Anthropic and OpenAI take opposite paths to AI profitability

As two of the most valuable private AI companies, OpenAI and Anthropic, move into the public markets, investors are facing a fundamental question about AI economics: Does profitability come from consumer scale or corporate profits? The answer, evident in both companies’ financial reports this week, is enterprise.

OpenAI secretly filed for an IPO, working with Goldman Sachs and Morgan Stanley, aiming to go public as early as September 2026 at a valuation of more than $1 trillion. At the same time, as part of its latest funding round, Anthropic revealed to investors that it expects second-quarter 2026 revenue to be $10.9 billion, more than double the first quarter’s $4.8 billion, and that it expects its first-ever operating profit for the same period to be $559 million. This one-two punch frames the debate over AI profitability, with one company asking the public market to fund several more years of mounting losses, while the other is already poised for a profitable quarter.

Investors continue to look to Amazon for comparison. The company lost billions of dollars over the years before becoming one of the most profitable companies in history, and the template of spending aggressively and capitalizing on platform changes is being used as the basis for the AI bull case. While this analogy is misleading, the numbers behind it show where AI profitability comes from. Amazon accumulated approximately $3 billion in cumulative losses over a six-year period before posting its first annual profit in 2003. OpenAI is expected to accumulate hundreds of billions of dollars in losses before reaching positive cash flow around 2029 or 2030. The scale is 100 times different, the cost structure is different, and the path to profitability, for those who find it, is through the enterprise rather than through consumer adoption.

AI profitability gap between OpenAI and Anthropic

As I wrote last December, OpenAI’s infrastructure ambitions were already threatening the credibility of its revenue base. The company ended with about $20 billion in annual revenue by 2025, while pledging $1.4 trillion in infrastructure spending, which was later revised down to about $600 billion by 2030, according to CNBC. Even the trimmed numbers look dwarfed by what comparable-stage technology companies have historically committed to operating spending.

In April 2026, Anthropic surpassed OpenAI in annual revenue run rate to $30 billion, up from $1 billion 15 months earlier. More important than the execution speed itself is the structure behind it. Approximately 85% of Anthropic’s revenue comes from enterprise and developer customers. OpenAI’s composition is backwards, with about 85% tied to consumer subscriptions for ChatGPT, and about 95% of users paying nothing. OpenAI’s computing spending is expected to reach $121 billion in 2028 alone, with losses of $74 billion in the same year. In contrast, Anthropic projects positive cash flow of $17 billion in 2028 on revenue of $70 billion, with gross margins approaching 77%.

This difference is due to client configuration. Enterprise customers earn 3-5 times more per token than consumer users, have more deterministic query patterns, are cheaper to serve, and have stronger contracts. More than 500 companies currently spend more than $1 million annually on Anthropic’s Claude platform, and eight of the Fortune 10 are customers. That’s the basis of a profitable business. The free tier’s consumer base of 900 million weekly users generates huge inference costs without commensurate returns, but that’s not the case.

What does AI profitability look like if public markets take notice?

The ongoing IPO filing will force the private market to make a long-overdue reckoning. OpenAI is preparing to ask public investors to value the company at more than $1 trillion, but predicts it will lose $14 billion in 2026 and become unprofitable by 2029 or 2030. The listing will test how much faith investors have in the AI boom, with some already questioning whether generative AI can deliver returns commensurate with the trillions of dollars being poured into the field. Sarah Friar, OpenAI’s chief financial officer, said the company was not yet ready for public market scrutiny and expressed concern about the timing.

Anthropic’s IPO story is very different. The second quarter operating profit numbers include model training costs, the expense most often cited as a structural barrier to AI profitability. This milestone comes a full two years earlier than Anthropic told investors last summer. The key proviso is that the operating profit forecast does not include stock-based compensation, which can be significant enough for companies that have raised billions of dollars in private capital to erase margins on a GAAP basis. Additionally, Anthropic may not be able to maintain profitability for the full year due to planned increases in spending on computing and model training. But what this quarter shows is that this business model can generate operating profits at a decent revenue scale. That’s something worth showing to the public market.

The problem with OpenAI is structural. Amazon survived several years of losses because customers paid before suppliers, generating near-consistently positive operating cash flow and keeping the company solvent during the dot-com crash without having to continually raise capital. OpenAI does not have an equivalent mechanism. That burn is related to the operational costs of running inference at scale, and is dependent on the continued availability of private capital at an unprecedented pace. Amazon raised about $8 billion during its existence. OpenAI raised $122 billion in private capital in its latest round in March alone, but needs even more money to move forward with its data center construction plans. Analysts at HSBC estimate there is a $207 billion funding gap for growth plans. Public markets, with their quarterly revenue pacing and intense analyst scrutiny, are less forgiving for those kinds of structural gaps than the sovereign wealth funds and private equity firms that have traditionally funded AI buildouts.

Questions about AI profitability before S-1 filing

As I pointed out in January when I predicted the trajectory of AI in 2026, the era of pure scaling is giving way to something more disciplined. Corporate buyers are demanding measurable returns and selective capital deployment, replacing the build-first attitude of the past three years. That change is reflected in the financial differences between these two companies. Once institutional investors are shown the actual numbers in an S-1 filing, the discrepancy becomes impossible to ignore.

Anthropic’s trajectory for 77% gross margin by 2028 is more akin to the economics of enterprise software than AI infrastructure. A relevant IPO comparison might be Salesforce or ServiceNow rather than Amazon’s retail business. In contrast, OpenAI’s open market litigation is based on the thesis that agent AI, an autonomous system that performs multi-step tasks, becomes the structural unlock that turns consumer reach into revenue for companies. Although this theory is plausible, this change has not yet occurred on a large scale. Meanwhile, Anthropic’s rapid revenue growth has narrowed the scope for OpenAI to establish its desired evaluation criteria before rival applications.

The race for AI profitability ultimately comes down to whether and how quickly the cost of providing intelligence can be lowered below the revenue generated by deploying it. As of this week, one of these companies has demonstrated that it is possible. The other is to convince public market investors that it will happen on a timeline that extends beyond the end of the 20th century. It’s not about Amazon, that’s the bet on the table right now. This is very difficult to price and very important for the future of AI investing.

Source link