AI: A Real Revolution Wrapped in Bubble Behaviour
The AI trade in 2026 carries unmistakable echoes of the dot‑com era: explosive valuations, record VC flows and a belief that one technology will rewrite every business model.
But the capital plumbing and earnings base look very different from 1999, creating a regime where parts of AI look priced for disappointment while the underlying platform shift is real and investable.
For investors, the key risk is not that “AI is the next internet” and therefore doomed to implode but that capital has front‑loaded decades of hoped‑for cash flows into a handful of names, leaving a long tail of would‑be winners structurally over‑funded.
What’s Similar: Valuations, Narrative and Funding Manias
On several dimensions, the AI boom does resemble a late‑1990s‑style melt‑up.
Narrative dominance and TAM euphoria
Commentators note that AI is now pitched as the solution for “virtually every aspect of modern life", from operating systems to pharma, with parallels to the dot‑com promise that the internet would transform all commerce.
A Fortune review of 2025 price action points out that dozens of AI‑linked companies have reached hundreds of billions of dollars in valuation despite many having “thin or experimental” revenue lines, reminiscent of 1999’s revenue‑lite dot‑coms.
Funding blow‑out
A data‑driven comparison by Intuition Labs estimates that AI VC funding reached roughly 258.7 billion dollars by 2025, with AI‑related equity market value heavily concentrated in a few firms such as Nvidia and OpenAI.
That same work flags “sky‑high valuations, investment per employee, and market concentration” as classic bubble indicators, even as it acknowledges stronger fundamentals than in the 1990s.
Leverage creeping higher
A BIS bulletin on “Financing the AI boom” notes that AI firms, historically funded by retained earnings and equity, are increasingly turning to debt, raising concerns that shocks could be amplified if expected returns fail to materialise.
The report warns that this shift to leverage introduces vulnerabilities for financial intermediaries and that equity markets may be overestimating future AI cash flows.
These factors give the current cycle a familiar late‑stage feel: abundant capital, stretched multiples in peripheral names and rising balance‑sheet risk.
What’s Different: Earnings, Adoption and Who’s Writing the Cheques
The strongest argument against a straight dot‑com rerun is that today’s AI leaders are profitable incumbents, not pre‑revenue start‑ups.
Profits and cash funding the capex
BlackRock and iShares analyses both stress that today’s AI build‑out is “primarily being funded by established tech giants with massive balance sheets and cash flow", not by loss‑making start‑ups leaning on speculative equity.
iShares notes that AI‑linked tech valuations, while elevated, remain well below dot‑com extremes and that AI data‑centre capex is “building for real demand, today", supported by observable customer pipelines.
Real, large‑scale demand for compute
Demand for AI compute is “growing exponentially with no observed slowdown", driven by large‑language models, recommendation systems and enterprise automation that are already in production, not just pilots.
Reuters reporting around Nvidia’s 2025 earnings underlines this: the company was expected to post revenue growth above 50% year‑on‑year on the back of genuine cloud‑provider and enterprise orders, not promotional deals.
Concentration in a few system‑critical firms
Intellectia’s April 2026 review notes Nvidia still holds about 92% of the data centre GPU market, having reported record fiscal‑2026 revenue of 215.9 billion dollars, up 65% year‑on‑year, with its data centre segment alone generating tens of billions per quarter.
This level of dominance and profitability has no real parallel in 1999, when even the hottest internet names had fragile business models and thin margins.
In short, the core of the AI trade sits on top of cash‑generating oligopolies, not on Pets.com‑style experiments. That doesn’t eliminate bubble risk, but it reshapes it.
Where the Bubble Risk Actually Lives
The more serious risk is not “AI = dot‑com” in aggregate, but a bifurcation between a solid core and a frothy periphery.
Fidelity highlights five bubble indicators to watch — earnings, growth quality, valuations, capex intensity and interest-rate regime — and several are now flashing amber for parts of the AI complex:
Valuation stretch beyond earnings growth
As 2025–26 progressed, multiple commentators flagged that equity markets may be overestimating long‑run AI cash flows relative to even very strong recent earnings, especially in second‑tier hardware and application names.
Capex arms race and debt build‑up
NPR reports that tech companies are pouring billions into AI chips and data centres, with a growing reliance on debt and “risky tactics” to fund infrastructure and M&A, raising the prospect of overcapacity if demand normalises.
The BIS bulletin warns that AI‑related loans could be “as risky as the average private‑credit loan", suggesting lenders are not necessarily pricing unique AI risks appropriately.
Narrative contagion into marginal names
Fortune and CKGSB analyses both point to a long tail of listed and private companies rebranding around AI with little underlying capability, echoing late‑1990s “.com” name pivots.
If a bust comes, history suggests it is this outer ring of story‑driven AI names, not the core cash machines, that will absorb most of the damage.
Systemic Risk: What the IMF and BIS Actually Say
Official institutions see AI as both a stabiliser and a potential amplifier of financial risk.
The IMF’s 2024 Global Financial Stability Report chapter on AI notes that AI can improve risk management, deepen market liquidity and aid surveillance, potentially reducing some traditional stability risks.
At the same time, it warns about herding, procyclicality and model-risk amplification if many market participants rely on similar AI models and datasets, particularly for trading and credit allocation.
The BIS paper on AI financing underscores that rising leverage in AI‑exposed firms could transmit disappointment in AI returns into broader credit stress, especially if banks and private‑credit funds have concentrated exposures.
That combination looks less like 1999’s largely equity‑contained collapse and more like a sector‑specific shock that could propagate via debt and derivatives if not monitored.
Boom, Bust, Then Build‑Out: Likely Path From Here
The weight of current evidence points to a three‑phase sequence rather than a single catastrophic pop:
Boom:
We are in it: rapid multiple expansion for AI leaders, a long tail of speculative capital chasing anything with an “AI” label, and escalating infrastructure capex across cloud and semiconductor supply chains.
Shake‑out / mini‑bust:
As in the dot‑com era, a revenue and cash‑flow reality check is likely to hit peripheral players hardest.
A plateau in enterprise AI adoption or a cyclical downturn could trigger repricing of second‑tier names and privately funded AI start‑ups that lack differentiated IP or distribution.
Long build‑out:
The underlying thesis — that AI will be embedded across software, hardware and services — is consistent with observed adoption and profitability at today’s leaders.
Post‑shake‑out, survivors will own durable infrastructure and data moats, much as post‑2002 survivors of the internet bust built today’s platform layer.
In that sense, AI is likely both the next dot‑com boom and the next dot‑com aftermath: a period of excess that ultimately leaves behind a more concentrated, system‑critical group of firms.
What to Watch: Bubble vs Regime Shift Signals
For a 2026‑era investor, the most informative signals over the next 12–24 months are:
Earnings vs expectations at core AI names (chips, cloud, hyperscalers) — sustained 50–70% revenue growth and margin resilience argue for a regime shift; deceleration with capex still rising would look bubble‑like.
AI‑linked credit growth and loan performance — accelerating AI‑sector leverage with no commensurate cash‑flow improvement is the classic precursor to a boom‑bust dynamic.
Breadth of enterprise adoption — broad, profitable use cases across sectors point to structural change; concentration in a handful of pilots and consumer apps would echo 1999’s shallow monetisation.
On balance, the data today support a real technological supercycle with pockets of classic bubble behaviour, not a pure speculative mirage. The next phase will decide how much of the current market value is future cash flow pulled forward — and how much is simply air waiting to escape.