AI bubble in the context of digital infrastructure dynamics
The digital infrastructure and tech markets are witnessing an unprecedented capital deployment cycle in which billions of dollars are flowing into AI compute, energy, and network capacity. The apparent tension between this massive infrastructure buildout, enterprise commitments to allocate 10–40% of budgets to AI, ~USD 20bn in current revenues from AI companies, and ~USD 400bn in expected new AI-related debt issuance has prompted widespread debate over whether this momentum reflects a sustainable technological transformation or a potential bubble.
Following the release of ChatGPT in late 2022, AI has emerged as the dominant theme in equity markets, attracting a disproportionate share of capital into a small group of AI-linked stocks. J.P. Morgan’s AI Index, comprising 38 AI-related firms across technology, retail, and real estate, accounted for approximately 44% of the S&P 500’s total market capitalisation as of November 2025.
Within this group, eight firms — Nvidia, Microsoft, Amazon, Alphabet, Meta, Broadcom, Oracle, and Tesla — represent almost one third of the index, roughly doubling their share from three years earlier. The scale of this concentration recalls the late 1990s, when the largest tech firms accounted for ~25% of the S&P 500 before the dot-com bubble burst.
Valuation dynamics reinforce these concerns. AI-linked equity valuations have moved well beyond historical norms, with the median P/E ratio for AI-exposed firms at around 55x, well above both the technology sector’s three-year average and the long-run S&P 500 range of 18–25x.
Forward valuations for leading AI firms remain elevated, trading at a median 12-month forward P/E of approximately 29x versus around 20x for the broader S&P 500, implying a 50% premium for each expected dollar of earnings, along with greater sensitivity if expectations reset or execution disappoints.
Beyond valuations, the “bubble” debate is increasingly shaped by structural interdependence across the ecosystem. Deep linkages between developers, hyperscalers, hardware suppliers, and infrastructure platforms create an opaque investment environment, where near-term losses are justified by uncertain long-term cash flows and valuation signals may be reinforced by internal financing dynamics, delaying corrective market forces.
In our view, this opacity increases bubble-related risk by delaying market discipline and enabling capital misallocation, particularly in hardware-heavy or highly leveraged segments.
Monetisation risk further amplifies the perceived bubble signal. The gap between investor expectations and visible monetisation largely drives AI bubble concerns, as valuations increasingly discount future revenues that have yet to materialise as clear cash flows.
While adoption is progressing, much of AI’s expected economic value remains embedded in long-term projections. This leaves valuations more exposed if monetisation remains implicit (i.e., customers do not pay directly for AI services or revenues rely mainly on ecosystem effects).
While players further down the value chain may face challenges in monetisation, established players such as hyperscalers and chipmakers benefit from clear, well-embedded monetisation pathways from traditional cloud and IT services, mitigating some of this risk.
Concentration and valuation stretch are rising faster than visible cash flows.
Why the market is questioning sustainability
The market is experiencing a record level of capital investment in compute, power, and network capacity, which could raise risks due to delayed returns and rapid asset depreciation. Combined capex by Amazon, Meta, Alphabet, and Microsoft is estimated at around USD 344bn in 2025, up from USD 200bn in 2024 and nearly three times 2022 levels.
AI-related investments are approaching levels last seen during the dot-com telecom expansion. Concerns are further heightened by the scale of debt financing required to support future investment, with more than USD 1.15tn expected to be financed through private credit, asset-backed securities, and corporate debt by 2030.
While scale does not in itself imply a bubble, history suggests that capital formation at this pace raises downside risk in the presence of valuation stretch, delayed monetisation, and greater reliance on external financing.
Overall, the “bubble” narrative is driven less by a lack of innovation and more by the combination of concentration, valuation stretch, ecosystem opacity, and delayed monetisation, dynamics that can amplify downside in a repricing scenario, particularly for capital-intensive and venture-dependent segments.
Furthermore, realised returns are already showing a clear “layered” structure. Most realised gains continue to flow to companies at the core of the AI value chain, with Nvidia capturing a large share of industry profit growth and Microsoft and Alphabet beginning to see meaningful earnings uplift through cloud and productivity services.
Returns beyond this group remain less consistent, showing that the ecosystem is not progressing uniformly and reflecting the typical variability expected in early-stage technology transitions. AI’s upside is not evenly distributed. Cash flows and ROI concentrate upstream (chipmakers and scaled cloud platforms).
While downstream segments remain more exposed to slower monetisation and tighter funding conditions, the payback profile of AI infrastructure remains long and uncertainty persists. However, profitability is already emerging among leading platforms.
Azure’s AI revenues have grown by nearly 40% year-on-year in 2025, Nvidia continues to post margins above 50%, and leading platforms collectively generate more than USD 110bn in quarterly free cash flow. This suggests value is accruing first to core infrastructure and platform providers, reflecting scale advantages and integrated monetisation pathways, rather than purely speculative capital chasing narrative momentum.
Efficiency gains further support the case that the buildout reflects rational sequencing rather than structurally weak returns. GPU throughput per watt has increased materially, inference costs are falling, and data centre efficiency is improving as workloads mature. This suggests that early capital intensity may partly reflect transitional inefficiencies rather than permanent value destruction.
Enterprise adoption reinforces this view: a Wharton study finds that 74% of enterprises already report positive AI ROI, with 80% expecting further gains, indicating that productivity benefits are materialising beyond experimentation.
All in all, our view is that AI does not resemble a traditional bubble but rather the early infrastructure phase of a structural technological shift, with risks and outcomes likely to be uneven across the value chain.
Importantly, it can be argued that valuation patterns do not fully align with system-wide euphoria. Not all AI-linked firms face the same expectations; investors distinguish between mature platforms and those with growth potential. This points to an increasingly asymmetric risk profile across AI verticals, with differentiation driven by business model strength, cash-flow visibility, and position within the AI value chain.
Taken together, we expect a more nuanced outcome than a classic “boom-and-bust”. Even in a tightening scenario, risk is likely to transmit through slower deployment, greater selectivity, and consolidation, rather than a reversal of structural demand for compute, power, and connectivity.
The key implication for investors is therefore not whether AI demand disappears, but whether capital is deployed with sufficient discipline, conservative ROI underwriting, and a clear understanding of where monetisation is direct versus ecosystem-driven.
As a result, we expect a more selective repricing across the stack, rather than a broad unwind typical of traditional bubbles.
Long-term digital infrastructure investors
For long-term infrastructure investors, the key question is not whether AI demand disappears, but whether absorption and returns arrive later than underwritten. The main risk is overbuilding ahead of visibility, particularly for single-tenant campuses and GPU-dense facilities where exit timing and pricing depend on sustained tenant expansion and capital availability across the ecosystem.
So, is AI a bubble from a digital infrastructure angle?
Taken together, these elements reflect the dual nature of the ongoing AI bubble debate. While speculative patterns and extreme valuations evoke historical parallels with past bubbles, the structural, financial, and operational foundations of today’s AI sector appear stronger and more resilient. The evidence suggests that, rather than a speculative hype, the AI boom may reflect a genuine, albeit uneven, process of technological transformation with profound market implications.
The market may envisage different potential outcomes. In the most pessimistic case, AI investment could prove overextended, echoing past episodes of capital misallocation and triggering a sharp correction in equity markets. A second possibility is a protracted geopolitical race, in which the United States and China double down on AI as a strategic priority, fuelling state-backed spending and fiscal expansion. In our view, AI could ultimately deliver on its transformative promise, driving a new wave of productivity and innovation, albeit with disruptive effects on labour markets and policy frameworks.
