Expanding AI’s production possibilities frontier

From a macroeconomic perspective, AI presents both significant opportunities and notable challenges.

It has the potential to boost economic growth by increasing total factor productivity (TFP) and expanding the production frontier. At the same time, it raises concerns about job displacement and the uneven distribution of gains across workers, firms, and countries.

The integration of AI into the economy also adds complexity to monetary policy, particularly through its effects on inflation, the labour market and the neutral rate.

This article examines the macroeconomic implications of AI through several key channels: productivity, capital expenditure, labour markets, and inflation, before turning to the implications for monetary policy. Our central view is that AI will gradually lift productivity and exert disinflationary pressure over the medium term, although in the near term, AI is inflationary. The transmission will be slow, and we incorporate conservative estimates in our macro forecasts.

Productivity

AI can influence economic growth through multiple channels. Most notably, it has the potential to boost long-term growth by increasing productivity, although the magnitude and timing of these gains remain uncertain. In parallel, AI may accelerate capital deepening by stimulating investment in digital infrastructure, while also partially offsetting the drag from demographic ageing.

A central mechanism is AI’s impact on TFP. By improving decision-making, optimising resource allocation, and automating both routine and complex tasks, AI expands the production possibilities frontier. In more advanced scenarios – where AI systems increasingly automate innovation and production processes – growth effects could become even more pronounced. However, historical experience suggests that such gains are not immediate. As with past general-purpose technologies, such as electricity and IT, implementation lags, organisational adjustments and learning costs can delay measurable productivity improvements. Over time, as adoption broadens, the impact on trend growth could become substantial.

In the US, we have raised our real gross domestic product (GDP) growth forecast for the next decade by 0.4 percentage points, with higher productivity growth offsetting the diminishing contribution from labour supply. In the euro area, we expect a smaller uplift, reflecting (i) lower AI adoption rates – roughly half the US level on most enterprise surveys – (ii) a smaller share of high-exposure occupations, and (iii) a more fragmented digital single market that slows diffusion. The UK sits between the US and the euro area, given higher services intensity and faster adoption in financial and professional services.

Capital deepening and demographics

AI is also reshaping the structure of production by increasing capital intensity. Firms are investing heavily in hardware, data infrastructure and software systems, driving significant capital deepening. This raises output per worker but also shifts the balance between capital and labour. On its own, AI capital expenditure is unlikely to be very labour-intensive – most of the cost is equipment-related (chips, cooling, power systems), and relatively little labour is required to operate data centres once built.

At the same time, AI may help offset demographic headwinds. Many advanced economies face ageing populations, which reduce labour supply and constrain potential growth. By automating tasks and augmenting human capabilities, AI can sustain output even as the workforce shrinks. This mechanism is particularly evident in healthcare, manufacturing, logistics, and elder care. Empirical evidence shows that ageing is strongly correlated with robot adoption, particularly in Japan, Germany and South Korea.

Labour market impacts

AI is often portrayed as a force for job destruction, but a growing body of research points to a more nuanced picture: limited broad-based disruption so far, with changes occurring primarily at the task level rather than through wholesale job elimination. Several studies show modest changes in hiring for AI-exposed roles, with effects concentrated in specific segments. These include some displacement, especially among younger or entry-level workers in highly AI-exposed occupations or routine cognitive roles. Some of this displacement is offset by productivity-driven growth elsewhere, particularly in analytical, technical and creative jobs. To date, the impact has manifested more as a slowdown in hiring demand than as large-scale layoffs. For example, one study¹ found that openings for routine, automation-prone roles fell by 13% after the release of ChatGPT, while demand for more analytical, technical and creative jobs grew by 20%.

In terms of labour supply, there is no clear evidence yet of significant AI-driven shifts in aggregate participation or willingness to work. AI could influence labour supply through several channels: by reducing the time burden of household chores (potentially increasing participation), improving job search and matching efficiency, or lowering barriers to skill acquisition through tutoring tools. Conversely, it might reduce labour supply if AI-enhanced leisure becomes more appealing or if job insecurity discourages workforce entry. The primary observable effect so far is concentrated among younger workers, where slower training pathways could indirectly affect long-term labour supply by limiting traditional career entry points and on-the-job training opportunities.

Inflation and price dynamics

AI affects inflation through multiple, and sometimes opposing, channels. In the near term, AI capital expenditure in the US has already created acute price pressures in electricity, memory chips, transformers and construction materials – key inputs for the data centre buildout. Since this capex is not labour-intensive, it is not creating wage pressures. We estimate AI is contributing around 0.3 percentage points to core personal consumption expenditure (PCE) inflation in the US. Over the longer term, supply-side effects could lead to disinflation through capacity expansion and softer unit labour costs.

