Memory prices are up sixfold. Companies and consumers will foot the AI bill

Inflation in the digital economy has its own source, and it is not oil. Corriere Comunicazioni published an analysis citing the WEF: memory prices rose roughly sixfold over the past year, by Morgan Stanley's estimate. Within a few quarters that shift has reset the entire cost calculation for hardware.

Gartner's April forecast put the average rise in DRAM prices at 125 percent and NAND at 234 percent in 2026. Analysts saw a possible turning point only in the second half of 2027. The scale has no precedent. Memory sits in every device, from a server with an accelerator to the phone in your pocket.

The effect already shows in demand forecasts. Gartner expects PC shipments to fall 10.4 percent and smartphone shipments 8.4 percent in 2026. Institutional customers pay more for the same hardware. Consumers push purchases back.

A second thread is the concentration of value in the supply chain. According to SIA and Deloitte, processors, memory, networking chips, storage and power chips account for about 95 percent of the hardware value in an AI compute rack. That is why the whole investment is so sensitive to the prices of a few component categories.

The scale of spending is striking in absolute numbers too. By 2028 global data centre infrastructure spending is to exceed 4 trillion dollars, of which as much as 2.8 trillion could go to semiconductor makers. HBM memory, crucial for training and inference, is today a market worth more than 30 billion dollars.

For end users the conclusion is unpleasant but simple: until the supply chain raises output, there will be no relief on price. Memory prices have become a macroeconomic variable. They feed into companies' inflation baskets, into cloud bills and into the IT budgets planned for the year ahead.

The global semiconductor market is to be worth more than 1.3 trillion dollars in 2026, and as late as June forecasts were revised upward, to 1.56 trillion dollars. According to SIA and Deloitte, processors, memory, networking chips, storage media and power chips account for roughly 95 percent of the hardware value in a single AI server rack. These are exactly the components whose prices are rising fastest, which is why their higher cost feeds into the rest of the bill.

The consequences for users show in sales forecasts. Gartner expects PC shipments to fall 10.4 percent and smartphones 8.4 percent in 2026. Devices are simply getting more expensive. Costlier conventional memory, together with HBM chips worth more than 30 billion dollars that AI accelerators absorb, narrows the manufacturing capacity available to the consumer market.