By Akarsh Choudhary
Sam Altman, CEO of OpenAI, recently speculated that the AI industry was in a bubble, stating “investors as a whole are overexcited about AI”, drawing comparisons to the Dotcom bubble of the early 2000s.
Is history repeating itself?
Investors in the 1990s were captivated by the rise of information technology, pouring their money into several companies without caution. A “.com” domain was the barrier to entry. Often, companies’ business models were not tenable, and would fail to generate meaningful returns. Eventually, the bubble popped, companies with market capitalization of millions became worthless, causing market activity to grind to a halt and recession became all but imminent.
Now in 2025 you would be excused for experiencing déjà vu. Investors are not holding back on their bets with AI and AI adjacent companies, which are experiencing ballooning evaluations since the release of ChatGPT – 3.5 in 2022. Capital expenditure is forecasted to double for the likes of Google and Microsoft, rising to a total of $400bn by 2027. Worldwide AI spending is predicted to reach $1.5tn by 2025 according to Gartner.
In the 1950s America and the USSR were engaged in the space race. A victory meant technical and ideological superiority over the other. America emerged victorious after spending roughly $280bn dollars (adjusted for inflation) on the Apollo program. A race for humanity’s next big leap is taking place.
The reason why companies such as Meta and OpenAI are spending billions is in order to build the superior AI, one that could operate at or even beyond the intellectual scope of human intelligence. There is significant monetary gain on the table, and almost as importantly legacy. Like the industrial revolution, history will never forget the impact AI will make on humanity.
Signs of strain
There is certainly good reason to be optimistic about the prospects for AI, with its potential yet to be realised across several industries. It is generally agreed that AI will boost labour productivity, by automating (often tedious and error prone) tasks and it is gradually being incorporated into the workplace with AI calendars, note taking in meetings and data analysis. A McKinsey and Company report found that generative AI could add the equivalent of $2.6 trillion to $4.4 trillion in economic value to the global economy annually, across 63 use cases.
The problem is that growth in the economy and the rise in stock prices are being largely driven by spending and speculation, and it is unclear how any of these companies will make the huge sums of money they are investing back. In fact, an MIT report stated that 95% of companies trying AI have seen no returns from it. In the last 12 months ten lossmaking AI startups reached a collective valuation of $1tn, and although Spotify didn’t make its first full year profit until 2024, it never reached a valuation of this magnitude.
While many AI companies haven’t made any profit they have already begun making billion dollar deals with each other. OpenAI has entered a deal with Nvidia worth $100bn. Nvidia has a stake in AI startup Coreweave which itself partially provides infrastructure needs to OpenAI. The financing has become increasingly complex in these circular deals, inflating valuations and parallels are being drawn to the arrangements that caused the Dotcom bubble.
Another issue lies with smaller AI startups who create what are known as “wrappers”. Essentially, they re-package another companies’ large language model (LLM) with their own design and experience. They take in an input, send it to one of the popular LLMs (ChatGPT, Claude etc), they may add some additional processing and then output the results. Not much additional value is created as they still rely on somebody else’s LLM for the core product, only with some bells and whistles.
Too big to fail
Many economists have been surprised at the resilience of the US economy, despite Donald Trump’s audacious tariffs and active curtailing of the federal government. This has been in part because of the AI boom. A basket of 30 AI related stocks now accounts for 43 percent of the S&P 500’s market capitalization.
It is difficult to say what a potential crash would look like. Layoffs and a slowdown in consumption would likely follow, plunging the USA and its connected markets such as the UK, Canada and Japan into a severe recession. The effects of a crash could take decades to subside.
Regardless of a bubble bursting, income inequality is likely to worsen. Low-skill jobs are under threat from AI disproportionately compared to higher skilled jobs. Although, this could encourage workers to pursue higher skilled and more creative jobs. Just as the industrial revolution created several new crafts and opportunities, new careers could potentially emerge from the ashes of old jobs.
Beyond the bubble
Technology has always persisted through boom and bust cycles. The Dotcom bubble did not stop the rise of the internet; similarly, an AI bubble would not mean the end of AI. We need to prepare for the potential fallout that AI can cause, from economic growth stagnating to mass layoffs.
Instead of relying on AI to drive growth, we need to diversify our attention, for example in lowering the barrier to entry for education so that workers can re-train if AI displaces their jobs. Deals between AI companies should be sufficiently regulated to ensure demand does not appear inflated and markets remain competitive. Workers and consumers alike need to step up pressure on governments to take a more practical approach. The current US administration has done little to limit the power of the tech giants engaging in these risky deals or to even regulate the technology itself. It does not matter if there is an AI bubble, AI is here to stay, but appropriate guardrails are not being set up which could lead to a painful couple of decades.
The views expressed in this article are the author’s own and may not reflect the opinions of The St Andrews Economist.
Image Credit: CEPR.org