Chipflation: AI is pushing chip prices up, and businesses are picking up the tab

For decades the electronics industry could count on an unwritten rule: for the same performance, chips and memory cost less. That rule has broken. The phenomenon now has a name, "chipflation", and it describes the end of the steady fall in component prices.

A third of global sales

Demand for artificial intelligence is driving the change. Gartner data cited by the World Economic Forum shows that in 2025 semiconductors built for AI accounted for nearly a third of the sector's global sales, while HBM memory alone passed 30 billion dollars. Memory has become the new bottleneck in the supply chain. Morgan Stanley analyses cited by the WEF record price increases over the past year.

The mechanism is easy to describe and hard to fix. When production capacity is limited, manufacturers steer it towards the components with the highest margins, meaning those demanded by AI accelerators. Prices then rise even for companies that never use AI at all. Delivery times for PCs, smartphones and consoles stretch out. Cloud service costs climb, and the budget companies can devote to the rest of IT shrinks.

What a square millimetre costs

An analysis picked up by the Chinese tech press puts a figure on the gap. According to Kernel Insight data cited by IT之家, 1b-class DRAM chips cost about 0.654 dollars per square millimetre of silicon, against 0.424 dollars per square millimetre for a TSMC N2 logic process: a 54% gap running against memory. The same survey gives 1b DRAM a density of 0.436 Gb per square millimetre, which places these memories in the sub-10 nanometre node bracket, the one used for DDR5 and for high-bandwidth applications.

"The AI industry cannot exist without memory," observed the chief executive of SK Hynix, according to IT之家.

On timing, Acer chairman Jason Chen said the shortage will not last until 2030, but that prices will keep rising at least until the middle of 2027. That is a window of almost two years, in which the "components" line item will weigh on balance sheets again in a way not seen since the supply crisis of 2020-2022.

Who pays the bill

For Italian and European businesses the practical consequence is twofold. The cost of every project that needs accelerators or large amounts of memory grows. At the same time, room to manoeuvre narrows in ordinary hardware renewals, from laptops to mid-range servers. When memory is the scarce factor, whoever controls production capacity also controls the timing and prices of other people's innovation. Europe has begun to address that at the industrial level, but not yet at the level of procurement.