1Q26_ Quarterly Outlook Report_Final_EN - Flipbook - Page 37
T H E P LUMB LI N E | A RETU RN TO F I RS T PRI N CI PL ES
surrounding its adoption. Several factors could materially alter the trajectory outlined above,
chief among them physical and energy-related constraints.
Energy availability represents a key potential bottleneck. According to estimates from the
International Energy Agency, global electricity consumption from data centres is expected to
double by 2030 as AI workloads expand. Scaling constraints extend beyond electricity supply
alone and include limitations around cooling capacity, grid infrastructure, and specialized
materials. These physical constraints can slow the pace of AI deployment, delaying widespread
diffusion, and extending the period during which early infrastructure owners capture a
disproportionate share of economic value. In such a scenario, the capital deepening phase
discussed earlier may persist for longer than expected, allowing narrow market leadership and
AI-driven equity outperformance to endure. Once these constraints begin to ease, the pace of
adoption could accelerate sharply, with the second derivative of AI diffusion inflecting higher.
Data centre power demand is expected to surge
1000
TWh
800
600
400
200
0
2020
2022
Other infrastructure
Cooling
2024
Other IT equipment
2026
2028
Conventional servers
2030
Accelerated servers
Source: IEA, Scotia Wealth Management
In the near term, energy and infrastructure constraints may also prove inflationary, particularly if
rising power demand coincides with insufficient supply.
Innovation is (eventually) disinflationary
Historically, major technological innovations have ultimately proven disinflationary by lowering
marginal production costs and compressing prices over time. We expect similar results for AI,
though the path toward that outcome may not be smooth or immediate
In the near term, AI adoption may introduce inflationary pressures. Rapid growth in demand for
data centres, compute capacity, energy, and specialized physical inputs such as advanced chips
will be met with finite supply, temporarily lifting inflation as productivity benefits remain confined
to a narrow subset.
As AI diffusion broadens and productivity gains extend beyond early infrastructure owners,
marginal production costs should decline. Since the post-dotcom era, corporate profits have
consistently outpaced labour costs, with unit labour costs rising more slowly than output. When
productivity and profits grow faster than unit labour costs, the marginal cost of production falls,
36
