NORBERT BIEDRZYCKI

The AI Investment Paradox: Record Capital, But Where Are the Returns?

The AI Investment Paradox: Record Capital, But Where Are the Returns?

“You can see the computer age everywhere but in the productivity statistics.” Robert Solow wrote that in 1987. It is worth rereading in 2024.

The amount of capital flowing into AI this year is without precedent. The four largest American technology companies, Microsoft, Alphabet, Amazon and Meta, are on track to spend well over $200 billion on capital expenditure in 2024, most of it linked to data centres and AI infrastructure. Nvidia briefly became the most valuable company in the world in June. Frontier model developers are raising rounds measured in billions rather than millions.

And yet, the question that dominated investor conversations this summer was not about growth. It was about returns.

The $600 billion question

In June, David Cahn of Sequoia Capital published an analysis that framed the debate precisely. If you take the current run rate of spending on AI infrastructure and assume that the companies buying it need to earn a reasonable margin, the AI ecosystem would need to generate around $600 billion of annual revenue to justify the investment. Actual AI revenue today is a fraction of that.

A few weeks later, Goldman Sachs published a research report with a title that captured the mood: too much spend, too little benefit? In it, MIT economist Daron Acemoglu estimated that AI might add only around 1 percent to US GDP over the next decade, far below the optimistic forecasts circulating in the market.

These are not fringe views. They come from some of the most sophisticated investors and economists in the world.

What I see on the ground

In my work with enterprise leadership teams and portfolio companies, I see a consistent pattern. Almost every organisation has launched generative AI pilots. Very few have moved them into production at scale. Those that have often report meaningful gains in specific processes, such as customer service, software development and document handling, but the gains have not yet aggregated into visible changes in P&L.

The reasons are rarely technical. They are organisational. Processes have not been redesigned. Data is fragmented. Accountability for outcomes is unclear. Employees use AI tools individually, but the organisation has not changed how work is structured. Technology can be improved quickly. Organisations are much harder.

Signs of a shakeout

Meanwhile, the model layer is already showing signs of strain. In March, Microsoft hired most of Inflection AI’s team, including its co-founder Mustafa Suleyman, in a deal that left the startup a shell. Similar arrangements followed at Adept with Amazon and at Character.ai with Google. Stability AI, one of the darlings of 2022, saw its founder step down amid financial difficulties. These are quiet acquisitions of talent, not triumphant exits.

The lesson for investors is that being technologically impressive is not the same as building a sustainable business, particularly when your largest competitors are also your suppliers of compute and your potential acquirers.

How to think about it

I do not believe the AI boom is a bubble in the sense that the technology lacks value. I do believe that the timing of returns is being underestimated, and that the distribution of returns will be highly uneven. History suggests that during infrastructure cycles, early investors in capacity often see poor returns, while those who use the infrastructure later, when it is abundant and cheap, capture most of the value.

The market is beginning to distinguish between companies creating technological breakthroughs and companies building profitable businesses. That distinction will define the next phase.

Investment question: Is the AI investment boom creating sustainable economic value, or are investors financing an infrastructure race whose returns remain uncertain?

Read next

  1. The Business of AI 2020–2026: about this series

  2. The New AI Power Struggle: OpenAI, Microsoft, Google and the Battle for the AI Platform

    The Business of AI series, written in 2026

  3. The Physical AI Opportunity: Investing Beyond Software

    The Business of AI series, written in 2026