Anthropic : first quarterly profit, 10.9 billion USD in revenue, and valuation at 900 billion

Anthropic : first quarterly profit, $10.9 billion in revenue, and $900 billion valuation

🔎 AI has just crossed a symbolic milestone — but at what cost?

Anthropic has just announced figures that defy all startup logic. The company's first operational profit in its history: $559 million. Projected revenue for Q2 2026: $10.9 billion, representing a 130% increase compared to Q1 ($4.8 billion). And meanwhile, the company is reportedly in negotiations to raise an additional $30 billion at a $900 billion valuation.

These figures, reported by Bloomberg and Dataconomy, have had a bombshell effect in Silicon Valley. For the first time, a major player in generative AI is demonstrating that the model can not only generate revenue, but also deliver operational profit.

Except that critics are already mounting. Analysts point to an accounting artifice behind this "profit" and remind us that compute costs directly threaten Q3. International Business Times emphasizes that this $900 billion valuation — which would surpass that of OpenAI ($852 billion) — relies largely on this narrative of sudden profitability.

The question is no longer whether AI will generate profits, but whether Anthropic's are real or manufactured for a historic fundraising round.

The essentials

• Anthropic reports its first operating profit: $559 million in Q2 2026.
• Revenue explodes by 130%, jumping from $4.8 billion (Q1) to a projected $10.9 billion (Q2).
• The company is negotiating a $30 billion raise at a $900 billion valuation, surpassing OpenAI.
• Analysts dispute the reality of this profit, calling it an accounting artifact tied to the capitalization of R&D expenses.
• Compute costs could wipe out any margin as early as Q3 2026.

Recommended tools

| Hostinger | Web hosting for integrating AI APIs | Starting at €2.99 (May 2026, check on hostinger.com) | AI startups and developers |

The numbers in detail — $559M profit, $10.9B revenue

Anthropic does not formally publish its accounts — it is a private company. The figures come from leaks and internal briefings relayed by Bloomberg.

The jump from $4.8 billion to $10.9 billion in a single quarter is extraordinary. For comparison, very few tech companies have experienced such revenue acceleration outside of a bubble period.

What catches investors' attention is the ratio: $559 million in operating profit on $10.9 billion in revenue, representing an operating margin of about 5.1%. This is modest compared to the margins of Google Cloud (30%+) or AWS (35 %+), but for an AI company spending billions on compute, it is presented as a miracle.

The reality is more nuanced. Operating profit excludes a number of expenses. And this is precisely where the criticism sets in.

Why profitability arrives (seemingly) so quickly

Two factors explain this sudden profitability. The first is the leverage effect of very high-volume enterprise contracts. Anthropic has signed massive agreements, the most notable being with Google Cloud. This historic $200 billion contract locks in infrastructure while guaranteeing predictable recurring revenue.

The second factor is accounting. A significant portion of Anthropic's R&D expenses is capitalized rather than expensed. In plain terms: instead of immediately recording the costs associated with training models like Claude Opus 4.7, the company spreads them over several years as intangible assets.

This practice is legal and common in tech. But it artificially inflates the quarter's operating profit. Quasa.io was particularly vocal on this point, claiming that Anthropic "lies" about its quarterly profit by omitting this accounting reality.

The debate is not futile: if investors value Anthropic at $900 billion based on a profit that disappears when the accounting is adjusted, the bubble is inflating on fragile foundations.

Claude's role in enterprise adoption

Revenue growth is not a mirage. It is driven by massively accelerating enterprise adoption. Claude has become the go-to model for many large enterprises, particularly in regulated sectors (banking, insurance, healthcare).

Why Claude rather than GPT-5.5? The answer comes down to two words: security and compliance. Anthropic has built its positioning around "safe" AI, and enterprises are responding. Claude Opus 4.7 (Adaptive) achieves an agentic score of 94.3, while Claude Sonnet 4.6 sits at 81.4 in agentic and 83 in general — scores that are sufficient for the majority of enterprise use cases.

The acquisition of Stainless for $300 million also played a key role. This acquisition allowed Anthropic to lock down SDK access, making the integration of Claude into enterprise systems smoother while complicating things for OpenAI and Google.

The net result: enterprise contracts signed in Q1 generate their first recurring revenues in Q2, creating a wave effect that partly explains the 130% surge.

This momentum is reinforced by the enterprise joint ventures launched by Anthropic and OpenAI, each injecting $10 billion to deploy AI in SMBs and large corporations. Anthropic is capturing a significant market share there thanks to its reputation for reliability.

The $900 billion valuation: rational or delusional?

A $30 billion raise at a $900 billion valuation would place Anthropic above OpenAI ($852 billion). This is a major symbolic shift in the AI race.

To justify this valuation, you have to look at the multiples. At $10.9 billion in annualized quarterly revenue (i.e., ~$43.6 billion at an annual rate), a 20x multiple yields approximately $870 billion. The multiple is therefore in line with tech norms for a high-growth company — provided the growth is sustained.

