What is different today, and what is not

Recent earnings from major U.S. hyperscalers reveal both strength and strain. Revenue growth remains robust, but free cash flow has declined as capital expenditure has accelerated. Firms are turning to leasing, special-purpose vehicles, and new debt to sustain the AI arms race ¹ . Analysts estimate that these companies now represent 75 percent of S&P 500 returns, 80 percent of earnings growth, and nearly 90 percent of capital-spending growth since late 2022 ⁴ . 

Global AI infrastructure investment is near US $400 billion this year ¹⁴ . By 2030, cumulative spending could exceed US $5–7 trillion ² . Roughly sixty percent of this capital will go to semiconductors and computing hardware, a quarter to energy infrastructure, and the remainder to construction and land ² . These are extraordinary numbers for a technology still early in commercial adoption. 

The psychology is familiar. In Extraordinary Popular Delusions and the Madness of Crowds, Charles Mackay wrote that people “go mad in herds but recover their senses one by one.” Investors rarely err in recognising an idea’s potential; they err in extrapolating its immediacy and profitability. Kindleberger later formalised this as the “euphoria stage,” when optimism becomes self-reinforcing and credit expansion sustains valuations beyond fundamentals ¹¹ . 

As Robert Shiller observed in Irrational Exuberance, financial booms are not simply valuation events but social epidemics of belief. Rising prices feed stories, and stories feed more buying. Each generation invents a new justification for why old valuation rules no longer apply. In that sense, AI is not just a technological theme but a narrative powerful enough to bend investment behaviour around it. 

As Alphabet’s CEO Sundar Pichai put it, “AI is more profound than fire or electricity.” ¹⁷ Such statements exemplify the “new-era” narrative that Shiller described, a belief that the rules of valuation and prudence no longer apply when confronting a supposedly world-altering innovation. 

Historical rhyme and the capital cycle 

The railway boom of the nineteenth century offers the clearest analogy, indispensable to economic development, but financed too much capacity too quickly. The 1990s telecom and internet build-outs followed a similar pattern. More than US $500 billion was spent laying fibre⁶.

Financial historians William Quinn and John Turner note that every major technological boom shares this pattern ⁶ : innovation, abundant credit, speculative overshoot, and eventual consolidation. Bubbles often leave behind valuable infrastructure. The technology succeeds; the first wave of investors rarely does.

From a macroeconomic perspective, distinguishing between nominal and real investment is critical. Nominal capital expenditure may exaggerate momentum when falling hardware prices mask real volume growth, a dynamic Capital Economics identifies again today as computing costs decline. Corporate spending appears dramatic, but the macro effect is more muted, a reminder that productivity gains typically materialise only after broad adoption.

Capital Economics also estimates that AI could lift productivity growth by 1–3 percentage points annually over the next five years if adoption is broad-based, potentially the most significant acceleration in two decades ²⁰ . Yet these gains depend on efficient capital allocation and the ability of firms to translate technological promise into measurable output.

Edward Chancellor’s Capital Returns explains why ⁵ . High returns attract capital, capacity expands faster than demand, and future returns fall. Firms with the fastest asset growth have historically underperformed those with more disciplined investment. AI leaders are entering that phase now: asset growth is surging, free cash flow is compressing, and competition is escalating as each firm races to defend its perceived moat.

Where the fragility lies 

Balance-sheet stretch

The financing model of the AI boom is evolving from self-funded to leveraged. Until recently, hyperscalers could finance expansion entirely from cash flow. Now, new debt issuance and long-term power-purchase agreements are becoming common ¹ .

Capital allocation now reflect conviction as much as competition.

Executives themselves are openly acknowledging the trade-off between prudence and survival. As Mark Zuckerberg, CEO of Meta Platforms, admitted, “If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously. But what I’d say is I actually think the risk is higher on the other side.” ^{1 5}

Larry Page, Co-founder of Google / Alphabet, was even more blunt: “I’m willing to go bankrupt rather than lose this race.” ¹⁶

Microsoft’s CEO Satya Nadella has echoed this sentiment, declaring that “we are in the race to build the world’s most powerful AI infrastructure and we will not be outspent.” ¹⁸ The competitive impulse to dominate, rather than to optimise returns, has become a defining feature of this phase.

Funding the boom matters

The macroeconomic implications of this cycle will depend on how it is financed. For now, most hyperscalers fund AI capital expenditures from operating cash flow. Free cash flow has narrowed but remains positive, though rising use of leases and structured finance signals the first signs of leverage, a late-cycle dynamic that warrants attention.

If the boom remains self-financed, the risks will stay contained within equity valuations and corporate balance sheets. If it shifts toward borrowing, the effects will propagate through credit markets, tightening financial conditions and amplifying downturn risks when investment slows.

Turning point and the path to a golden age

History shows that financial manias and technological booms tend to follow a recognisable pattern, moving through stages of enthusiasm, disruption, and eventual renewal. As Carlota Perez observed in Technological Revolutions and Financial Capital (2002), each major technological surge follows a four-phase pattern: installation, frenzy, turning point, and deployment.

The early installation phase introduces breakthrough technologies and supporting infrastructure, while the frenzy phase sees finance race ahead of production, driving speculation and overinvestment. The turning point emerges when markets and policymakers force a realignment between financial and real capital, and the deployment phase delivers the lasting productivity and social benefits that define a “golden age.” ¹⁹

Perez’s framework suggests that the AI cycle has moved from installation to frenzy. The build-out of data centres, chips, and power systems has become self-reinforcing as firms compete for dominance, but valuation metrics and financing structures are now stretching beyond immediate profitability. The next turning point, as Perez described, will hinge on how quickly finance recouples with productive investment, a process shaped by funding sources, policy adaptation, and public confidence.

