Why the AI boom can’t be compared to the dotcom bubble
Fears that the AI boom is heading for a dotcom-esque crash have been simmering for quite some time now, as tech stocks wobble and valuations reset.
But new data suggests the comparison is flawed, and risks missing what is actually happening in the market.
According to Redpoint Ventures’ 2026 market update, today’s AI cycle is being driven by real demand, real revenues and real, physical constraints.
At the peak of the dotcom bubble, telecom companies had built huge amounts of fibre capacity that was barely used.
Utilisation was below three per cent, revenues lagged far behind investment, and infrastructure, for the most part, was funded through debt.
But the current AI buildout looks very different. OpenAI and Anthropic are respectively generating over $20bn in annual recurring revenue, while more than 90 per cent of new data centre capacity is already committed before construction has even started.
Hyperscalers are expected to spend over $700bn on infrastructure in 2026 alone, but crucially, that spending is being pulled forward by demand, in a way that it wasn’t for telcos in 2000s.
User adoption also completely dwarfs the early internet era, with ChatGPT reaching around one billion monthly active users in its first four years, compared with roughly 70 million internet users over a similar timeframe back at the turn of the century.
Demand driving the boom
In the late 1990s, firms could cheaply overbuild infrastructure. Laying additional fibre had pretty low marginal cost, which encourages speculative expansion that ultimately overclubbed demand.
Today, the physical constraints of power and connectivity have made that kind of overshot far harder.
Data centres are expensive, hugely power hungry, and difficult to scale without committed customers prior to the fact.
That doesn’t mean there aren’t risks, and markets have already shown signs of strain. The so-called ‘SaaSpocalypse’, for example, wiped roughly 30 per cent off major software stocks, while valuation multiples have tumbled.
Redpoint found public software revenue multiples have dropped to around four times, among the lowest levels in years.
Elsewhere, analysts at Capital Economics argue that if there is a bubble forming, it may be in earnings rather than valuations, with tech profits growing so quickly that sustainability is now the key question.
That shift has been seen in how investors are pricing the sector. Redpoint has estimated that 85 to 95 per cent of a typical software firm’s valuation actually comes from long-term expectations.
There are also emerging fault lines within the market. AI is acting as a tailwind for infrastructure providers, boosting demand for chips and data, while simultaneously disrupting parts of the software stack, where products are more easily replaced.
But even here, the pattern differs from the dotcom era. Rather than hundreds of unprofitable startups chasing speculative growth, capital is concentrated in a smaller group of highly profitable companies with global scale.
That concentration creates its own risks, particularly if spending slows or consolidates, but it is a different kind of vulnerability to the widespread collapse seen in 2000.