AI’s disinflationary impact is likely to be particularly strong in services – the most persistent component of inflation. Since many service activities rely on cognitive tasks, they are highly exposed to AI-driven productivity gains. Evidence from firm-level studies, including work summarised by the International Monetary Fund (IMF), points to clear cost savings in finance, healthcare, and customer services. Conversely, AI may have ambiguous effects on competition: it can increase price transparency and reduce search costs (lowering markups), but it may also reinforce market concentration as firms with superior data and computational resources gain pricing power. In such cases, cost savings may not be fully passed on to consumers.

Power, compute and energy constraints

The physical infrastructure required to scale AI is emerging as a binding macro constraint. Three interconnected bottlenecks deserve attention: power demand, generation and compute supply. These constraints argue for (i) a more inflationary near-term profile than a pure productivity story would suggest, (ii) significant capital expenditure spillovers to utilities, midstream gas, nuclear and grid-equipment manufacturers, (iii) tension with decarbonisation trajectories, and (iv) a growing role for energy availability as a determinant of where AI capacity is built – with consequences for the geographic distribution of productivity gains across regions and countries.

(continue reading below)

Indeed, obstacles to data centre construction in the US – due to social resistance, regulatory hurdles and resource constraints – have kicked in just as demand is being driven up by growing corporate adoption of AI and the emergence of agentic AI platforms. This mismatch between supply and demand is reflected in the rising cost per token, which is the basic unit of data that an AI model processes.

Monetary policy implications

AI introduces four distinct challenges for monetary policy.

The first is measurement. AI-driven improvements in product quality, customisation and digital services are imperfectly captured in consumer price index (CPI) and GDP statistics. As highlighted by the IMF (2024), this could understate real output and overstate inflation by 10–30 basis points annually, complicating output-gap and natural rate (r*)² estimates.

Second is the natural rate. The direction is contested and is an active debate inside the Federal Reserve. Former Federal Reserve Chair Jerome Powell has argued that AI likely raises r* in the near term, because the demand side – the massive physical buildout required to power AI – is running ahead of any productivity payoff, and the disinflationary benefits of AI remain theoretical for now. Federal Reserve Chair Kevin Warsh has suggested instead that stronger productivity from AI can support non-inflationary growth, giving room for rate cuts without sparking inflation. Our base case sits closer to the Powell framing: a modest +25 basis point drift in r* over the next decade, front-loaded as capital expenditure runs ahead of productivity gains. The countervailing forces – profit concentration, rising inequality, higher precautionary savings – are real but unlikely to dominate over a 10-year horizon. This argues for a higher terminal Fed funds rate than in pre-2020 cycles.

Third is the Phillips curve. Algorithmic pricing, increased price transparency, and the substitution of capital for labour in services may further flatten the Phillips curve (Bank for International Settlements, 2025), making inflation less responsive to traditional measures of slack and more responsive to supply-side shocks (e.g., energy, semiconductors, transformers). This raises forecast uncertainty and the value of structural – rather than purely cyclical – inflation indicators.

Fourth, the transmission mechanism. AI-intensive firms rely heavily on intangible capital, internal cash flow and equity financing rather than bank credit. This weakens the interest-rate channel – large hyperscalers continue investing through hiking cycles – and shifts marginal sensitivity toward the cost-of-equity and financial-conditions channels. Central banks may need to monitor equity valuations and credit spreads more closely as policy transmission proxies and may find that conventional rate moves have asymmetric effects across the capital- and labour-intensive segments of the economy.

Conclusion

AI is likely to become one of the most consequential macroeconomic forces of the coming decades. Its effects will be neither uniform nor immediate; rather, they will unfold gradually through multiple channels.

Our central view is that AI will gradually lift productivity and exert disinflationary pressure – particularly in services – while raising near term capital expenditure-driven price pressures in electricity, advanced semiconductors, and construction inputs. However, the transmission will be slow: as with previous general-purpose technologies, measurable TFP gains typically materialise with a five- to 10-year lag after the deployment phase, reflecting implementation costs, organisational adaptation and skill acquisition.

We therefore incorporate conservative estimates in our macro forecasts: 0.5 percentage points annual productivity boost in the US over the next decade, with smaller uplifts of 0.1 to 0.3 percentage points in Europe and Japan, respectively. The risks around this baseline are two-sided but skewed to the upside over a 10-year horizon.

As with previous general-purpose technologies, the ultimate macro impact of AI will depend not only on the pace of technological progress, but also on the ability of economies, institutions and physical infrastructure to adapt.

¹ Harvard Business School, Enhance or Eliminate? How AI Will Likely Change These Jobs, February 2026.

² The natural rate is defined asthe real short-term interest ratethat neither stimulates norslows down economic growth. It represents the equilibriumrate at which the economy operates at full employment and inflation remains steady.