AI investments still clearly exceed anticipated productivity benefits
Summary
- Investor sentiment in the AI sector has shifted positively in 2026, with hyperscalers like Amazon, Microsoft, Google, and Meta investing over 700 BUSD in data centers, equating to about 2% of US GDP.
- The AI boom has surpassed historical investment milestones, with only the 19th-century railway mania exceeding current levels, though analysts suggest this could change by the decade's end.
- Despite optimism, 85% of AI ecosystem revenue is from "capex recycling," raising concerns about the sustainability of investments without external revenue generation.
- Warning signs include potential IPOs from SpaceX, OpenAI, and Anthropic, alongside possible monetary policy tightening by central banks, suggesting caution in the AI investment landscape.
This content is generated by AI. You can give feedback on it in the Inderes forum.
Automatic translation: Originally published in Finnish 09/06/2026, 04:54 GMT. Give feedback here.
Investor sentiment in the AI sector has turned upside down within a year. The growth in cloud service revenues and the surge in demand for computing capacity have alleviated concerns about overcapacity, but the same structural problems still persist in the background.
The capital flow directed towards AI infrastructure in the United States is reaching historical proportions. In 2026, hyperscalers, Amazon, Microsoft, Google, and Meta, will inject over 700 BUSD into data centers, which corresponds to of around 2 percent of the US GDP.
Cumulatively, the AI boom has already surpassed the Apollo space program, the construction of telecommunications infrastructure in the 1990s, and the shale oil revolution. Only the 19th-century railway mania still exceeds the current pace, but if analysts' estimates materialize, even that milestone could be broken by the end of the decade, as noted by Dario Perkins in a recently published TS Lombard article, whose figures are also discussed here.
Rapid turnaround in sentiment
Just a while ago, the market was skeptical about AI investments. Hyperscalers' investment rates rose to over 100 percent of revenue, free cash flow collapsed, and indebtedness grew. Lessons from the Dot-com bubble were remembered from history: even though the internet changed the world, pioneering companies suffered huge losses from overcapacity.
However, at the beginning of 2026, the mood shifted. The annual growth rate of cloud service revenues rose to 370 BUSD, and order backlogs swelled to nearly 1.5 BUSD. Token processing multiplied, and the entire ecosystem is suffering from an acute hardware shortage. The adoption of agent-like AI increased token consumption by 5 to 30 times, which strengthened belief in the AI revolution and has added bottlenecks within the industry.
However, optimism may be premature. Of around 85 percent of the AI ecosystem's revenue still comes from "capex recycling," meaning hyperscalers invest in infrastructure, which is reflected in suppliers' revenue and "other income" as an appreciation in the value of cross-shareholdings. Also, 85 percent of the demand for computing capacity comes from within the ecosystem.
Common sense, however, would necessitate that companies and consumers are willing to pay for genuine productivity benefits. With current user numbers (40% of companies use AI for more than an hour a week) and efficiency assumptions, AI could generate an annual productivity jump of 0.6 percentage points. This would correspond to annual efficiency benefits of around 200 BUSD. Naturally, the revision would be significant, but is it enough to justify an 800 BUSD annual investment?
Warning signs are also starting to appear in the economy: SpaceX, OpenAI, and Anthropic are planning IPOs, and central banks are considering tightening monetary policy, which could be a challenging combination. The AI theme currently has strong momentum, but caution is also warranted. To succeed in the long run, the ecosystem must demonstrate the ability to generate revenue from outside its own sphere.
