By Neha Sahni, head of CIO thought leadership and global market strategist at HSBC
Artificial intelligence often gets labelled as either inflationary or disinflationary. The reality is more nuanced, depending on where the economy sits in the adoption cycle. In the near term, AI can add to inflation pressure, thanks to a capex-heavy buildout that’s colliding with real-world constraints in chips, data centres and power.
Over time, if firms truly redesign how work gets done (rather than simply bolting new AI tools onto old processes), AI can flip from an inflation impulse into a productivity dividend. As such, AI’s inflation impact is time dependent, and hinges on whether the technology is efficiently deployed to change how organisations operate. For investors, the direction of travel matters, but so does the sequencing.
Capacity scarcity in the near term
In the next 12-to-24 months, AI’s inflation impulse is best understood as a supply and capacity scarcity story:
First, the buildout is tangible, urgent and resource intensive. A wave of data centre construction boom increases demand for advanced semiconductors, memory, networking equipment, and cooling infrastructure at a pace that supply chains cannot always match. When demand for memory chips and related components surges, bottlenecks form, and prices rise.
These pressures rarely stay confined to “AI hardware” as the same components are embedded in a wide range of everyday goods — from smartphones and laptops to vehicles and industrial equipment. Higher input costs can spill into prices via higher producer prices and, ultimately, to consumer prices.
Second, energy consumption is another near-term pressure point. Scaling compute requires large and reliable electricity. In regions where energy grids are constrained, additional demand can translate into higher electricity prices or higher investment costs for energy generation and transmission.
Even if the impact differs by region, energy sits near the base of the inflation pyramid: it hits household bills directly and raises operating costs across sectors, amplifying the inflationary impulse from the hardware buildout.
Third, the early-stage economics of AI tend to look more like investment rather than efficiency. Many firms initially spend on hardware, cloud capacity, software, and scarce talent without an immediate payoff in productivity.
When costs rise faster than output, businesses face a familiar decision to either accept the pressure through margin compression or pass some costs on through repricing. Either path can keep inflation stickier than it would be without the AI capex surge.
Shifting to disinflationary in the medium-to-long term
Over a three-to-five-year horizon and beyond, the story can invert. As adoption broadens with AI thoroughly embedded into everyday workflows, the inflation picture starts to resemble a positive supply shock. The mechanism is straight forward. If the same workforce and capital base can produce more output – or the same output at lower cost – unit costs fall. That’s inherently disinflationary even in a healthy demand environment, because it expands the economy’s supply potential.
Crucially, productivity dividend isn’t merely about automating tasks. It also comes from reducing error rates, speeding up cycle times, personalising customer solutions, and enabling better decisions at scale. These gains can show up as lower operating costs, faster throughput, and improved capacity utilisation — all of which ease price pressure.
The labour market channel also matters. As AI adoption spreads – particularly across service sectors – firms may slow hiring and rely less on incremental headcount. Even without mass redundancies, weaker wage bargaining power can dampen wage-driven inflation. If productivity rises while wage growth cools, unit labour costs fall, reinforcing a disinflationary impulse.
What to watch: Signposts for the ‘regime change’
The transition from an “input-cost shock” to a “productivity dividend” is unlikely to arrive with a clear announcement. It will emerge in data and corporate behaviour. These signposts include firm-level adoption metrics which show evidence that AI use is moving from pilots to scaled deployment across functions rather than remaining concentrated in tech teams.
Margins and pricing behaviour are also a sign, if firms begin reporting cost savings and improved efficiency, pricing pressure could ease, even if revenue holds up. More broadly, grid bottlenecks, regulation, or permitting delays could prolong the inflationary phase; efficient investment and expanded capacity could shorten it.
Lastly, any change in hiring, and wage trends in exposed sectors, specifically slower hiring, flatter wage growth, and changes in job design, which can signal that AI is altering labour demand.
Investing in AI through the phases
For investors, the timeline matters, because different parts of the market may lead in different phases. Exposure to the opportunity set can be captured through the AI buildout and various parts of the AI ecosystem, where demand is more visible.
Meanwhile, investing in the “real economy” beneficiaries, such as materials and industrials that supply the infrastructure, equipment and automation underpinning the buildout will be key to capture further opportunities.
AI’s growth ultimately depends on reliable power, resilient supply chains and infrastructure investment. Therefore, energy and resources may also stand to benefit as the cycle progresses.
The bottom line
AI may add to inflation pressures in the short run as capacity is built and constraints bind. But if adoption becomes widespread, and effective and embedded in how businesses operate, there is a credible path to a productivity-driven, more disinflationary outcome over the medium term. The opportunity, as ever, lies in being positioned for the current phase, while keeping sight of the transition and the consequences that follow.
