BCG Puts AI Buildout Price at $2.6 Trillion a Year by 2030

Global AI spending outside China must reach roughly $2.6 trillion annually by 2030 to sustain data centers, chips, model development and investor returns, according to estimates from the BCG Institute. That figure is not a forecast. It is the sum BCG calculates the ecosystem needs to absorb each year for the buildout to pay for itself.

The enterprise segment carries most of the load. BCG allocates $1.8 trillion of the total to enterprises, followed by consumers at $366 billion, hyperscalers' internal use at $204 billion and emerging technologies, including physical AI, at $200 billion. The estimates cover spending outside China and focus on infrastructure costs, with most new capacity under construction in the United States.

What is the $1.8 trillion enterprise number, exactly?

BCG itself flags the caveat. The company said the $1.8 trillion measures the spending needed to support the AI ecosystem, not a forecast of actual corporate outlays. That distinction matters. A requirement is not a commitment, and no company has signed up for a share of it.

In one scenario, companies would reallocate $857 billion from existing technology budgets and add $937 billion in new spending. The result would push enterprise technology budgets about 13% above International Data Corporation's IT spending forecast. AI's share of those budgets would rise from 5% to 25% by 2030. That implies the largest reallocation of corporate IT money in a generation, on assumptions BCG has not fully stress-tested in public.

Clark O'Niell, BCG's global lead for AI, infrastructure and cyber, frames the boardroom problem narrowly. Finance chiefs seeking more AI investment need to show boards productivity gains, cost reductions, new revenue or competitive advantages. He also conceded that some strategic benefits may be difficult to measure through traditional return-on-investment calculations. That is a notable admission from a firm selling AI advisory work.

"The question isn't just whether AI delivers an immediate financial return," O'Niell said. "It's also what the cost of falling behind could be."

Is this boom different from the last one?

BCG argues the AI buildout is more resilient than previous infrastructure booms, in part because hyperscalers have funded much of their investment from cash flow rather than debt. The claim has a basis, but it describes the past, not the future. O'Niell said investors should watch whether infrastructure funding shifts toward debt and whether deployed computing capacity generates enough revenue. Those are precisely the two signals that preceded earlier busts.

The arithmetic is demanding. BCG assumes AI chips have an average lifespan of five years. On that basis, it estimates each gigawatt of capacity entering service in 2026 must produce about $15 billion in revenue just to cover its costs. Weaker demand could trigger price cuts and losses for model providers and data-center operators. Nothing in the source establishes that current gigawatts are clearing that bar.

What should buyers do about falling prices?

O'Niell advised corporate customers to avoid long-term price-per-token commitments while AI prices continue to decline. Flexible contracts, he argued, could allow companies to benefit from future cost reductions. The advice cuts against the multi-year capacity deals that hyperscalers and chip suppliers have promoted. Buyers who lock token prices now may simply pay a premium for certainty they do not need.

The direction of pricing is the one firm fact in this debate. Declining per-token costs reward delay and punish early long-term commitments. BCG's own guidance treats that as settled.

Achievable, but not guaranteed

O'Niell's summary is candid, and it deserves to be quoted in full.

"The $2.6 trillion is achievable," O'Niell said, "but it's not guaranteed."

The gaps are easy to list. The estimate excludes China, which limits its claim to be global. The enterprise figure is a requirement, not a forecast. The revenue hurdle of $15 billion per gigawatt rests on a five-year chip lifespan assumption. And the resilience argument depends on cash-flow funding that could shift toward debt at any point before 2030.

What happens next turns on demand. If deployed capacity fails to generate the revenue BCG's model requires, expect price cuts across model inference and data-center services, followed by losses among the operators that built ahead of demand. If enterprises do reallocate toward that 25% budget share, the spending arrives and the arithmetic holds. O'Ni'll's own watch list — funding sources and revenue per deployed gigawatt — is the scoreboard to follow through 2026.