Enterprises Would Carry $1.8T of a $2.6T AI Bill by 2030

A BCG Institute analysis dated October 29 says “roughly $2.6 trillion in annual end-market spending will be needed by 2030” to pay for AI infrastructure, and puts $1.8 trillion of that on enterprise customers. The rest splits into $366 billion from consumers, $204 billion from hyperscalers and $200 billion from emerging technologies such as physical AI, BCG says.

BCG estimates installed AI capacity will grow from 17.8 GW in 2025 to 120.3 GW in 2030. In 2030, hyperscaler capex of $1.15 trillion compares with gross debt issuance of $390 billion in a BCG exhibit, so debt equals about 34% of that year’s capex by our arithmetic. BCG calls its figures a revenue requirement, “not a revenue forecast”.

Business Pill · Usage-Based Pricing

A short Business Engineer explainer on usage-based pricing, where customers pay per use. It teaches the general idea only and does not discuss BCG’s analysis.

How BCG Builds the $2.6 Trillion

BCG says it built “a bottom-up cost model which follows the money through three layers: data centers, model companies, and the customers who ultimately pay”. Its figures are annual, global excluding China, and in US dollars.

At the data center layer, BCG puts the requirement, depreciation plus return to investors plus operating costs, at $2.27 trillion in 2030. Adding model-company costs takes it to $2.32 trillion, and a 10% operating margin on external customers to $2.55 trillion, which BCG rounds to $2.6 trillion. BCG says the 10% “is illustrative, not a prediction”.

BCG estimates that a gigawatt of capacity online in 2026 needs to earn $15 billion a year to cover its costs, including the cost of debt and a return for equity investors. A BCG exhibit puts the 2026 cost of building 1 GW at $42.1 billion, including core IT and the data center shell.

What Was Posted on X

This post on 8 November quoted BCG’s analysis, including its estimate that installed AI capacity will grow from 17.8 GW in 2025 to 120.3 GW in 2030. The figures in this piece are from BCG’s report itself, dated 29 October.

Who BCG Says Pays

Enterprises carry $1.8 trillion, about 70% of the $2.55 trillion requirement by our arithmetic. BCG says that in its illustrative example $857 billion is reallocated to AI from existing technology budgets and $937 billion would need to be added, “just 13% of the existing IDC IT spending forecast”.

BCG estimates annual AI spending at about $21,200 for an onshore tech worker at a tech company, falling to about $900 for an offshore knowledge worker at a nontech company. For the $366 billion consumer share, BCG says “direct subscriptions are secondary” and that “Better-targeted advertising and agent-mediated commerce could monetize AI through far larger existing spending pools”.

The Debt Question

BCG says the five hyperscalers driving the build-out, Amazon, Alphabet, Meta, Microsoft and Oracle, expect to commit about $4.6 trillion in AI capex between 2022 and 2030, “nearly $3.9 trillion of it after 2025”. BCG says they “have funded the current build largely from internally generated cash”, and that this suggests the build-out “may be more resilient than prior booms”.

Our recent coverage tracks the debt side: hyperscaler bond issuance of $296 billion this year, against Goldman Sachs’s $250 billion forecast, credit default swap spreads on Nvidia, Oracle and CoreWeave, and Apollo’s $75 billion in chip financing deals. BCG’s 2030 exhibit and the 2026 bond tally are different measures and different years.

BCG’s $42.1 billion per GW covers chips, servers, shell, land, fiber and owner-paid power. Anthropic’s $3.4 billion purchase of 1.7 GW of gas generators works out to about $2.0 billion per GW of generation equipment alone. The two cover different scopes and are not like-for-like.

The Structural Read

As we read the model, the $2.6 trillion is what the system would need to earn, not what BCG expects it to earn, so the payer split is a test of plausibility rather than a forecast.

As we read the enterprise share, about 70% of the requirement rests on companies reallocating or adding technology budgets, which makes enterprise AI adoption the variable the rest of the model depends on.

As we read BCG’s debt exhibit next to our earlier coverage of 2026 bond issuance, the two describe different years and measures, so they show the direction of borrowing rather than a single trend line.

Three Implications

FOR ENTERPRISES BCG puts $1.8 trillion of the 2030 requirement on enterprise customers, with $937 billion of it new budget in its illustrative example.

FOR HYPERSCALERS BCG says the five largest expect about $4.6 trillion in AI capex between 2022 and 2030, nearly $3.9 trillion of it after 2025.

FOR WHAT TO WATCH BCG lists nine indicators to monitor, including intelligence demand elasticity, AI chip utilization and the emergence of new use cases.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Capex Map and the State of the Supercycle.

The essay frames its analysis as “How the money flows, where it cracks, what it buys, who can absorb it, and what happens when revenue can’t catch up.”

As we read it, BCG’s model answers the last of those questions with a payer split; the framework’s other questions are not tested by the figures in this piece.

What Is Not Established

BCG describes its figures as a revenue requirement, not a forecast, and its 10% margin as illustrative. The payer split is based on expert interviews and BCG’s own models. The $1.15 trillion and $390 billion come from a BCG exhibit; we did not see the underlying annual series. The model excludes China. The post on X that drew attention to the report on November 8 quotes BCG; it adds no figures of its own.

The Bottom Line

BCG’s analysis says AI infrastructure needs about $2.6 trillion a year in end-market spending by 2030, with $1.8 trillion of it from enterprises. Its 2030 exhibit shows hyperscaler capex of $1.15 trillion against $390 billion of gross debt issuance, about 34% by our arithmetic, and BCG says hyperscalers have funded the build so far largely from internal cash.