It has been an interesting week for AI economics reports. Torsten Slok, chief economist at Apollo Global Management, which manages over $1 trillion in assets, reported that AI was having “zero financial payoff to corporate America outside tech”. Specifically, revenue per worker has been flat across the 493 S&P 500 companies outside the “Magnificent Seven” (Alphabet, Amazon, Apple, Meta, Microsoft, NVIDIA and Tesla). Indeed, only the Magnificent Seven show increasing profit margins. There has not yet been an obvious economy-wide acceleration in productivity attributable to AI.
(source: Apollo Global Management, October 2026)
This report merely confirms others. A 2026 National Bureau of Economic Research study reported that 90% of executives found no impact of AI on productivity over a three-year span. Industry analyst Gartner reported in October 2026 that just 3% of organisations exceed cost-saving goals via AI productivity, and 38% of companies are over budget on AI.
Separately, Gartner reported that likely AI revenues through 2030 will fall far short of the amounts required to justify the industry’s enormous capital expenditure. Gartner estimates that AI data centre owners need cumulative revenues of $8.5 trillion by 2030 to cover their capital spending, financing costs and achieve a 12% return on invested capital (ROIC), its assumed hurdle rate. Yet Gartner projects at most $3.6 trillion in revenue attributable to AI data centres over this period, a shortfall of $4.9 trillion.
The so-called “neocloud” vendors, such as CoreWeave, Nebius and Lambda, are particularly vulnerable, since they lack the diversified revenue streams of the major cloud providers. Gartner predicts that over 80% of neocloud vendors will cease operating independently by 2030. Building data centres is certainly capital-intensive, and the combined revenue of neocloud vendors in 2025 was just $25 billion, while analyst Mordor Intelligence estimates 2026 neocloud revenues at $35 billion. This represents a 40% increase, but even if revenues continued growing at that rate, they would reach only $134 billion in 2030, with cumulative revenues of $383 billion from 2026 to 2030. Gartner itself forecasts that the broader AI-optimised infrastructure-as-a-service market will reach $174 billion in 2030.
Meanwhile, Fortune Business Insights estimates that the global AI inference market will grow from $104 billion in 2025 to $118 billion in 2026, an increase of just 14%. This is a different market, which includes conventional AI inference as well as generative AI, but it provides some context for the scale of the infrastructure investment boom. Failing to achieve a 12% ROIC does not necessarily imply bankruptcy, but for highly leveraged neocloud providers facing rapid hardware depreciation, persistently inadequate returns could have much more serious consequences. The fundamental question is whether future demand for AI computing will grow quickly enough to justify the extraordinary sums currently being invested.
We are now almost four years into the generative AI era, since ChatGPT was launched publicly in November 2022. That event spurred an enormous frenzy of investment, expectation and excitement. Yet, four years on, only 3% of enterprises have exceeded their AI productivity cost-saving targets, and a whole generation of neocloud vendors, which sprang up to satisfy AI demand, face severe economic headwinds. We may soon see the IPO of Anthropic, whose S-1 filing included an $80.7 billion loss in the first half of 2026; admittedly, most of this was due to accounting adjustments relating to convertible debt rather than underlying operating losses (which were $693 million), but it is still quite a number for a company whose revenue in the first half of 2026 was $16.3 billion. The same document devotes 92 pages to financial risks and has met with a mixed reaction. The coming months promise to be interesting as the rubber starts to hit the AI road.