The problem: a $900 billion valuation assumes that revenues will continue to grow at this pace for several quarters. However, compute costs are also increasing exponentially. Each new model (Claude Opus 4.7, and then the next generation) costs significantly more to train than the previous one.

International Business Times notes that the profitability narrative makes the valuation "more logical" in the eyes of investors, but that the duration of this profitability remains the true unknown.

If Q3 confirms the profit, the valuation will hold. If the profit disappears under the weight of compute, the $900 billion will look like a bubble peak.

Compute costs: the sword of Damocles for Q3

This is Anthropic's weak point, and every serious analyst mentions it. Training and inference for the latest generation of models cost fortunes.

Claude Opus 4.7 (Adaptive) is Anthropic's flagship model, with an agentic score of 94.3. But every query on this model consumes a massive amount of GPUs. The $200 billion Google Cloud contract provides preferential access to TPUs, but it doesn't erase costs — it smooths them out.

Q2 benefits from an optimization effect: deployed models have been trained, training costs are capitalized, and inference generates revenue. But Q3 will have to absorb the training of the next generation of models, and that bill will be colossal.

Anthropic is not alone in this situation. OpenAI faces the same pressures with GPT-5.5 (agentic score of 98.2) and GPT-5.4 Pro (91.8). Google internalizes part of its costs through its own cloud, giving it a structural advantage. But for an independent player like Anthropic, every training cycle is a financial gamble.

The question for Q3 is not "Will Anthropic be profitable?" but rather "By how much will profit contract?"

Anthropic vs OpenAI vs Google: the AI race in May 2026

Market dynamics have changed in six months. OpenAI was dominating unquestionably a year ago. Today, the landscape is tripolar.

In pure score, OpenAI remains in the lead with GPT-5.5. But Anthropic has gained the advantage on two criteria: operational profitability and enterprise adoption. Google, with Gemini 3 Pro Deep Think (95.4 in agentic, 90 in general) and Gemini 3.1 Pro (87.3 in agentic, 92 in general), benefits from the ecosystem but suffers from a less clear positioning on the "safe AI" segment.

The real shift is strategic. Anthropic has chosen the path of creating autonomous AI agents as the main vector for enterprise revenue. OpenAI is betting on general public and enterprise versatility. Google is playing on vertical integration (Search, Cloud, Android).

Anthropic's profit changes the game: it proves that a "safety-first" model can be profitable, which many skeptics denied just 12 months ago.

Autonomous agents: the real growth engine

Behind the aggregated figures, there is a crucial strategic detail. Anthropic's revenue growth is not driven by the consumer chatbot. It is driven by enterprise deployments of autonomous agents.

An autonomous AI agent, unlike a simple chatbot, executes complex tasks iteratively: data analysis, decision-making, triggering actions in external systems. This consumption model generates revenue per task, not per query. And the volumes are astronomical.

Claude Opus 4.7 (Adaptive) was specifically designed for this use case. Its "adaptive" architecture adjusts the level of reasoning based on the complexity of the task, optimizing inference costs while maintaining quality.

Companies that create their first autonomous AI agents are mostly doing so on Claude today, not because it is the best model in absolute score (GPT-5.5 is above), but because Anthropic's safety framework reduces risks in critical production environments.

It is a fragile competitive advantage — OpenAI and Google are catching up fast — but it explains a significant part of the 130% growth.

AI avatars and emerging new use cases

Beyond autonomous agents, Anthropic is exploring new interaction formats. AI avatars represent an emerging market where Claude could take a strong position thanks to its conversational robustness.

An AI avatar is a visual and vocal representation of a language model, capable of interacting with users in real time. Applications range from customer service to personal assistants, as well as training and education.

Anthropic does not yet dominate this segment — competition is fierce there and lightweight models like Claude Sonnet 4.6 (83 in general) are often sufficient. But Anthropic's safety infrastructure becomes a selling point when avatars interact with the general public, where the risk of going off the rails is at its highest.

This segment remains marginal in the $10.9 billion in Q2 revenue, but it contributes to the diversification narrative that investors want to see to justify a $900 billion valuation.

The accounting critique: why some say it's a mirage

All the figures above are real in the sense that they come from reliable journalistic sources. But the way they are presented is the subject of growing controversy.

Quasa.io published a scathing analysis that boils down to one argument: capitalizing R&D expenses turns a negative result into a positive one. According to this reading, if Anthropic expensed all its training costs immediately, the $559 million operating profit would turn into a loss.

This debate is technical but essential. The capitalization of R&D expenses is governed by accounting standards (ASC 730 in the United States). A company can capitalize costs related to software development if it demonstrates that the product will be profitable. Anthropic considers that its models meet this criterion.

The problem: when a company capitalizes massively, it creates an asset on its balance sheet that will have to be amortized. This amortization will reduce future profits. In other words, the Q2 profit is partly a loan taken from subsequent quarters.

