The debate about artificial intelligence almost always revolves around one simple promise: machines will do more work for us, and do it faster; costs will fall, and prices will follow. The logic sounds persuasive until productivity growth is broken down into its component parts and we examine how it actually passes through the economy. The conclusion is far less convenient: greater efficiency does not guarantee deflation. Under certain conditions, it can fuel inflation – precisely when governments and central banks least expect it.
The key distinction is between a one-off jump in productivity and a sustained acceleration in its growth. If productivity rises once and then levels off, the economy receives a temporary gift: more goods and services are produced with the same inputs, their prices naturally decline and, after a while, everything returns to its previous trajectory.
The story is very different when productivity begins to grow faster year after year and people believe the change will last. What shifts is not a one-off income gain but expectations of permanently higher income for years to come – and it is that expectation, rather than today’s paycheque, that shapes spending and investment decisions. Households and companies quickly incorporate future income they have not yet earned into their plans. They spend and invest now. Demand outruns supply, the economy’s equilibrium interest rate rises and, unless the central bank raises its policy rate in step, inflation accelerates rather than slows.
Economists call this anticipatory demand: demand responds to the promise of future productivity before that productivity actually appears in factories and offices. In this sense, today’s excitement over AI – companies reallocating budgets, hiring for future automation and building faster revenue growth into their forecasts – is already boosting demand, regardless of when or to what extent the promised productivity gains eventually show up in the data. The history of major technologies suggests that the gap between investment and measurable results can last for years. Businesses first spend on retraining, process redesign and equipment. The returns arrive only later. Until that gap closes, demand has already increased while supply has not.
This is the lens through which the US information-technology boom of the late 1990s should be reconsidered. Productivity growth was presented at the time as a gift without side effects: more output, less inflation, everyone wins. In reality, core inflation crept higher over the following years until the 2001 recession, because stronger productivity lifted the equilibrium interest rate and monetary policy had to catch up with this new reality by tightening. It is telling that the debate around that boom – much like today’s discussion of AI as a general-purpose technology – usually stops at the question of output growth and rarely moves on to what it means for interest rates and prices.
There is a second trap: speed. The same cumulative productivity gain over ten years can have a completely different effect on prices depending on how quickly it materialises. If the entire gain arrives at once, supply moves ahead of demand, prices for domestic goods and services fall and the equilibrium interest rate barely shifts. The economy simply receives a one-off windfall. If the same gain is spread over time, people begin spending against future income before that income appears in actual output. The result is reversed: prices and interest rates rise. The more gradual and protracted AI’s path into the economy proves to be – and research into the adoption of major technologies suggests that this is the usual pattern – the greater the chance that it will push prices up along the way rather than bring them down.
The third twist concerns where the breakthrough occurs, and here the picture becomes more complicated still. If productivity rises in sectors exposed to global competition – commodities or industrial exports, for example – their prices are already set in the world market, so greater domestic efficiency has little direct effect on them. The additional income generated by these sectors instead spills into domestic demand for housing, services and transport, driving inflation precisely there. The effect is strongest when productivity in the export sector rises rapidly and all at once.
If, by contrast, the productivity gain in the export sector unfolds gradually and is priced into expectations in advance, a different channel opens: the exchange rate. Expectations of higher export earnings in the future strengthen the national currency today, before the gains have actually materialised. A stronger currency makes imports cheaper and cools inflation. Similar logic, but in reverse, played out around Britain’s referendum on leaving the European Union. Markets treated it as bad news for the future productivity of the country’s export sector, and sterling immediately depreciated even though no factory had yet shut down or lost output. The conclusion is simple and uncomfortable: an exchange rate can move on expectations alone, long before an economy has actually produced or lost anything.
When a technological leap occurs in sectors serving only the domestic market – retail, consumer services or construction – the logic is much simpler. Their prices do fall, and the faster productivity rises, the larger the decline. Automation in extraction or export manufacturing therefore does not necessarily make life cheaper for consumers at home and can sometimes do the exact opposite, while automation in services almost always works to reduce prices for the end customer.
For an economy such as Kazakhstan’s, where a significant share of productive capacity is tied to commodity exports and a substantial proportion of goods and technology is imported, these scenarios are not an abstract model but a practical map of risks. Rapid, large-scale automation in extraction and processing could lift incomes in those industries faster than in services. In that case, the logic points not to cooler domestic prices but to stronger inflation through demand for housing, construction and services. Meanwhile, the tenge and import prices could remain relatively stable, masking the pressure building within the economy. If markets instead come to believe in advance that the productivity of Kazakhstan’s exports will grow steadily over an extended period, the effect may operate through the currency. Expectations of future earnings would strengthen the tenge today and pull down the imported component of inflation, creating the illusion that everything is going to plan even as the policy rate remains out of step with the new reality.
In such circumstances, needs to watch more than headline inflation and the exchange rate. It must also track where the equilibrium interest rate is moving as different sectors of the economy become more digital. A technological development of exactly the same magnitude can, depending on the sector, the speed of adoption and whether markets believed in it beforehand, produce either a stronger tenge and import deflation a year later or a hidden overheating of demand for housing and services. An interest-rate decision based solely on the current inflation picture risks falling behind the curve, just as the Federal Reserve did in the late 1990s.
The final conclusion is inconvenient for anyone seeking a simple story about technology. Productivity can be either disinflationary or inflationary, depending on whether it represents a one-off leap or a sustained acceleration, whether it arrives immediately or unfolds over time, which sector of the economy it transforms and whether markets have already priced it into the exchange rate. The only compass that genuinely works amid this uncertainty is neither the current inflation rate nor the exchange rate in isolation, but whether the economy’s equilibrium interest rate has changed. Those waiting for an AI-driven deflationary miracle may first need to prepare for the opposite – and certainly should not lower their guard merely because prices remain quiet for now.
Shyngys Yerbolat, EconomyKZ Research Group, exclusively for EconomyKZ.org
