AI has reached its prove-it moment

​The technology can, and is, delivering huge productivity gains. Investors still need to establish who will capture the profits - and how much they should pay for them.

Artificial intelligence has reached the point where it could deliver everything its promoters promised but could still disappoint investors who paid too much for the wrong companies.

The real question for AI is whether the promised productivity gains will translate into revenue, and whether this is sufficient to repay the massive amounts of capital being poured into data centres and chips. And, if AI stocks are over-valued, are we likely to see rotation into something less risky?

The gap between AI spending and productivity gains

In its June 2026 outlook, J.P. Morgan Asset Management put consensus capex estimates for five U.S. hyperscalers at $697 billion for the year. It estimated that capex would absorb 93% of their operating cash flow, compared with 33% in 2023. These are estimates, to be sure, but capital appears to be chasing one singular goal, reminiscent of the railroad mania of the 1800s.

To understand whether this is money well spent, we need to analyse the promised productivity benefits, now and in the future. Virtually every company on the planet is involved in AI in some form in the hope of achieving productivity efficiencies. 

Those benefits are patchy and not all that convincing: The National Bureau of Economic Research surveyed nearly 6,000 executives in the US, UK, Germany and Australia between November 2025 and January 2026. Some 89% reported no impact from AI on labour productivity over the preceding three years.

That doesn’t necessarily mean their investment in AI will not deliver, only that there is a significant gap between what companies expect and what they have measured.

This also does not mean AI is destined to fail. AI adoption might simply take longer while the price of AI falls. Another factor weighing on corporate adoption is competition – if every business is using similar tools, the competitive edge is neutralised while customers get the benefits through lower prices.

This has prompted market commentary around an “AI bubble” and its inevitable deflation.

Investors may be measuring the wrong market

Veteran Wall Street investor Jordi Visser sees a much larger opportunity beyond the productivity gains businesses are measuring today. He argues that demand for AI should not be judged solely by the number of people paying for a chatbot, but by the potential emergence of an economy in which autonomous agents conduct business with one another.

In Visser's view, this future could involve tokenised stocks trading 24/7, alongside cryptocurrencies and stablecoins facilitating transactions between autonomous agents. We’re not quite there yet but getting close. The ability for investors to develop their own indices and ETFs using AI tools like Grokbot poses a fascinating (and decentralised) future for investment.

Visser developed a “46 Name” crypto index comprising six stocks and 40 tokens, all equally weighted, including Coinbase, Robinhood, Circle, PayPal, BTC, ETH, SOL, ZEC, TAO and NEAR. The index is up 50% year-to-date, even while Bitcoin is down 3.2%, suggesting that investor interest is spreading beyond Bitcoin.

The chart below shows the outperformance of Global X's AI & Technology fund (AIQ) against the S&P 500 over three years. AIQ is up 136% over three years, compared to 86% for the S&P 500. 

It’s clear where investors believe the future is headed.

AI’s proving ground 

If AI allows a small team to perform work that previously required a much larger organisation, its economic value extends well beyond the subscription revenue collected by the model provider.

There is some evidence of gains from AI in particular settings. Research by Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that AI assistance increased customer-support productivity by nearly 14%, with larger gains among less experienced and lower-skilled workers. 

Microsoft reported more than 30 million paid Microsoft 365 Copilot seats in its July 2026 results. Azure revenue exceeded $100 billion for the financial year, although Azure includes conventional cloud services as well as AI.

It's clear that customers are willing to pay for AI, though whether those investments deliver a lasting competitive advantage remains to be seen.

The benefits are not going to be evenly spread. Visser’s argument is that smaller businesses may be able to achieve efficiencies quicker with AI, while more established companies may struggle with their existing costs and falling prices. There is also a risk that computing bottlenecks may slow adoption, and hence weaken the return on capital.

Rotate into what?

This raises the question – if AI is over-valued, where will the smart money move?

Mature businesses with steady earnings may look like the obvious choice, but those earnings may come under threat from competitors who are now able to enter the market more cheaply.

Chip suppliers such as Micron and AMD are riding a wave of euphoria at the moment, but they are not immune to any potential slowdown in customers’ investment plans.

Semiconductor producers -Year-to-date stock performance

The table above shows an increasingly crowded semiconductor chip sector with one-time Wall Street darling Nvidia delivering a rather pedestrian 16.8% year-to-date return against some extraordinary competition.

Micron has been the standout stock of the year with a YTD return of 218%, built largely on the back of tight memory supply. AI processors need fast access to data, creating demand for high-bandwidth memory, or HBM. Micron benefits from both that demand and tighter supply across the memory market, which supports higher selling prices.

AMD offers its own EPYC brand of server processors and Instinct AI accelerator which plonks it firmly in the expanding data centre market. The company reported second-quarter data-centre revenue of $6.7 billion, up 107% year on year, driven by demand for both product lines. It has also announced a collaboration with Anthropic for the deployment of up to two gigawatts of AMD computing capacity.

Intel has reignited investor hopes following the turnaround under CEO Lip-Bu Tan, with second-quarter revenue rising 25% to $16.1 billion – despite reporting a loss under standard accounting rules – while data centre and AI revenue increased 59%. 

Arm designs processor technology and licenses it to chipmakers, earning licensing fees and royalties on chips using its designs. Its technology powers most smartphones and is expanding into data centres and AI. It has also moved into selling its own data-centre processors, with manufacturing outsourced. Arm forecasts that its new processors could generate about $15 billion in annual revenue by fiscal 2031.

What investors should watch

These stocks have clearly front-run the promise of future revenue, and the slowdown in more mature AI stocks such as Nvidia provides a sobering lesson: the next phase of the AI trade calls for a more discerning approach.

To maintain current valuations, investors will want to keep an eye on demand, including renewals and spending by existing customers. Revenue growth alone does not equal profits, so a more important question is: how profitable is that growth?

Underlying this is the customer experience: are business customers reporting measurable savings or additional output after allowing for integration, supervision and errors?

A rotation need not mean abandoning AI altogether. Capital could move from expensive suppliers towards reasonably valued businesses using AI to improve their operations. Alternatively, investors could favour companies with dependable cash generation and less exposure to technological disruption.

Neither outcome is inevitable. Both depend on what the earnings and cash-flow figures reveal.

The real test for AI investors

AI’s prove-it phase will not be settled by an impressive demonstration. It will be settled by paying customers, sustainable margins and returns on the money invested. Growing demand is encouraging, but investors need to see whether revenue can grow fast enough to justify the enormous spending on chips, data centres and power.

The gains may also spread beyond the companies building the infrastructure to businesses using AI to cut costs, improve services and expand without adding large numbers of staff. Some of the eventual winners may attract little attention today.

Visser may be right about the scale of the opportunity. But a technology that transforms the economy does not make every company attached to it a good investment. Investors still have to get the company, the price and the timing right.