The Big Problem Every Artificial Intelligence Company Faces

The AI ​​industry is facing multiple issues at once, and they all feed into the same major, snowballing problem. There is a need for revenue that needs to reach such a ridiculous level ($6 trillion) that it would have to rival or surpass the auto industry. This represents 1% of global gross domestic product (GDP) in 2024, or around $3 trillion. Big labs are facing copyright crises because the technology behind generative AI, large language models, is categorically based on theft, now admitted by top Microsoft officials (via TechCrunch).

Next, the necessary AI data centers, which pollute or destroy infrastructure and drive up prices in the United States, must be quickly built for the Western AI industry to move forward. Predictably, they are either trapped by bureaucracy, planning, material shortages and public pushback after their negative impact was widely spread on social media. A community tracking site, Brockovich AI Data Center Reporting, estimates that fewer than 140 are fully operational, with demand being exceeded according to data center operators and other AI companies.

Despite all these social issues around AI, companies are ultimately only concerned with achieving revenue targets, which seemingly fall from the sky. Right now, based on the information we’ve seen, there is no path to profitability unless almost everything lines up perfectly. Guess what doesn’t line up perfectly? Let’s take a look at some numbers and use the big AI lab Anthropic, which is developing Claude and threatening the world by trying to go public, valued at $2 trillion.

Anthropic, whose financial information in its IPO document, an S-1, was disclosed (via Reuters), showed that it lost $42 billion in 2025. However, $34 billion was realized via write-down, reducing the value of the asset while keeping it on the books. Recent reports of $65 billion in revenue in 2026 are questionable because no one is reporting actual profits. However, a closer look at the numbers reveals that Anthropic – in 2025 – had actual revenue of $4.6 billion, with an operating loss of $8 billion. That means they spent $2.75 for every dollar.

As David Gerard, author of Pivot to AI, points out, these figures have also been slightly “rigged”. A Reuters report details that Anthropic actually records “the total value of marketplace contracts,” or platforms purchasing access to Claude, and then reports the money earned by the platform itself as a “marketing cost.” There is no discernible way to detail the company’s recent dealings outside of the opaque and hard-nosed reporting by some mainstream media outlets, in which it openly told investors that its “adjusted” operating profit would make it profitable.

It’s also worth noting that this figure comes after Anthropic moved to a token billing method. It now charges API, enterprise, and enterprise customers per token, or the way LLM breaks things down during generation. This surprised several companies, including Amazon and Microsoft, who should have told their engineers to slow down their use, even though they had pushed it a few months earlier.

The AI ​​industry has supported its customers since exploding into the market in 2023, subsidizing them with extraordinary investments, but the real cost is now bearing down on it, and it’s not pretty. Anthropic’s top-of-the-line model, Fable 5.1, currently costs $50 per million tokens. Programmers, for example, can burn millions of tokens in a week by having the model generate entire reams of code. Google recently made the decision to limit the free version of Gemini to “Flash-Lite” as it seeks to generate more revenue in an industry that is bleeding cash.

Added to this are questions about circular financing, because Nvidia has a system to finance the purchase of its hardware by its customers. The company, alongside Microsoft, Oracle and other major players in the AI ​​field, participates in circular financing. For example, Microsoft received $30 billion from Anthropic for compute space, but then had to pay the huge sum to Anthropic for access to its Claude models. Meta also joined Microsoft to reduce its reliance on AI.

Questions also arise about Nvidia’s graphics card shipments: where are they? If Nvidia has managed to ship billions or even trillions of hardware, but the data centers are still under development or non-existent, and by the time companies build them, it will be time to upgrade the hardware (ranked to three years in some cases). Where are the data center GPUs? Where does all this money come from? Once again, the famous quote from “All the President’s Men’s” from “Deep Throat” persists: “Follow the money.”