Capital Economics estimates that the current wave of AI investment has added roughly 0.2 percent to global GDP over the past year and could lift growth by 0.3–0.5 percent per year through 2027. Longer term, successful deployment could raise productivity growth by up to 1.5 percentage points annually ²⁰ across advanced economies, a pattern consistent with Perez’s “deployment phase” of broad diffusion.

For policymakers, the turning point is a window to lay the foundations for a sustainable golden age. Aligning regulation, education, and infrastructure with the new technological paradigm helps diffuse benefits beyond early adopters. Failure to adapt, Perez warned, risks entrenching inequality and prolonging stagnation after the frenzy ends.

Asset-heavy transition

The same companies once prized for asset-light scalability are now allocating 20 to 35 percent of revenue ⁵ to physical capital. Rising depreciation and reduced buybacks reflect this structural shift. Historically, sectors that become capital-intensive see declining returns on capital and compressed valuation multiples.

Duration mismatch

Global AI compute demand is projected to reach 200 gigawatts by 2030, roughly triple today’s capacity. To justify the required investment, annual data-centre revenues would need to rise from US $20 billion today to about US $2 trillion, leaving an US $800 billion shortfall ³ . Even with expected efficiency gains, the gap illustrates how far current valuations depend on unproven revenue streams.

Market concentration and valuation heat

The largest seven U.S. firms now account for more than a third of the S&P 500’s market capitalisation. Capital Economics notes that only about a quarter of S&P 500 companies have outperformed the index this year, the narrowest leadership in decades, reinforcing that market breadth has thinned to bubble-like levels. They describe current conditions as “a bubble starting to inflate but unlikely to burst for years,” sustained by genuine technological promise and concentrated capital flows.

Energy and infrastructure bottlenecks

AI-optimised data centres are exceptionally power-intensive. By the end of the decade, their electricity demand could rival that of Japan ¹³ . Utilities are already warning of grid constraints and rising costs. The physical strain on power systems mirrors the financial strain on balance sheets, both symptoms of an overextended investment cycle.

Lessons for Canadian portfolios

Canadian investors are indirectly exposed to this cycle through structural reliance on U.S. mega-caps. To access innovation, domestic portfolios have concentrated in the same narrow set of global growth leaders, effectively import U.S. balance-sheet risk.

AI is likely to reinforce U.S. economic dominance and widen productivity gaps across regions. For Canada, this concentration of technological capacity and market capitalisation heightens reliance on U.S. economic momentum and exposure to its eventual corrections.

This mirrors Canada’s trade dependence on the United States: when one counterparty dominates outcomes, flexibility erodes. If the AI complex stumbles, Canadian portfolios will move in tandem. Managing that correlation, not abandoning the theme, is the challenge.

What advisors will be asked

Is AI a bubble?

Owen Lamont defines a bubble as a self-sustaining rise in prices over time that results in the speculative trading of an obviously overvalued asset ⁷ . It requires three conditions:

1. Overvaluation — prices are broadly acknowledged as excessive;
2. Feedback loop — price gains attract new participants and amplify optimistic narratives;
3. Speculative trading — investors knowingly buy overvalued assets expecting to resell at higher prices.

By that definition, the AI cycle has not yet crossed into a fully developed bubble but clearly exhibits the first two elements. Valuations are stretched and feedback loops are evident: corporate spending drives enthusiasm, enthusiasm drives prices, and rising prices justify still more spending and executives’ own rhetoric reinforces this feedback. When industry leaders declare they are “willing to go bankrupt rather than lose this race” or that the risk of under-investing outweighs the waste of “misspending billions,” exuberance has clearly migrated from markets to boardrooms.

The dot-com era provides a useful parallel. When nominal information and communication technology (ICT) investment collapsed after 2001, real investment fell only modestly because falling equipment costs sustained capital formation. If history rhymes, a future AI correction would likely hit equity valuations far more than real economic activity.

A speculative boom funded by cash flow can deflate with limited macroeconomic fallout, as happened after the dot-com bust, when losses were largely confined to equity holders. A boom financed by borrowing, however, creates liabilities that spread through the financial system, transforming a valuation correction into an economic event. This distinction will determine whether this AI cycle ends as a market correction or a broader contraction.

Portfolio implications

In my opinion, the way to manage these risks is to:

• Maintain AI exposure and include mandates that broaden beyond infrastructure builders to include diversified beneficiaries. 
• Identify portfolio managers that emphasize capital discipline and free-cash-flow sustainability as key selection criteria.
• Retain allocations to non-correlated and alternative strategies as protection against concentrated risk.
• Remember the historical precedent: every major innovation has endured a correction before delivering lasting returns.
• I believe that participating through the correction, not avoiding it, is the path to long-term success.

Closing thoughts

Mackay wrote that “men go mad in herds but recover their senses slowly, and one by one.” ¹⁰ As Shiller later observed, speculative narratives rarely collapse overnight; they deflate as investors reconcile belief with experience.

The recovery this time will take the form of rediscovering valuation discipline after years of growth at any price. AI will reshape productivity across industries, but its financial cycle will follow the same human rhythm as every past innovation: overconfidence, excess, correction, and renewal.

As Perez reminds us, the true test of every technological revolution lies not in its frenzy but in its deployment. The challenge for investors and policymakers alike is to ensure

that the AI boom transitions into a genuine golden age, one where capital and innovation reinforce, rather than exhaust, each other. History suggests that sustained productivity gains will depend on maintaining discipline between financial ambition and real investment capacity.

Corrado Tiralongo (he/him)
Vice President, Asset Allocation & Chief Investment Officer
Canada Life Investment Management