AI Demand Outpaces Moore's Law: Semiconductor Import Prices Hit Historic Highs

AI compute demand has driven hedonically-adjusted semiconductor and computer import prices to levels with no precedent in the available historical data.[3][1] Import prices rose 3.6% in May 2026 alone and are up 14.4% year-over-year.[3] The increases reflect genuine supply shortages — buyers are paying more per unit of compute, not simply upgrading to better chips — and the underlying data is pre-tariff, meaning actual buyer costs are higher still.[4] A concurrent memory shortage, particularly relevant to AI accelerator supply chains, adds a separate dimension to the constraint picture.[5]

Because import prices are hedonically adjusted to strip out performance gains, a rising index means the cost of a fixed amount of compute is increasing — the direct reversal of the historical norm. For roughly two decades, AI and other compute-intensive workloads got cheaper capacity for free via Moore's Law; that dynamic has now run backward, at least in the near term, and tariff effects layer additional cost on top of the underlying market price increase.

For roughly two decades ending around 2020, hedonically-adjusted import prices for computers and semiconductors fell steadily — by about 52% in total — as Moore's Law reliably delivered more compute per dollar with each chip generation.[1] Hedonic adjustment matters here: the price index accounts for improvements in chip speed and capacity, meaning a rising index cannot be explained away by buyers simply upgrading to faster hardware.[2] When the index falls, compute is genuinely getting cheaper; when it rises, buyers are paying more per unit of compute, independent of product generation.

That index is now rising sharply. In May 2026 alone, semiconductor and computer import prices increased 3.6%, bringing the year-over-year gain to 14.4%.[3] SemiAnalysis describes the magnitude as "so far from anything in the historical record that 'fastest ever' doesn't do justice to it," with AI demand identified as the primary driver.[3] The framing is that AI compute demand is growing faster than Moore's Law is reducing costs in the short run — the industry is consuming chips faster than fabrication technology can cheapen them.[1]

Critically, SemiAnalysis draws a distinction between two mechanisms that could produce rising hedonically-adjusted prices: demand for higher-performance chips (partly absorbed by hedonic adjustment) versus genuine supply shortages that push up the cost of existing compute capacity. Their argument is that the current situation reflects the latter — actual scarcity, not a product mix shift.[4] A separate sardonic observation about "the great memory shortage of 2026" points to high-bandwidth memory as a specific bottleneck, directly relevant given HBM's role in AI accelerators.[5]

One additional layer is policy-driven cost. The import price data cited is calculated before tariffs are applied, so the 14.4% year-over-year figure reflects underlying market prices only.[4] A White House presidential action in January 2026 adjusted semiconductor import tariffs under Section 232,[6] and the ITIF has since argued those tariffs could undermine U.S. economic growth.[7] The full cost burden on buyers — particularly U.S. AI infrastructure builders — therefore exceeds what the headline import price index captures. Earlier in 2026, 14 suppliers raised prices and extended lead times,[8] consistent with the broad supply-constraint picture the import price data reflects.