The analysis adds a structural layer to an inflation backdrop already strained by energy, which argues against easing and supports the case for central banks on tightening paths, including the Fed and the Bank of Japan, to keep a hawkish bias into 2027. Japan’s projected 0.3 point boost is notable ahead of this week’s BoJ meeting, where the key question is whether underlying inflation has reached 2%. For bonds, a temporary but well-timed price shock could keep front-end yields supported. For equities, the finding that AI lifts prices more than growth outside Asian supply hubs feeds into investor scrutiny of returns on AI spending. Oil remains the bigger swing factor, but AI now adds to the pressure from the goods side.
—
Speaking of inflation, coming today:
—
AI is showing up in inflation faster than in growth, and Goldman Sachs reckons the price bump will be temporary, though not before central banks have to sit through it.
Summary:
- Goldman Sachs says the AI infrastructure buildout will raise inflation in developed markets, with effects varying by region and fading through 2027
- Memory prices currently add about 19bp to US core PCE; direct pressures elsewhere are around 0.1 point for now
- Peak boost is put at 50bp for US core PCE and 20bp for US core CPI; non-US developed markets see an average 25bp lift to core inflation into early 2027, possibly up to 33bp
- Japan is projected at around 0.3 point and the UK at about 0.2 point
- GDP boost is modest at 0.3% on average across developed markets through 2026, with Australia highest at 0.6%
- Goldman cites the concentration of data centre input production among Asian exporters as limiting wider spillovers
The global buildout of artificial intelligence infrastructure is adding to inflation across developed economies, according to new analysis from Goldman Sachs, which sees the effect peaking at around 50 basis points on US core PCE inflation before fading through 2027. The bank stressed that the impact will vary by region and should prove temporary.
The main channel is direct price pressure from the hardware behind AI. Goldman estimates that memory prices alone are currently adding about 19 basis points to US core PCE, where those components carry a larger weight in the index. Outside the US, direct pressures are smaller for now, at around 0.1 percentage point in most economies, but the bank expects them to build as higher costs pass through to consumer electronics prices, which carry larger weights in non-US inflation baskets than in American measures.
On that basis, Goldman projects that price pressures directly linked to AI will lift year-on-year core inflation in non-US developed markets by about 19 basis points by the first quarter of 2027, compared with a peak boost of 50 basis points for US core PCE and 20 basis points for US core CPI. Taking all effects together, the bank estimates an average lift to core inflation of 25 basis points in non-US developed markets through early 2027, with potential increases of up to 33 basis points relative to long-run trends. Japan stands out at around 0.3 percentage point, with the UK at about 0.2 percentage point.
The indirect effect, through stronger economic activity, looks limited. Goldman estimates the AI buildout will lift GDP across developed markets by a modest 0.3% on average through the end of 2026, with Australia seeing the largest boost at 0.6%, the euro area 0.2% and the UK 0.1%. The bank attributes the small spillover to the concentration of data centre component production among Asian exporters, which keeps much of the economic benefit within a few supply hubs.
The analysis matters because it adds another source of upward price pressure for central banks already dealing with higher energy costs. Goldman’s view that the effect fades as memory price pressures ease suggests policymakers may treat it as a temporary shock, but its timing, peaking into early 2027, overlaps with a period when many are still judging whether inflation is sustainably back near target.
This article was written by Eamonn Sheridan at investinglive.com.
