strongest consumer-facing entry point. ChatGPT is one of the most important applications of the entire AI era—but the flip side of a consumer gateway is massive usage costs.

A Q&M Dental-level entry point means Q&M Dental-scale API calls: the more users there are and the deeper they engage, the more staggering the inference bills become. In the short term, a mass-market super-app doesn’t automatically translate into attractive profit margins.

Anthropic’s approach resembles that of an enterprise AI vendor. Unlike ChatGPT, it hasn’t captured a mass-market entry point, but it has earlier and more intensely focused on enterprise clients and developer use cases. Its revenue is more concentrated, its average revenue per customer is higher, and its customers have clearer reasons to pay—making it easier to build stable revenue streams around APIs, team plans, enterprise offerings, programming tools, and agent workflows.

The AI bubble narrative meets its first financial counter-evidence

Anthropic’s profitability this time isn’t just about the company making money; more importantly, it marks the first time the AI bubble narrative has encountered a hard-to-ignore financial counterargument.

In many cases, the AI bubble thesis stems from the difficulty of establishing viable unit economics.

Large language models may be highly capable, but they’re genuinely expensive. Training these models demands enormous computing power, and inference services require ongoing payments. GPUs, electricity, data centers, and research teams—all represent heavy capital expenditures. Worse still, unlike traditional software companies, where marginal costs decrease as user numbers grow, AI companies face rising inference bills as more users make more frequent calls to their models.

If every model upgrade means higher training costs and every increase in users leads to greater inference expenses, then even rapid revenue growth merely amplifies the scale of losses.

This is why many worry that large-model companies could ultimately become capital-market cash vacuums: continuously raising funds, buying more GPUs, expanding data centers, yet struggling to convert technological hype into actual profits.

Leading model companies, including OpenAI and Anthropic, are all telling the same story: models will become the new gateway, agents will reshape workflows, and computing power will emerge as new infrastructure. Yet no matter how grand the narrative, one question remains unavoidable: when will positive numbers finally appear on the income statement?

Now, Anthropic has at least provided a provisional answer.

It hasn’t proven that all AI companies can become profitable, nor does it mean the large-model industry has escaped its capital-burning phase. But it has achieved something critically important:A company competing at the forefront of large language models could indeed outpace cost growth with revenue growth in real commercial markets.

Claims of an AI bubble aren’t entirely unfounded—valuations are too high, capital expenditures are enormous, and many applications still haven’t found sustainable monetization. These are real issues. Even Anthropic’s recent profitability doesn’t guarantee it will remain consistently profitable; future commitments for compute contracts, model training, inference demand, and data center investments could easily erode those profits again.

However, if a leading model company can double its revenue while achieving operating profit, it becomes much harder to argue simplistically that large-model companies can only survive on fundraising and merely burn money into GPUs and parameters.

The more precise question now becomes: which use cases can generate revenue, which customers are willing to pay, which model companies can control costs, and which can translate technical advantages into financial ones.

Anthropic’s profitability is positive news for all large-model companies.

At the very least, it sets a precedent: as long as the use case is sufficiently mission-critical, customers are willing to pay, and the product integrates into real workflows, large language models can indeed generate profit.

As a result, other large-model companies will increasingly align their commercialization strategies along this same path.

Aggressively developing productivity tools and pursuing enterprise services is a viable route. While chatbots can deliver scale and user access, they don’t necessarily translate into strong profit margins.

OpenAI has actually been moving in this direction over the past few months.

Although ChatGPT remains its strongest consumer-facing entry point, OpenAI is no longer content with being just a high-traffic chat product. It is actively rebuilding its enterprise offerings—from dedicated sales for major clients and industry-specific deployments to partnerships with consulting firms and integration into internal corporate systems—with the goal of embedding AI into companies’ core workflows.

OpenAI itself has also publicly emphasized this shift. In April this year, Denise Dresser, OpenAI's Chief Revenue Officer, noted in a corporate AI article that enterprise business already accounts for more than 40% of OpenAI’s revenue and is on track to reach parity with consumer business by the end of 2026.

