AI is draining the memory market: APPLE is already preparing for the next shortages
To contact us: editorial@fw.media
TL;DR
After GPUs, a new bottleneck is emerging: memory. AI infrastructure is absorbing a growing share of global DRAM production.
Apple is seeking to secure supplies from Chinese manufacturer CXMT, not to reduce costs, but to guarantee sufficient production volumes.
DRAM prices are expected to rise sharply: by 40% to 50% in the third quarter of 2026, followed by a further 30% to 40% increase in the fourth quarter.
The risk is becoming geopolitical. A decision by Washington targeting CXMT could directly affect iPhone production over the next 18 months.
AI is reshaping the memory market. By 2027, between 15% and 20% of capacity currently allocated to consumer electronics could be redirected towards data centres.
Each new generation of AI servers consumes more memory, with greater volumes of HBM for GPUs and DDR5 for host processors.
Shortages are moving along the value chain. After GPUs, advanced packaging, electricity and data centres, memory is becoming the new limiting factor.
The next critical resources could be optical interconnects, electrical equipment and cooling systems.
This transformation is already benefiting several French industrial groups, including Schneider Electric, Legrand, Soitec and STMicroelectronics.
For two years, the technology industry has been consumed by a single obsession: GPUs. NVIDIA’s accelerators became the symbol of the artificial intelligence rush, triggering a shortage of advanced packaging capacity, an explosion in data-centre investment and a global race for electricity. Another critical link is now emerging, and this time the issue is not computing power, but memory.
Apple is stepping up efforts to preserve its ability to source DRAM from Chinese memory manufacturer CXMT, even as Washington considers tightening restrictions on certain Chinese technology companies.
Market forecasts point to sustained pressure. Jefferies expects DRAM prices to rise by between 40% and 50% in the third quarter of 2026, followed by a further increase of 30% to 40% in the fourth quarter. Prices are expected to continue rising in 2027 before a potential rebalancing from 2028 onwards, provided that new manufacturing capacity actually comes online. Behind this inflation lies a more structural shift: a growing share of the world’s memory production is now being absorbed by artificial intelligence infrastructure.
The market is therefore entering a new phase. After GPUs, electricity and data centres, memory is emerging as the next limiting factor in AI development. For electronics manufacturers, the question is no longer simply how to negotiate the best prices, but whether the necessary components will be available at all.
APPLE reveals a paradigm shift: in the AI economy, availability is becoming more valuable than price
Apple’s behaviour is probably the clearest indication of this change. The Cupertino-based company has distinguished itself through exceptionally sophisticated supply-chain management. Supplier diversification, continuous cost negotiations and upstream investment in certain strategic partners have made this industrial discipline one of the company’s main competitive advantages.
Unlike its 2022 project with YMTC for NAND memory, which was primarily motivated by price competitiveness and supplier diversification, Apple’s discussions with CXMT are not intended to reduce costs. They are an attempt to secure an additional option on a resource that could become scarce.
This makes the CXMT issue critical for Apple. Decisions taken in Washington regarding the US Entity List could now directly influence the number of iPhones the company is able to produce over the next 18 months.
Artificial intelligence is steadily consuming the world’s memory supply
The current pressure is being driven by a much deeper transformation: the rapid expansion of artificial intelligence infrastructure.
Between 15% and 20% of memory capacity currently allocated to consumer electronics could be redirected towards data centres as early as 2027.
Each new generation of AI servers simultaneously increases the number of installed GPUs, the amount of HBM integrated directly into accelerators and the volume of DDR5 required by host processors. As models grow larger and autonomous agents multiply the number of simultaneous workloads, memory requirements are increasing alongside computing power.
This dynamic mirrors the pattern already observed with GPUs. In 2023, the shortage concerned accelerators. In 2024, it shifted to advanced packaging capacity. By 2025, electrical infrastructure and data centres had become the new limiting factors. In 2026, memory is joining that list.
The AI economy is entering a phase in which bottlenecks are gradually moving towards less visible but indispensable components. After GPUs and memory, optical interconnects, electrical equipment and cooling systems could become the next critical resources.
This global demand is already benefiting French industrial companies such as Schneider Electric, Legrand, Soitec and STMicroelectronics, which supply several of the essential building blocks underpinning the AI economy.
