Apollo Economist: AI Boom Delivers 'Zero' Results Outside Big Tech

By Daniel Okafor

Gartner projects a $4.9 trillion gap between the cumulative revenue AI data centers need through 2030 and what they are likely to earn. That figure anchors a growing body of skepticism about whether the AI investment boom can pay for itself. At the center of the debate is Torsten Slok, chief economist at Apollo Global Management, who argues the boom has produced almost nothing for corporate America outside a handful of tech giants.

What did Apollo's analysis find?

Slok compared the so-called Magnificent Seven — Nvidia, Microsoft, Apple, Alphabet, Amazon, Meta and Tesla — with the other 493 companies in the S&P 500. Revenue per employee and profit margins at the M7 keep trending upward. The aggregate metrics for the other 493 companies have been essentially flat.

Slok said that "despite massive investment and enthusiasm for AI, the financial results delivered by U.S. companies outside the technology sector are zero." He added that economy-wide U.S. productivity data still shows no conclusive evidence that AI has lifted the growth rate.

The metric carries limits. Business conditions and sector-specific characteristics shape revenue per employee and margins together, so the numbers alone cannot prove AI's effects either way. The comparison group of 493 companies also includes technology firms outside the M7. The point of the analysis is narrower: the AI effect is not yet visible in the earnings of major U.S. companies.

Why do companies treat token spending as a cost?

Slok flagged a shift in how ordinary companies pay for AI. He said that "ordinary U.S. companies have begun to cut spending and now see token spending not as a source of growth but as a cost that pressures profit margins."

Tokens are the basic unit AI models use to process information. Companies that use external AI services pay usage fees based on volume processed. If adoption does not generate revenue growth or cost savings above those fees, the spending becomes a burden.

Token spending by venture capital-backed companies continues to rise, according to Slok. Ordinary companies take a different view. They judge that they do not necessarily need the latest AI models and show little inclination to pay high costs. Slok predicted that "as long as AI agents and token spending remain confined to certain parts of the economy, AI's impact on the macroeconomy will be very limited."

Survey evidence points the same way. Gartner polled 2,002 chief information officers and technology executives between May and July this year. Only 13% said their companies had realized substantial or transformative value through AI. Alicia Mullery, a vice president-level analyst at Gartner, said "volume is not value," and advised companies to define expected outcomes before adopting AI and set criteria for discontinuing underperforming initiatives.

How big is the data center revenue gap?

Gartner's numbers quantify the infrastructure risk. The firm assumes AI companies will spend about $1 trillion on infrastructure this year and invest an additional $6 trillion from 2027 to 2030. On that path, more than $3.5 trillion in AI infrastructure investment could face asset impairment risk by the end of 2030.

The revenue math is the problem. Gartner projects data center operators need cumulative revenue of $8.5 trillion through 2030 to cover capital expenditure and financing costs and deliver a 12% return on invested capital. Revenue attributable to data centers from AI services for enterprises and consumers over the same period will reach only up to $3.6 trillion. Will Sommer, a senior director analyst at Gartner, said that "AI revenue from enterprises and consumers alone will make it difficult to earn sufficient returns on all planned AI capital spending."

Neocloud companies, whose businesses concentrate on AI computing infrastructure, face the sharpest pressure. Amazon and Microsoft can fund AI investment with revenue from diverse businesses. Specialists cannot. Gartner forecast that more than 80% of neocloud companies will cease independent operations by 2030 under cost pressure and difficulty differentiating. That is a long-term risk analysis — it confirms no actual losses or closures, and the revenue gap is not a confirmed loss amount.

How have markets reacted?

The caution has reached trading desks. Meta's share price fell about 7% after it announced further expansion of AI capital spending, and its credit default swap premium, which reflects default risk, soared to a record high. AI cloud company CoreWeave saw its shares and bonds slump on concerns over data center investment funded by large-scale borrowing. Australian data center firm Firmus Grid scrapped an IPO that targeted a valuation of about $30 billion after U.S. investors shunned it, and Nvidia-backed nScale drew criticism during its IPO process for an excessively high valuation.

Who decides the industry's direction?

Slok also questioned AI's structural limits. He said that "AI may be an excellent calculator, but most of the intelligence that matters in corporate management is a matter of judgment," arguing AI does not adequately handle responsibility, trust, organizational culture and ethical choices.

He stressed that "the direction of the AI industry will be decided not by the M7 but by the other 493 S&P 500 companies." Whether those ordinary users of AI can lift revenue and margins will determine the industry's success or failure. Until gains concentrated in the Magnificent Seven spread to the rest of the index, the sustainability of the AI boom will stay under scrutiny.