The $5 Trillion Question NVIDIA Can't Answer
NVIDIA reclaimed $5 trillion in market capitalisation on 25 April 2026. The number is extraordinary on its face. It becomes almost numbing when set against the infrastructure wave the company sits inside. McKinsey projects global AI-driven data centre investment reaching $5.2 trillion by 2030, climbing to $7.9 trillion in a high-demand scenario. The figure that once described the entire annual output of a mid-sized economy is now a rounding error in a decade-long capital deployment programme — and one company supplies the compute layer that makes the whole edifice run.
The investment case feels self-evidently strong. Compute demand is insatiable, NVIDIA holds the dominant position, hyperscalers are locked into multi-year capex commitments, and the productivity gains promised by artificial intelligence will justify every dollar spent. The market cap, on this reading, is the least interesting part of the story.
The investment case collapses on contact with the supply chain.
Most of the capital flagged in McKinsey's projections does not flow to technology companies. Power substations, cooling infrastructure, fibre backbones, industrial real estate — these are built by utilities, construction conglomerates, and engineering firms with no meaningful exposure to AI software. When public capital mobilises for AI infrastructure, it arrives as grid upgrades, rezoning approvals, and municipal construction contracts. The financial engineering that inflates tech valuations — buybacks, options compensation, multiple expansion — does not constitute broad economic growth in any meaningful sense.
The productivity argument deserves a fuller hearing. Infrastructure built in the 1990s internet wave required enormous capital deployment that appeared wasteful on virtually every five-year reckoning. The technology nevertheless became load-bearing for the modern economy. A parallel outcome for AI infrastructure is not implausible, and this publication does not dismiss it. The McKinsey projections are directionally credible: the scale of the opportunity is real. The question this publication refuses to defer is who collects the return.
Energy demands are not speculative. A 100-megawatt data centre facility is a permanent load on a regional grid, not a construction event. If McKinsey's figures hold across the decade, electricity demand from AI infrastructure will reshape industrial power markets within years, not decades. Grid stability, residential electricity pricing, and manufacturing competitiveness are all downstream of decisions being made right now by capital allocators optimising for GPU throughput. That the energy costs are largely externalised to utility ratepayers and taxpayers is not an accident of the market — it is a design feature of how public infrastructure investment is being routed through private procurement frameworks.
The geopolitical dimension compounds the structural risk in ways that standard financial analysis does not capture. AI infrastructure is increasingly framed by Western governments as a national security capability, not an economic one. When procurement decisions are filtered through security exemptions, cost-effectiveness becomes secondary to supply chain control. That creates a structural subsidy for incumbents — and a geopolitical risk, since concentration of any critical infrastructure in a single corporate ecosystem introduces fragilities that market valuations systematically underprice.
NVIDIA's $5 trillion valuation is the product of a capital deployment wave that will reshape energy markets, industrial geography, and geopolitical competition for a generation. The investment is real. The productivity gains may be too. The question the market cap cannot answer — who absorbs the costs when the infrastructure is built, and who collects the returns when it delivers — is the one that deserves a public answer before the next trillion-dollar capex cycle begins.
This publication noted the NVIDIA milestone and the McKinsey infrastructure projections. Wire coverage framed the $5T mark as a vindication of AI investment; this article takes the framing at face value and follows the money downstream.