In May, OpenAI also directly established OpenAI Deployment Company to help businesses actually deploy AI systems. According to OpenAI, it will embed engineers specializing in cutting-edge AI deployment directly into client organizations to jointly transform critical workflows.

This approach closely mirrors Anthropic’s enterprise strategy: model companies can no longer simply wait for customers to figure out how to use their products on their own—they must proactively enter enterprise environments and turn AI into a purchasable solution.

Anthropic, however, has moved faster and with greater focus along this path.

On the eve of its IPO, Anthropic has seized the initiative

The timing of Anthropic’s profitability is highly significant, coming just as several of the most-watched AI companies are approaching their IPOs.

SpaceX has already entered the formal IPO process, aiming to list on Nasdaq as early as mid-June; OpenAI is also accelerating its IPO preparations, with foreign media reporting it could confidentially file documents as early as late May and target a listing as soon as September; although Anthropic has not yet publicly filed its prospectus, it has already hired the Silicon Valley law firm Wilson Sonsini, engaged with investment banks, and is seeking another massive funding round ahead of its anticipated IPO.

All three companies now have valuations in the trillions: SpaceX is targeting an IPO valuation of approximately $1.5 trillion (Wall Street Journal) to $1.75 trillion (Reuters); OpenAI’s current valuation stands at roughly $852 billion (Reuters, May 21); and Anthropic’s latest funding round valued it at around $900 billion (Financial Times, May 21).

AI companies are collectively heading toward public markets, and at this critical juncture,Anthropic’s sudden announcement of a quarterly operating profit is not only positive financial news but also a strategic first move in shaping its IPO narrative.

OpenAI has the strongest consumer-facing entry point, SpaceX has the hardest infrastructure, and Anthropic has delivered what capital markets know best—and can least afford to ignore: profit.

In the primary market, investors can pay for vision, for technology, for the founding team, or even for the 'next platform.' But in the secondary market, stories ultimately must translate into financials. Public-market investors may tolerate losses from high-growth companies, but they prefer those losses to be manageable, with a clear path to profitability and evidence that unit economics can improve as scale increases.

In an IPO narrative, profitability carries significant weight—especially when concerns about an AI bubble persist and markets still worry that large-model companies are merely capital black holes. A leading model company reporting quarterly operating profit first would dramatically differentiate its public listing story.

But this doesn't mean the road ahead is smooth or that all challenges have been resolved.

One quarter of profitability does not equate to permanent victory.Anthropic has demonstrated that a frontier-model company can turn a profit in a single quarter—but it has not yet proven it can generate stable, long-term earnings.

The era of massive AI spending is far from over. SpaceX disclosed that Anthropic has agreed to pay $1.25 billion per month for compute services related to Colossus and Colossus II, under a contract running through May 2029—a figure that is clearly substantial.

This quarter’s profitability resulted from rapid enterprise adoption of Claude, which drove revenue growth that temporarily outpaced cost growth, pushing the company above the breakeven line. However, as future compute contracts, model training, data center investments, and enterprise delivery costs continue to rise, whether this profitability can be sustained remains to be validated over a longer period.

After going public, capital markets won’t focus on just one quarter—they’ll repeatedly ask: Can revenue double again? Can profits be retained? Can the compute bill be controlled? And will enterprise customers renew their contracts?

Anthropic has secured an exceptionally strong opening act,But the ultimate test for cutting-edge model companies isn't earning profits in a single quarter—it's whether they can continue to make money through the next round of model upgrades, the next wave of compute capacity expansion, and the next enterprise procurement cycle. $Artificial Intelligence (LIST2136.US)$$Artificial Intelligence (LIST23586.HK)$$Technology (LIST20763.US)$$Technology (LIST20840.HK)$$Star Tech Companies (LIST2518.US)$

1257 Views