Artificial intelligence has already won the expectations battle. Investment is flowing into semiconductors, data centres, computing infrastructure and electricity systems. Companies are reorganizing strategies around AI, governments increasingly view computing capacity as strategic infrastructure, and investors are assigning substantial value to businesses expected to benefit from the technology.

The scale is striking. Stanford's 2026 AI Index reports that U.S. private AI investment reached $285.9 billion in 2025. The technological progress behind this investment is also difficult to dismiss. But from an economic perspective, the more important question is whether these advances will translate into sustained productivity growth.

We have seen this pattern before

The history of technological change gives us good reason to be optimistic about AI, but also good reason to be patient. Businesses invested heavily in computers throughout the 1970s, 1980s and early 1990s while computing power improved rapidly. Yet aggregate productivity remained surprisingly weak. Robert Solow captured the puzzle in 1987 when he observed that the computer age could be seen everywhere except in the productivity statistics.

The picture eventually changed. U.S. labour-productivity growth accelerated substantially in the second half of the 1990s, rising to around 2.8% annually during 1995–2000 compared with 1.6% during 1990–1995.

The delay matters. Buying a new technology does not automatically make a business more productive. Firms need to redesign workflows, train workers, reorganize production and discover where the technology creates real commercial value.

Economists Erik Brynjolfsson, Daniel Rock and Chad Syverson describe this process as the Productivity J-Curve. Investment in a general-purpose technology may come well before its full productivity benefits because the organizational changes needed to use it effectively take time. The internet boom offers a related warning. The internet ultimately transformed the economy, but many dot-com-era expectations about how quickly individual companies would become profitable proved unrealistic.

A technology can be genuinely transformative while expectations about its timing, valuations and returns still run ahead of economic reality.

The productivity gains are real, but still uneven

There is already credible evidence that generative AI can improve productivity in specific activities. A widely cited study of 5,179 customer-support agents found that access to a generative-AI assistant increased issues resolved per hour by about 14% on average, with larger gains among less experienced workers. This is economically meaningful evidence. But it is important to distinguish between making a task faster, making a business more productive and raising productivity across an entire economy.

Suppose AI allows an employee to complete a task in half the time. The task has clearly become more efficient, but the business benefit depends on what happens to the time saved. Does the employee produce more, improve quality, reduce costs, serve more customers or help create new products and revenue? If not, a large improvement in task-level efficiency may produce a much smaller improvement in firm-level productivity.

Moving from successful AI applications to higher economy-wide productivity requires these gains to spread across firms, industries and business processes.

Recent U.S. data are encouraging, but they do not yet settle the question. Private nonfarm business labour productivity grew 3.0% in 2024 and 2.2% in 2025, compared with average annual growth of about 1.5% during 2007–2019. Yet total factor productivity increased by only 0.8% in 2025. It is too early to attribute the recent improvement primarily to AI because pandemic-related restructuring, capital investment, labour-market adjustment, remote work and other changes are occurring at the same time.

The evidence therefore supports a measured conclusion. AI is already generating productivity gains in some activities, but an economy-wide AI productivity revolution has not yet been established.

Investment, resources and the business case

That distinction matters because the resources being committed to AI are substantial. AI infrastructure requires chips, servers, data centres, software, networks and electricity. These investments have an opportunity cost. Capital committed to expanding computing capacity cannot simultaneously be invested elsewhere. For businesses, the relevant question is therefore no longer simply whether AI is useful. It clearly can be.

The harder question is whether the additional value created will justify the capital and operating resources required to produce it.

This is also where the productivity debate connects directly with sustainability and business strategy. AI may appear almost weightless to someone using a chatbot or automated tool, but the infrastructure behind it is highly physical. The International Energy Agency projects global data-centre electricity consumption to increase from roughly 485 TWh in 2025 to around 950 TWh by 2030, while electricity use from AI-focused data centres is expected to triple. In the United States, data centres could account for nearly half of additional electricity demand through 2030.

These trends do not mean that AI is economically or environmentally undesirable. They mean that productivity, cost and sustainability cannot be treated as completely separate questions. Electricity, water, computing equipment and grid capacity are business inputs as well as environmental resources.

An AI investment that produces modest productivity improvements while requiring rapidly rising infrastructure and energy costs presents a different business case from one that generates substantially more economic value from the same resources. More efficient models, better use of computing infrastructure and cleaner electricity can therefore improve both the sustainability profile and the economics of AI adoption.

For executives and investors, this shifts the focus from AI adoption itself to the quality of that adoption. The objective should not be to use more AI simply because the technology is available. It should be to identify where AI creates measurable value relative to its full cost, including financial, organizational and resource-related costs. In that sense, sustainability is not an additional constraint placed on the AI business case. Resource efficiency is part of the business case.

The economic test of the AI boom

The debate over whether AI is hype or reality is therefore too simple. AI can be transformative while some expectations surrounding it are excessive. History gives us reasons for optimism because general-purpose technologies often require years of complementary investment before their largest productivity effects appear.

But the same history gives us reasons for caution. Being right about the importance of a technology does not mean being right about its valuation, the speed of adoption or who will ultimately capture the economic returns.

The AI boom should ultimately be judged by observable outcomes: sustained increases in output per hour, stronger total factor productivity, lower production costs, new products and services, and productivity improvements that spread beyond technology companies into the wider economy. For businesses, another measure belongs alongside these indicators: how efficiently capital, energy and other resources are converted into economic value.

AI investment is booming, and some productivity gains are already visible. The unresolved question is whether those gains will become large and widespread enough to justify the extraordinary investment now taking place.

Will the AI investment boom become a sustained productivity boom? That, not the latest model benchmark or investment announcement, is the economic test that matters.

Sources

  1. Stanford Institute for Human-Centered AI, AI Index Report 2026
  2. U.S. Bureau of Labor Statistics, productivity and multifactor productivity data
  3. Brynjolfsson, Rock and Syverson, The Productivity J-Curve
  4. Brynjolfsson, Li and Raymond, Generative AI at Work
  5. International Energy Agency, Energy and AI
← Back to Insights