Good morning. By 2030, the AI ecosystem will need about $2.6 trillion in annual end-market spending to support data centers, chips, model development, and investor returns, according to a BCG Institute analysis.
Enterprises would account for $1.8 trillion of that total, followed by consumers at $366 billion, hyperscalers’ internal use at $204 billion, and emerging technologies such as physical AI at $200 billion. The estimates represent global spending, excluding China, and reflect infrastructure costs concentrated mainly in the U.S., where most new capacity is being built.
The $1.8 trillion is a measure of what the AI ecosystem would need from enterprises, rather than a forecast of actual spending.
In one illustrative scenario, BCG estimates that companies could contribute that amount by reallocating $857 billion from existing technology budgets and adding $937 billion in new spending. That would put enterprise tech budgets roughly 13% above International Data Corporation’s IT spending forecast, with AI’s share of those budgets rising from 5% today to 25% in 2030.
The scenario underscores how much the AI industry’s growth depends on corporate spending. For CFOs seeking additional AI investment, that means presenting boards with evidence of productivity gains, cost reductions, new revenue or competitive advantage, said Clark O’Niell, BCG’s global lead for AI, infrastructure and cyber. Some strategic benefits, however, may be difficult to capture in traditional ROI calculations.
“The question isn’t just whether AI delivers an immediate financial return,” he said. “It’s also what the cost of falling behind could be.”
That argument deserves scrutiny, too. Fear of falling behind can be a legitimate strategic consideration, but it shouldn’t become a substitute for a business case.
BCG considers the AI build-out as more resilient than past infrastructure booms, partly because hyperscalers have financed much of their investment through cash flow rather than debt. But the ability to finance infrastructure doesn’t guarantee that it will generate adequate returns.
O’Niell identifies two indicators worth watching. The first is how infrastructure investment is financed, with a sustained shift toward debt signaling greater risk. The second is whether deployed compute generates sufficient revenue. As frontier labs enter public markets, their disclosures could reveal whether revenue is keeping pace with compute costs.
The economics leave limited room for disappointment.
BCG assumes AI chips have an average five-year lifespan and estimates that each gigawatt of capacity coming online in 2026 must generate about $15 billion in revenue to cover its costs. Weaker-than-expected demand could trigger price wars and losses for model providers and data-center operators.
For corporate buyers, those pressures have a practical implication. O’Niell advises most to avoid long-term price-per-token commitments as AI prices continue to fall. Contract flexibility could help CFOs capture future cost reductions.
The $2.6 trillion is achievable, he said, “but it’s not guaranteed.” And even if enterprises provide the spending, the question remains whether they will capture enough of the economic value to justify it.
Sheryl Estrada [email protected]
The post Who pays for AI’s $2.6 trillion build-out? Look to corporate budgets appeared first on Fortune.
