With the near perfect electoral record of anti-data center ballot initiatives, degrowth advocates congratulated themselves arguing they helped deflate the AI bubble. Even before the election, the reported valuations of expected IPOs by Anthropic and OpenAI reportedly approaching $2 trillion, raised legitimate questions about whether they set their sights too high. So is AI in a bubble?
It is difficult to say AI is in a bubble when that phrase encompasses such a wide range of businesses from established industrial firms to start ups across a range of industries. Arguing AI is a bubble is too vague and amorphous to mean anything tangible when it is linked to so many processes across industries from old world industrial firms to founders in a garage.
When people argue “AI is a bubble”, what many implicitly mean, without realizing it, what they are really saying is that Anthropic and OpenAI valuations are divorced from the reality of their firm. Here the critics are on firmer but still shaky ground as reasonable people will have differences of opinion on financial valuations especially for still rapidly growing firms with scant if any profitability.
Anthropic reportedly enjoyed 47% growth between the second and third quarter of 2026 with an annualized revenue run rate(ARR) of $74 billion. OpenAI is reportedly slightly beneath that at $50 billion ARR. With revenue continuing to grow in the double digits every quarter for companies already doing $50-74 billion ARR, arguing for valuation of $1-2 trillion is not as crazy as it sounds.
This does not mean these companies are without weaknesses. While Anthropic and OpenAI may want to be treated like standard software as a service allowing higher valuations with high margins, the reality is they are much more like standard industrial due to significant cost of goods and physical asset requirements. Additionally, there are very real challenges to their business from open source or low cost models whether from China or elsewhere. The reality is there is effectively no moat and their product commoditizes rapidly. As those valuations are based upon continued rapid growth and swallowing up other industries, there are more than a few very real risks to Anthropic and OpenAI.
Critics who want to argue AI is a bubble by pointing to OpenAI and Anthropic may have a point but their argument is by no means without holes itself.
However, Anthropic and OpenAI are not the universe of AI in the United States, their industry, or all industries. If we start with simplistic measures of demand for AI, token consumption continues to explode. Depending on how you want to measure token consumption, token output in the United States is up more than ten times in the past year. While OpenAI and Anthropic may face a variety of business pressures to justify their lofty valuations, lack of demand certainly is not one of them. In reality, as any regular user of these products as well as the leaders of these firms will tell you, the biggest challenge is being compute constrained.
Another argument that critics make is that the data center firms financing is rickety. This is even a point that AI leaders and bankers like Goldman Sachs CEO have made noting so many operators rushed into the sector there is going to be a shakeout at some point. So there is some kernel of validity to this argument. However, this overlooks many key points.
For instance, if demand continues to increase significantly, even if at a slower rate than recent trends, even the current build out is unlikely to meet demand meaning these assets will be utilized. Even if AI growth slows dramatically bankrupting some smaller firms, this not only seems unlikely to spread widely, but would represent buying opportunity for larger cash rich firms.
For all the hand wringing about the financial investment being made, an often forgotten problem facing the large established tech firms from Microsoft to Google prior to the AI boom was not their precarious financial position but rather that they were generating too much cash and had too much cash on their balance sheet with no place to put it. Even today, the tech giants are extremely conservatively leveraged with very strong balance sheets. Even if there was a shakeout among data center operators, there is no evidence it would have a material impact on the broader AI or technology sector.
Other sectors are rapidly adopting or looking to adopt AI across a range of industries. Most people think of AI simply as a chatbot or paper writer but the basic technology is actually used in a variety of other ways. Many of these other uses require less compute than text generation but add them up from medicine to agriculture as older industries figure out how to integrate AI technology and it seems rather apparent adoption across the economy remains in its near infancy. Those older industries may not be using enormous compute now but they appear to be ramping up.
One of the more puzzling factors in all of this is tying AI to tangible productivity gains in the economic data. While the US economy has clearly seen an upswing in productivity starting in about 2023, coincidentally shortly after ChatGPT was first released, the straight economic data does not draw a clear line between AI and increased productivity. Some research links the increase in productivity less to labor enhancing technology and more to increased physical capital. One notable finding is that initial adoption of AI appears to reduce productivity before raising it again. This appears to happen because firms are not just adopting a plug and play technology but re-engineering their work flow.
However, a lot of research and data seems to have obvious holes similar to what Alan Greenspan complained of in the early days of the internet. For instance, productivity data shows minimal productivity gains to technology sectors like software. However, ask any coder from the vibe driven to the cubicle dweller at a Fortune 500 and they will tell you at length about how much they can accomplish with Claude Code or their favorite AI coder. Notably, we have seen employment declines in the IT sector and software even as more functions and products rush to market. It is difficult to square the empirical data with the observed reality especially as firms and consumers alike are demanding more AI.
The AI is a bubble argument is not entirely without basis, however, the arguments are at best over confident in their assertions. Demand continues to explode and the finances sector wide remain strong with firms in every sector continuing to think of new ways to integrate AI into their workflow. Like any new and rapidly growing technology, not every firm or use of the technology will exist in five years or even five months, however, the biggest problem facing the industry is how to manage the growth not whether people want the product.