This does not mean that Anthropic is "lying". But it does mean that the headline "first quarterly profit" must be read with a grain of accounting salt.

The implications for the AI market

Whatever one thinks of the accountant, Anthropic's announcement has concrete consequences for the market.

First, the pressure is mounting on OpenAI. If Anthropic, the small "ethical" company, can be profitable, why not OpenAI, which has higher revenues and a flagship model (GPT-5.5) that tops all the rankings? The answer is complex: OpenAI probably spends more on R&D and has a heavier cost structure. But Anthropic's profitability narrative forces OpenAI to show its own numbers.

Next, investors will demand profitability trajectories from all AI startups. The era of "burning cash for growth" will shorten. If Anthropic manages to achieve it after three years, the others no longer have an excuse.

Finally, enterprise customers will feel reassured. The fear that an AI provider might go bankrupt or be acquired was real. A profitable Anthropic at a 900 billion valuation is a long-term partner. This will further accelerate contract signings.

The theoretical context: chance or anthropic necessity?

The name "Anthropic" is not insignificant. It refers to the anthropic principle, a concept in physics and the philosophy of science which postulates that observations of the universe must be compatible with the existence of observers.

In the context of the company, the parallel is interesting. Anthropic fluctuations and the weak anthropic principle suggest that our position as observers biases our perception of phenomena. In the same way, Anthropic's "profitability" could be an artifact of perspective — a bias introduced by the choice of metric (operating profit rather than free cash flow).

Anthropic decision theory, developed by researchers linked to the company, explores how rational agents should act in a multiverse. Without falling into metaphysics, we can see it as a reflection on decision-making under deep uncertainty — exactly the situation Anthropic faces regarding its future compute costs.

The classical Everett model and questions of typicality and "freak" observers remind us that what seems "typical" in a sample can in reality be atypical on the scale of the multiverse. Is Q2 profit "typical" of Anthropic's trajectory, or is it a "freak observer" — a statistical anomaly in an otherwise negative series?

Even research on dark radiation fluctuations and their implications for cosmic warming finds an ironic echo: the "warming" threatening Anthropic is not cosmic but computational, and its "fluctuations" are those of its operating margins.

❌ Common mistakes

Mistake 1: Confusing operating profit and free cash flow

The operating profit of 559 million is not cash in the bank. It excludes financial charges, taxes, and — most importantly — it benefits from the capitalization of R&D costs. Free cash flow is very likely negative. Always check which metric is being used before quoting a profit figure.

Mistake 2: Extrapolating 130% growth over multiple quarters

The jump from 4.8 to 10.9 billion is exceptional because it corresponds to the materialization of contracts signed months earlier. Repeating this pace would mean reaching 25 billion in Q3 and then 57 billion in Q4, which is unrealistic. Growth will mechanically slow down.

Mistake 3: Ignoring Google's role in these results

Anthropic is not succeeding alone. The $200 billion contract with Google Cloud is a pillar of this growth. Without preferential access to Google's infrastructure, Anthropic's margins would be significantly lower. Anthropic's independence is relative.

Mistake 4: Directly comparing Anthropic's and OpenAI's valuations

The two companies have different structures. OpenAI has a capped for-profit structure, Anthropic is a Public Benefit Corporation. Investor rights, exit mechanisms, and governance differ. Comparing valuations without nuance is misleading.

❓ Frequently asked questions

Is Anthropic really profitable?

In operating profit, yes: 559 million in Q2 2026. But this figure benefits from the capitalization of R&D expenses, which artificially inflates the result. In free cash flow, the reality is probably different.

Why such a revenue increase in a single quarter?

The 130% jump (from 4.8 to 10.9 billion) is explained by the materialization of large enterprise contracts signed in Q4 2025 and Q1 2026, combined with the massive adoption of autonomous agents based on Claude Opus 4.7.

Is the 900 billion valuation justified?

At a multiple of ~20x annualized revenues, it is within tech norms for a hyper-growth company. But it relies on the assumption that growth and profitability are maintained, which is uncertain given compute costs.

Is Claude better than GPT-5.5?

In pure agentic score, no: GPT-5.5 (98.2) dominates Claude Opus 4.7 (94.3). But in enterprise adoption, Claude surpasses GPT thanks to its safety positioning and its integration infrastructure.

What are the risks for Q3 2026?

The main risk is the training of the next generation of models. The associated compute costs could reduce or even wipe out the operating profit, especially if revenues naturally slow down after the Q2 jump.

✅ Conclusion

Anthropic has proven that AI can generate operating profit — but the strength of this profit remains debatable, and its sustainability depends on a fragile balance between revenue growth and the explosion of compute costs. If the $900 billion funding round closes, it will be a massive bet that this first profitable quarter is not an anomaly but the start of a new normal. To follow the evolution of these models and their concrete integration, the best approach remains to create your first autonomous AI agent and judge for yourself.