Is the AI Boom a Bubble? The Case For and Against

Is the AI Boom a Bubble? Few questions in markets are more contested right now. Nvidia’s market capitalisation crossed US$5tn in October 2025 after quadrupling since 2023. The five major hyperscalers collectively plan to deploy roughly US$700bn in AI‑related capital expenditure in 2026 alone! Goldman Sachs projects US$7.6 trillion in AI infrastructure investment between 2026 and 2031.

Against that backdrop, the word “bubble” is being used freely, and not only by sceptics. Let’s take a look at 5 reasons why the AI boom might be a bubble, 5 reasons it might not be and our ultimate judgement.

Is The AI Boom A Bubble? Here Are 5 Reasons Why It Might Be

1. The infrastructure‑to‑revenue disconnect is severe

The most striking number in AI investing is not a valuation multiple. It is the gap between what is being spent and what is being earned. By the end of 2025, hyperscalers had committed close to US$400bn in annual AI infrastructure capex, while enterprise AI was generating only about US$100bn in revenue. That is a four‑to‑one mismatch between supply and monetised demand. OpenAI alone has committed to US$1.4 trillion of spending over eight years while reporting just US$13bn in annual revenue. History suggests that infrastructure build‑outs of this scale, particularly when debt‑funded, rarely end gently when revenue catch‑up disappoints.

2. Productivity gains are not showing up in the aggregate data

AI is pitched as a general‑purpose technology that will transform productivity. The macro data are not yet cooperating. A February 2026 NBER study found that 90% of firms reported no impact of AI on workplace productivity, even as executives projected a 1.4% uplift. McKinsey’s 2025 survey found that 94% of companies reported not seeing “significant” value from their AI investments despite near‑universal adoption. This echoes Robert Solow’s famous line that “you can see the computer age everywhere but in the productivity statistics.” The risk is that investors are pricing in productivity gains that either do not arrive or arrive far later and at far lower magnitude than the market assumes.

3. Valuation concentration is at historic extremes.

Nvidia’s weight in the S&P 500 reached about 7.3% at its 2025 peak. The Magnificent Seven collectively account for a share of the index with few precedents outside the late dot‑com era. Price‑to‑sales ratios in AI‑linked sectors are, per Morningstar’s 2026 Outlook, “nearing tech bubble levels.” Concentration of this kind means that any reassessment of AI’s near‑term earnings trajectory can transmit rapidly across diversified portfolios. The market is not just pricing in AI success; it is pricing in AI dominance.

4. The efficiency shock from cheaper models challenges the capex story

One of the least‑discussed risks is that AI itself is becoming cheaper faster than expected. DeepSeek’s January 2025 demonstration that frontier‑level capabilities could be achieved at a fraction of the assumed compute cost was a genuine shock. If inference efficiency continues to improve at pace, the assumption that hyperscaler capex must keep compounding may prove wrong. More efficient models mean less demand for GPU clusters, which directly challenges Nvidia’s revenue trajectory and the economics of the broader AI infrastructure stack.

5. The gains are highly concentrated, not broad‑based

PwC’s April 2026 AI Performance Study found that 74% of AI’s economic value is being captured by just 20% of organisations. The rest remain stuck in pilot mode with modest or unmeasurable returns. If AI’s commercial benefits accrue primarily to a small group of first movers with proprietary data and scale advantages, the total addressable market is smaller than consensus implies. A bubble does not require the technology to fail; it only requires that returns accrue to fewer participants than the market priced in.

5 reasons the AI boom is not a bubble

1. Today’s valuations are not the dot‑com era

The most important counterpoint is that today’s AI leaders generate real profits. Nvidia delivered roughly US$99bn in trailing twelve‑month net income with 53% net margins as of late 2025. The Magnificent Seven trade at about 28x forward earnings, roughly half the multiples assigned to tech at the dot‑com peak. Cisco, the Nvidia of its era, traded at 472x earnings in March 2000. Nvidia trades around 44–47x. These are elevated multiples on real earnings, not speculative revenue projections.

2. Capex is being funded from earnings, not debt

A hallmark of speculative excess is leverage. The dot‑com build‑out was funded heavily by equity issuance and debt that could not be serviced once revenue disappointed. Today’s AI infrastructure build‑out is largely funded from hyperscalers’ free cash flow. Meta spent US$72bn on capex in 2025 from a business generating substantially more in operating cash flow. The willingness and ability to self‑fund investment at this scale, without balance sheet deterioration, is a structural distinction from prior speculative episodes.

3. Early productivity evidence is beginning to emerge

Aggregate data lag, but sector‑level evidence is strengthening. Morgan Stanley found that top‑quartile AI‑exposed industries drove about 1.7 percentage points of US labour productivity growth in 2025. The San Francisco Fed’s 2026 research documents genuine cost savings in call centres, software development, financial management, marketing and healthcare. EY’s 2025 AI Pulse Survey found that 56% of companies reporting positive ROI also reported measurable improvements in financial performance. The lag between general‑purpose technology adoption and aggregate productivity uplift is well‑documented historically.

4. The applications layer is already generating returns

The infrastructure narrative dominates headlines, but the applications layer is where commercial AI is already working. Alphabet’s first‑quarter 2026 results showed that AI Overviews in Search are increasing query volumes rather than cannibalising them. In financial services, JPMorgan’s AI fraud detection and American Express’s real‑time personalisation are delivering measurable returns. In pharmaceuticals, AI‑designed molecules are moving into clinical development. These are revenue‑generating deployments at scale.

5. The counterfactual for not investing is existential

Hyperscalers are not investing because they expect immediate returns. They are investing because not investing is not a viable strategic option. A world where Microsoft does not deploy AI across its enterprise software stack, or where Google does not embed AI into Search, is a world where those businesses lose their competitive positions. The capex is partly defensive. The relevant comparison is not AI capex versus zero; it is AI capex versus competitive obsolescence.

Conclusion: The AI Boom Is Not a Bubble

The AI boom has a lot of hype and has real characteristics of speculative excess. Valuations are stretched, the infrastructure‑to‑revenue gap is large, productivity gains are arriving slowly and unevenly, and the efficiency trajectory of the models poses a genuine risk to the infrastructure investment story. These concerns deserve weight in any serious portfolio assessment.

But a bubble, in the classical sense, requires that the underlying value of the technology is substantially below what the market assumes. That case is getting harder and harder to make. The companies driving the AI boom are profitable, their investments are mostly self‑funded, and early evidence of commercial productivity gains, even if concentrated, is accumulating. The structural comparison to the dot‑com era does not hold at the aggregate level.

The more precise characterisation, in our view, is that AI is a genuinely transformative technology whose near‑term commercial return profile is less certain than current valuations imply. That is a recipe for volatility and selective disappointment, particularly among infrastructure suppliers whose revenue depends on capex assumptions that may prove too optimistic.

It is not a recipe for the kind of systemic collapse that the word “bubble” conjures. Investors would be better served by asking not whether AI is a bubble, but where in the AI value chain the returns are most likely to accrue, and at what price they are prepared to own them.