Is AI inflationary?
In our last open publication in early October, we warned that for the near term, much good news on the earnings front had been factored into technology valuations and risks looked skewed to the downside with Softbank (9984) one of our primary go-to shorts as a good hedge against some of the tech that had remained in our longs, mainly memory related names, those material suppliers that are benefiting from memory chip plants running at max capacity and those poised to benefit from the likely big memory capex we should see, naturally following this super cycle.
Although in Japan’s stock market we feel more comfortable now with recommending some AI plays, and memory remains the core of our tech composition, we are increasingly more troubled about the bigger picture. We see increasing evidence that AI may not be just a threat to white collar jobs, but in the nearer term, building these data centres are creating bottlenecks in the supply chain that are raising prices and may prove disruptive to the popular 2026 macro scenario for a more benign inflationary backdrop and thus, lower interest rates.
Looking at memory which is the heart of the AI semiconductor evolution, chip makers have been raising prices and experiencing capacity constraints since late summer as more resources have shifted towards HBM and high-end graphics memory to satisfy current AI data centre demand. This is in turn is setting the scene for a potential memory chip shortages in H2 of next year, especially for traditional DRAM applications such as PCs and smartphones and reportedly currently forcing these hardware makers to raise their prices by as much as 20% to pass on surging memory costs.
Interestingly, hard disk drives are also facing their own capacity constraints which has forced hyperscalers to adopt solid state drives which in turn has also driven up NAND prices. These capacity bottle-necks, also mirror another big AI proxy, TSMC (2330) which dominates AI chip production and supplies almost everyone, but one which faces its own chip and packaging capacity constraints and has raised its foundry quotes many times in the past 3 years for its cutting edge lines.
We suspect shortages we are witnessing in data storage solutions could be the canary in the coal mine, as we see much inflationary pressures brewing from AI infrastructure buildout, either more locally in terms of electric power and water prices, or more globally such as surging value of industrial metals such as copper. With AI investments reportedly having made up the bulk of US GDP growth in H1, we think investors should not underestimate the magnitude of their impact in exerting upward pressure on prices, and in wide array of industries.
From cement to high grade construction steels, to onsite power generators (from diesel to gas turbines), water pumps for cooling systems, and almost anything that is touched by these massive projects looks to be seeing some degree of supplies tightening, lead times extending with a more favorable pricing environment for the related manufacturers.
We fear that we’ll start to see higher prices feeding through from AI investments way before we see data centre overcapacity popping this bubble and slowing such outlays. Time will tell if AI becomes an inflation menace that we suspect it might, but at least for the next 12 months we see such shortages become even more acute and prices likely to rise much further. These trends may become problematic for central banks, especially as they would be considered demand-pull inflation that governors take more seriously than Covid-related supply disruptions that were seen as cost push and thus deemed more temporary.
Interestingly, there does seem to be a realisation among local communities that data centres could raise utility prices as they are always on and once they are built, they are unlikely to attract other tech firms to their locations as many hyperscalers have initially promised. With number of employees typically needed to run these centres also fairly minimal, many in the US are turning against plans for building them in their back yard.
