Is the AI bubble overstated?

The widely repeated claim that the artificial intelligence industry is in a bubble is wrong, or at least premature, according to a long essay that makes the case with company filings and earnings data. The author, who says he is not convinced the bubble exists, argues that the rapid growth of AI companies may be justified and that they still have room to expand.

How much are AI companies spending?

The essay starts with the scale of spending. AI companies have committed to spend at least $3 trillion so far, according to the author, and he estimates they have already spent $1.9 trillion, bringing the total to around $5 trillion. The debt and spending commitments run through complex financial vehicles, which the author says makes it hard to pin down the exact figures. He calls that opacity a serious problem and one of the strongest points in favor of the bubble argument.

Spending that may not keep climbing

Much of that money goes to building data centers, and the author says those costs are largely one-time purchases rather than ongoing operating expenses. Once the data centers are built, he argues, spending should drop sharply.

Are AI companies making profits?

Running AI, he says, is far less expensive than building it. Anthropic's S-1 filing states that the company earns $2.08 for every $1 it spends on compute. Ignoring infrastructure costs, the author writes, AI is very profitable. The filing also indicates that Anthropic has begun to bring in more than its compute and infrastructure investments combined. Companies could also cut costs quickly by ending free access to their products and pausing expensive research, he notes, something that is harder to do with leases and spending commitments.

The real long-term burden, in his view, is debt. He runs rough math: if the industry repaid $5 trillion over 20 years, that would be $250 billion a year. For comparison, he cites Alphabet's revenue of nearly $162 billion in 2019, or about $213 billion adjusted for inflation. He acknowledges the calculation is crude and that the required revenue could exceed $250 billion a year. Some debt is in convertible notes that can be repaid in shares. Anthropic recorded a $72 billion loss in the form of shares due to lenders in the first six months of 2026, which the author says reflects its estimated $1 trillion valuation and suggests it has enough equity to make lenders whole.

Revenue growth and public valuations

Anthropic's reported revenue grew from $41 million in 2023 to $386 million in 2024, then $4.585 billion in 2025 and $16.3 billion in the first six months of 2026, according to its S-1 filing. The author says it is safe to assume other large AI companies, including OpenAI, are seeing similar gains.

He also points to public companies. Alphabet, Meta and NVIDIA have seen their market capitalizations rise sharply, but their price-to-earnings ratios remain in the 20s. In past bubbles, including the dot-com boom, valuations were driven by companies that could not turn a profit or traded at P/E ratios in the hundreds. Alphabet's earnings show the gains are not just from renting out data centers. Google Services revenue rose $37.8 billion, a 12% increase, while Google Cloud grew $15.5 billion, or 36%. Meta's revenue rose by roughly $30 billion from 2023 to 2024 and another $30 billion from 2024 to 2025, the author says. He describes that as real profit from companies monetizing AI rather than investors chasing anything with AI in its name.

OpenAI, which buys a large amount of compute from other firms, is the one unknown in his view. Even if it struggles, he argues, competitors would likely consolidate and compete for market share rather than trigger an industry-wide collapse.

Risks the author still sees

The author does not dismiss the risks. He points to local opposition to data centers, regulatory barriers, the possibility of an autonomous AI system causing serious damage, and the heavy use of complex financing. If revenue growth slows, he warns, that financing could backfire.

He also addresses the idea that AI companies have no lasting advantage. He cites a 2023 leaked Google memo titled ‘We Have No Moat, And Neither Does OpenAI’ and compares the sector to the fast-food chicken sandwich war, in which no chain had a lasting edge yet still earned large profits. Companies can keep customers through brand loyalty and institutional lock-in, he argues, even without a moat.

Near the end, the author suggests that the bubble talk has less to do with money than with a wish to return to the past. ChatGPT launched a little less than four years ago, he notes, and AI has become common in daily life in a short time. Even if a bubble does burst, he says, it would resemble the dot-com crash, in which the underlying technology kept its value.

Este artículo fue escrito con la asistencia de IA.
News Factory APP - noticias agénticas para impulsar tu SEO y AEO.