AI has moved from research labs into everyday products at a speed rarely seen in technology. ChatGPT has reached hundreds of millions of users, companies are pouring billions into data centres and chips, and firms from software developers to banks are racing to build AI into their businesses.
But the scale of the investment has raised a harder question for markets: are investors funding a genuine productivity revolution, or are valuations running ahead of reality? With some AI-related companies priced for years of near-perfect growth, comparisons with previous technology booms are becoming harder to ignore.
The Scale of the Bet
In 2024 alone, private investment in AI in the US reached $109.1 billion, nearly twelve times higher than China’s $9.3 billion and twenty-four times the UK’s $4.5 billion. By the end of 2025, global spending on AI is projected to approach $1.5 trillion, according to Gartner.
AI has been called the “new electricity” – a technology so fundamental it promises to power everything. This faith has driven a truly monumental global infrastructure investment binge on a scale few could have predicted.
Tech giants like Microsoft, Google, and Amazon are collectively funnelling tens of billions annually into new data centres and specialised AI chips, a strategic bet that AI will drive future cloud and market dominance. This speculative frenzy is best symbolised by Nvidia, which became a trillion-dollar company in June 2023 and reached a $4 trillion valuation in July 2025 – becoming the first company in history to achieve that milestone.
Its valuation isn’t predicated on consumer product revenue, but on the assumed dominance of its H100 and Blackwell GPUs – the essential mining equipment of the AI gold rush. This has created a dangerous capital gap: a massive disconnect between the vast amounts being spent on building the ecosystem and the actual, measurable revenue being generated by most AI applications outside the chip and cloud providers.
The market is pricing in a return on investment that has not yet materialised for the majority of participants.
The Profit Illusion
The heart of the bubble argument lies not in the quality of the technology, but in the unsustainable unit economics of running it at scale. While training large language models is expensive, the long-term, profit-eroding problem is the cost of the computing power required to serve users – every time someone asks a question.
Running a single complex AI query can be ten to fifteen times more expensive than a traditional search engine query. For companies integrating AI into existing services – customer support bots, productivity suites, enterprise tools – this cost burden is substantial, and it routinely produces disappointing margins even when the technology itself is impressive.
Outside of the infrastructure providers, few firms are currently generating direct, sustainable cash flow from the revolution.
The barrier to entry for new applications is also rapidly shrinking, thanks to a proliferation of powerful open-source models. Many new AI start-ups are, in effect, thin wrappers around a handful of foundation models provided by the giants. If a competitor can achieve eighty percent of your performance for twenty percent of the cost using an open-source alternative, the unique value proposition – the “moat” – for these highly valued start-ups quickly collapses under market pressure.
The Ticking Clock
The promise of AI is clashing increasingly with real-world friction points that could trigger a sharp market pullback.
Energy is the most tangible. A single large data centre can consume the electricity of tens of thousands of homes, and the exponential growth in AI demand clashes directly with global net-zero commitments. In the UK and Europe, the pressure to secure enough power for expanding infrastructure is already pushing operational costs dramatically higher.
Simultaneously, the regulatory environment is tightening. The EU’s AI Act imposes strict governance and transparency requirements, adding significant compliance costs and development delays for companies operating across the bloc – friction that is not yet fully priced into speculative valuations.
The most potent catalyst for a correction, however, would be a disappointing earnings translation. The single greatest threat is if the “AI premium” fails to convert into real, net revenue growth for one of the major cloud players within the next two or three years. Should a giant like Microsoft or Google report that massive infrastructure spending has failed to improve margins – due to spiralling inference costs – investor confidence in the entire ecosystem would be shattered, forcing a sharp revaluation in AI-exposed equities.
Revolution Through Correction
Labelling AI a bubble doesn’t mean the technology is flawed or doomed to fail.
The internet bubble of the late 1990s burst spectacularly, wiping out the majority of start-ups, yet from its ashes rose Amazon and Google, companies that went on to define the modern web economy. The correction cleared out the pretenders and left the field to those with the strongest business models and deepest pockets.
The same pattern could repeat. A market correction would cleanse the speculative froth and expose the AI-washing pretenders. The firms that survive will be those that can turn AI tools into measurable productivity gains, build sustainable models where the cost of inference is manageable, and effectively navigate the enormous energy and infrastructure challenges ahead.
AI is undoubtedly real. So, too, is human overexcitement. Whether we’re witnessing the dawn of a new industrial age or the crest of a speculative wave depends not solely on the technology itself, but on how wisely, and sustainably, it is monetised. Revolutions are inevitable. Bubbles are optional.
