Key Takeaways
- • Global AI data centre capex is estimated at around $1 trillion in 2026, towards the upper end of published figures, with Nvidia expecting $3-4 trillion of annual spending by 2030.
- • Each $1 of annual data centre capex requires roughly $1.2 of annual customer value under a 15% return hurdle at each layer, or roughly $6 of customer value over the asset's ten-year life.
- • Assuming 30% annual capex growth to about $2.9 trillion in 2030, cumulative 2026-2030 capex reaches roughly $9 trillion, requiring around $5.4 trillion of annual customer value once the estate matures in the 2033-34 period.
- • Customers currently pay roughly $300 billion a year for AI services; the framework requires this to reach about $4.2 trillion at maturity, implying roughly 45% annual growth for seven years.
- • The framework concludes the bar is high and rising but not without precedent; speed of the required revenue ramp is the one element at the edge of historical precedent.
Summary
The build-out of AI infrastructure, covering chips, networking, land, buildings and power behind services such as ChatGPT, is estimated at around $1 trillion in 2026, according to the primary source's estimates. Nvidia, the largest supplier into the build, expects annual spending of $3-4 trillion by 2030. The central question the analysis poses is how much value AI must create for this spending to make economic sense, a question that can be sized with simple arithmetic and transparent assumptions rather than historical analogies to electrification, railways or the telecoms fibre bubble of 2000.
The framework models a three-layer value chain: data centre owners, typically hyperscale cloud providers, targeting a 15% return on capital at a 30% operating margin; model and software companies such as OpenAI and Anthropic charging about 1.5 times underlying infrastructure cost; and end customers requiring a 15% return on total adoption cost. Under these assumptions, each $1 of annual capex requires around $0.63 of infrastructure revenue, around $0.94 of customer bill, and around $1.19 of gross annual customer value. Over a roughly ten-year asset life, with depreciation, each capex dollar must ultimately be matched by roughly $6 of customer value.
Scaling spending at 30% annually from the $1 trillion base yields cumulative 2026-2030 capex of around $9 trillion. The analysis argues the estate should be judged once mature, around 2033-34, when net capital base falls to roughly $4.5 trillion and the whole estate must generate about $5.4 trillion of customer value annually. This equals roughly 20% of today's global corporate operating profit and about 4.5% of world GDP. The source identifies three potential channels: cost substitution, revenue acceleration and new value pools. Named activities could gross more than $26 trillion, with realistic capture of $2.2-3.7 trillion a year, covering roughly half the requirement. The source notes the framework excludes consumer AI spending, so the base could be conservative.
The source highlights a structural lag: data centres take 18-24 months to build, and enterprise deployment adds further delay, so a 2025 capex dollar may not show up as customer value until 2028 or later. The proposed observable test is whether AI services spending, currently around $300 billion a year by the source's estimate, grows roughly 45% annually for seven years to reach about $4.2 trillion. The source concludes the framework does not resolve whether the spending is rational. Three conditions must hold: capex growth must moderate, capacity must be built and powered on schedule, and customer value must arrive before the financing built around the assets must be rolled. The analysis closes by noting that aggregate gains may coexist with industry-level winners and losers, a theme it links to the authors' growth-gap investment philosophy. The material is marketing content for financial professionals and carries standard investment-risk disclosures.
The framework models a three-layer value chain: data centre owners, typically hyperscale cloud providers, targeting a 15% return on capital at a 30% operating margin; model and software companies such as OpenAI and Anthropic charging about 1.5 times underlying infrastructure cost; and end customers requiring a 15% return on total adoption cost. Under these assumptions, each $1 of annual capex requires around $0.63 of infrastructure revenue, around $0.94 of customer bill, and around $1.19 of gross annual customer value. Over a roughly ten-year asset life, with depreciation, each capex dollar must ultimately be matched by roughly $6 of customer value.
Scaling spending at 30% annually from the $1 trillion base yields cumulative 2026-2030 capex of around $9 trillion. The analysis argues the estate should be judged once mature, around 2033-34, when net capital base falls to roughly $4.5 trillion and the whole estate must generate about $5.4 trillion of customer value annually. This equals roughly 20% of today's global corporate operating profit and about 4.5% of world GDP. The source identifies three potential channels: cost substitution, revenue acceleration and new value pools. Named activities could gross more than $26 trillion, with realistic capture of $2.2-3.7 trillion a year, covering roughly half the requirement. The source notes the framework excludes consumer AI spending, so the base could be conservative.
The source highlights a structural lag: data centres take 18-24 months to build, and enterprise deployment adds further delay, so a 2025 capex dollar may not show up as customer value until 2028 or later. The proposed observable test is whether AI services spending, currently around $300 billion a year by the source's estimate, grows roughly 45% annually for seven years to reach about $4.2 trillion. The source concludes the framework does not resolve whether the spending is rational. Three conditions must hold: capex growth must moderate, capacity must be built and powered on schedule, and customer value must arrive before the financing built around the assets must be rolled. The analysis closes by noting that aggregate gains may coexist with industry-level winners and losers, a theme it links to the authors' growth-gap investment philosophy. The material is marketing content for financial professionals and carries standard investment-risk disclosures.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.