Hyperscalers spend on AI infrastructure globally; memory industry profits surge.

August 2026: early signs of AI return on investment?

Hyperscalers continue to spend, but the market is questioning the magnitude of the payoff from doing so. However, we are approaching a point where today’s investment should translate into materially stronger revenue growth.

In December 2025, I wrote about what the market was looking for from the technology sector. Put simply, investors wanted evidence that the enormous investment being made in artificial intelligence (AI) was beginning to generate attractive returns.

More than six months later, it is an appropriate time to revisit that debate. Since then, AI investment has continued to exceed expectations, with the industry’s largest companies announcing ever larger capital expenditure programmes. At the same time, evidence that AI is being adopted across the economy has continued to build. Yet despite this progress, investor concerns over whether these investments will ultimately generate attractive returns have persisted.

This article revisits that discussion. It examines how AI investment has evolved, where evidence of financial returns is beginning to emerge, and why the market remains cautious despite increasingly encouraging signs.

AI spending continues to be very high

The first step is to consider how AI investment has evolved since the end of last year. Unsurprisingly, spending has remained exceptionally strong and, if anything, has continued to surpass market expectations. The largest technology companies provide the clearest illustration of this trend. As the charts below show, capital expenditure remains well above historical levels, while analysts have continued to revise spending expectations higher throughout 2026.

Source: LSEG Datastream, Schroders, July 2026. Any reference to regions

securities is for illustrative purposes only and not a recommendation to buy or sell any financial instruments or adopt a specific investment strategy. T his material may contain “forward-looking” information, such as forecasts or projections. Any forecasts stated in this document are not guaranteed and are provided for information purposes only.

Funding this investment is becoming increasingly demanding. We are approaching the point where a significant proportion of operating cash flow, the cash businesses generate from their day-to-day activities, is being allocated to capital expenditure. In some cases, planned capital expenditure now exceeds operating cash flow. Should companies wish to maintain this pace of investment, they may ultimately need to rely on debt or equity financing.

Source: LSEG Datastream, Schroders, July 2026. Any reference to regions

securities is for illustrative purposes only and not a recommendation to buy or sell any financial instruments or adopt a specific investment strategy. This material may contain “forward-looking” information, such as forecasts or projections. Any forecasts stated in this document are not guaranteed and are provided for information purposes only.

This is not necessarily cause for concern. These investments are intended to generate additional revenue, profits and cash flow over time, meaning today’s elevated spending should be assessed alongside the future earnings power it is designed to create.

The overall picture therefore remains one of exceptionally high investment in AI infrastructure.

The market continues to question whether it is worth it

The extraordinary level of investment in AI is, naturally, leading to questions about the returns it will ultimately generate. As with many large investment cycles, there is a timing mismatch. These investments require substantial upfront capital, while the associated revenues, profits and cash flows will only emerge once new data centre capacity comes online. Investors therefore bear the cost today, with the benefits only expected to materialise over time.

In this instance, those concerns are amplified by the sheer scale of the investment. Not only are revenues deferred, but the quantum of capital being deployed is so large that it is materially reducing the free cash flow of some of the world’s most successful and cash generative businesses.

This dynamic is creating a clear divergence across the AI ecosystem. Companies supplying AI infrastructure have generally performed exceptionally well, benefiting directly from the surge in demand. Nowhere is this more evident than in the memory industry during 2026, where constrained supply and exceptionally strong demand have driven unprecedented growth in revenues, profits and free cash flow.

By contrast, many of the companies funding this investment have underperformed both the broader technology sector and the AI supply chain. That underperformance has also resulted in a meaningful de-rating of earnings-based valuation multiples, suggesting investors remain unconvinced that today’s investment will ultimately generate sufficiently attractive returns. This is reflected in the divergence between the SOX semiconductor index, a useful proxy for the AI supply chain, and the share price performance of the major AI spenders.

Source: LSEG Datastream, Schroders, July 2026. Any reference to regions

securities is for illustrative purposes only and not a recommendation to buy or sell any financial instruments or adopt a specific investment strategy. Past Performance is not a guide to future performance and may not be repeated.

Taken together, the market’s question appears relatively clear - will returns from this investment ultimately justify the scale of capital being deployed?

Is there evidence of ROI?

The short answer is yes, although the evidence currently comes through leading indicators such as revenue growth rather than conventional return metrics, and not yet to the extent required to fully reassure investors.

