$7.7 trillion AI bet: Can hyperscalers earn enough to justify the spending?

Six large hyperscalers are expected to invest $7.7 trillion between 2026 and 2030, including about $6.8 trillion above their pre-AI capital expenditure run-rate, according to UBS Global Research . To generate a 15% return on invested capital by 2031, they would need to produce roughly $1 trillion in additional earnings before interest and taxes.

At current cloud margins, that would require approximately $3.4 trillion in incremental revenue.

The scale of the investment has intensified scrutiny of whether AI infrastructure spending can generate returns commensurate with the capital being deployed. The concern has been amplified by weakness in bonds issued by some technology companies, particularly neoclouds such as CoreWeave.

But UBS EM Strategist Sunil Tirumalai said the credit market stress should not be confused with a broader breakdown in the AI investment thesis. Neoclouds account for only 5%-7% of expected AI data centre capital expenditure over the next four to five years and are “not central to the global AI capex story”, according to a note by the global investment bank.

Hyperscalers have historically generated returns of more than 40% before tax from asset light software, social media and e-commerce businesses. AI is changing that equation by making their financial profiles more asset heavy.

UBS expects this shift to result in a 20-25 percentage point decline in return on invested capital as AI becomes a larger part of their businesses.

Even after that decline, a 15% ROIC by 2031 would require around $1 trillion of incremental EBIT, slightly less than twice the group’s current combined EBIT. At current cloud margins, the implied revenue requirement is about $3.4 trillion.

That makes monetisation, rather than demand for AI services alone, the central investment question.

The spending can still generate acceptable returns, UBS said, but the calculation rests on two conditions: AI must deliver genuine productivity gains, and the AI ecosystem must capture a meaningful share of those gains through effective monetisation.

The global nominal economy is currently around $110 trillion and could reach approximately $136 trillion by 2031 if it continues to grow at the pace of the past decade, even without an AI-related boost.

UBS estimates that if AI adds 0.5%, 1% or 1.5% to annual productivity growth, global GDP could be $3 trillion, $7 trillion or $10 trillion higher, respectively, by 2031.

That additional economic output could provide the foundation for hyperscalers and other AI companies to generate the revenue needed to support their investment. But productivity gains alone will not guarantee that the companies funding the infrastructure capture the value.

UBS draws a comparison with the telecom and dotcom/cable industries. Those sectors arguably generated productivity benefits for the broader economy but failed to capture a commensurate share of the value created. Much of that value ultimately accrued to internet, e-commerce, social-media, streaming and smartphone businesses.

Hyperscalers, however, have a broader ability to capture value across multiple revenue streams than regulated or licensed telecom companies did, according to the note.

The recent market sell-off has been linked to several triggers, including higher bond yields among technology issuers. UBS said yields on hyperscaler bonds have risen this year, but higher US Treasury yields explain almost all of that increase.

The stress is concentrated among issuers viewed as having weaker balance sheets or significant contingent liabilities. However, the credit market pressure is not necessarily a signal that the entire AI infrastructure cycle is breaking down. Analysts say the risk is instead that capital spending continues to rise faster than monetisation, leaving investors to reassess the returns on the infrastructure buildout.

Power availability is another potential constraint. UBS’s global semiconductor team believes there is visibility on securing enough electricity to run the planned data centres, but meeting the targets will require smooth execution, supportive legislation and no delays in grid or off-grid capacity additions.

Against the broader volatility in emerging markets, Asia and global technology stocks, India has been among the markets that have held up since October 5, according to UBS’s EM markets comparison.

That relative resilience does not remove the global risk posed by a reassessment of AI spending . But it suggests that India’s market performance has, so far, been less exposed to the sharpest pressure points in the global AI capex trade.

For investors, the next phase may therefore be less about whether AI spending is large enough to transform the economy and more about who captures the returns.

The numbers leave little room for disappointment. A $7.7 trillion investment programme must ultimately produce $1 trillion of additional EBIT and the path to that outcome runs through productivity gains, revenue growth and the ability of hyperscalers to monetise AI at scale.