The first point to make is that AI adoption continues to accelerate. Whether through consumer chatbots, software development, digital advertising or AI embedded within enterprise applications, adoption is increasing rapidly as the technology evolves and demonstrates its usefulness. This is an important starting point. Widespread adoption is a necessary condition for the current investment cycle to generate attractive returns. Had AI failed to deliver meaningful productivity gains or user value, the scale of investment taking place today would undoubtedly prove difficult to justify. Fortunately, the evidence increasingly points in the opposite direction.

The more difficult question is whether that adoption is beginning to translate into company financials. In our view, there are two particularly useful areas to examine: the revenues being generated by the leading large language model (LLM) developers and the revenue growth of the hyperscale cloud providers investing most heavily in AI infrastructure.

Starting with the model developers, companies such as OpenAI and Anthropic are among the fastest-growing software businesses ever created. While precise figures vary across public disclosures and industry estimates, both appear on track to generate tens of billions of dollars of revenue during 2026, with growth expected to remain exceptionally strong. To illustrate the pace of that growth, Anthropic disclosed in May 2026 that its annualised revenue run rate had exceeded $47 billion, just five years after the company was founded.

Elsewhere, the hyperscale cloud providers are also beginning to see AI contribute more meaningfully to growth. Revenue growth has started to improve, while customer backlogs, representing committed future spending, continue to expand. Those backlogs suggest that demand for AI infrastructure remains strong and provide confidence that growth should continue as additional capacity comes online.

Source: LSEG Datastream, Schroders, July 2026. Any reference to regions

securities is for illustrative purposes only and not a recommendation to buy or sell any financial instruments or adopt a specific investment strategy

Encouraging as these developments are, they have not yet been sufficient to fully alleviate investor concerns for several reasons.

First, the scale of investment remains so significant that revenues still need to grow considerably before returns begin to approach historical levels. One useful measure in this regard is tangible asset turnover, which compares the revenue generated by a company’s physical assets with the capital invested in them. For the largest AI investors, this measure has come under considerable pressure as data centre investment has accelerated much faster than revenues. As utilisation improves and newly built capacity is put to work, this metric should begin to recover. For now, however, it serves as a reminder of just how much investment has been undertaken ahead of demand.

Second, revenue growth should be viewed as a leading indicator rather than definitive proof that AI investments are generating attractive returns. Ultimately, investors will judge these investments on the earnings, cash flows and returns on capital they generate. Strong revenue growth is therefore an important first step, but it must ultimately translate into higher profitability, stronger cash generation and improving returns on capital.

We remain comfortable with the current dynamic. Large investment cycles are rarely linear, and periods where upfront investment runs well ahead of future returns often prove uncomfortable for investors. However, we believe the industry is approaching the point where a growing proportion of today’s investment should begin translating into materially stronger revenue growth. Company commentary around AI demand and expanding cloud backlogs increasingly supports that view.

We also believe there are credible reasons why that revenue can ultimately generate attractive returns. Many of the largest AI investors operate highly profitable businesses with strong competitive positions, significant scale and established routes to monetisation. If AI-related revenues continue to grow, those characteristics should provide a pathway for that growth to translate into earnings, free cash flow and improving returns on capital.

That is not to say the investment cycle is without risk. The fact that AI is proving to be a useful technology does not necessarily mean the current level of investment is appropriate. History contains many examples of technologies that ultimately transformed the economy but still experienced periods of excessive investment along the way. Both statements can be true.

One important risk is customer concentration. A significant proportion of cloud providers’ committed demand ultimately depends on a relatively small number of leading LLM developers. Should those companies materially reduce their investment plans, the effects would likely ripple through the AI supply chain.

That said, the LLM companies are not the ultimate source of demand. Their investment decisions are themselves driven by businesses and consumers adopting AI applications across the economy. In other words, end-user demand ultimately determines the need for additional compute capacity, with model developers acting as intermediaries between users and infrastructure providers.

The evidence that AI investment is generating meaningful revenues continues to strengthen. Adoption is accelerating, frontier model developers are monetising at an extraordinary pace, and hyperscale cloud providers are beginning to see stronger growth from AI-related workloads.

While revenues are only the first step, we believe the largest AI investors are well positioned to convert that growth into earnings and cash flow in the future. Their scale, profitability and strong competitive positions provide a credible foundation for improving returns as utilisation rises and the revenue contribution from AI becomes more meaningful.

So, is there an AI ROI? It is still early b