Chipflation Explained: How AI’s Memory Appetite Is Driving Up the Cost of Phones and PCs

Memory chips, the components that power everything from smartphones and laptops to AI servers and medical devices, are at the centre of a growing price crisis that analysts have begun calling "chipflation." A recent report by Morgan Stanley describes it as the shift away from the historical trend of falling chip prices toward sustained, structural price increases that now pose a risk across the global digital economy.

For decades, memory got cheaper every year. The price of a gigabyte of DRAM — the most common type of working memory — fell by roughly a factor of ten every five years between 1957 and 2020, according to Morgan Stanley. That trend has sharply reversed. Memory prices have risen more than six-fold over the past year alone, the report said, and a new global supply crunch is making it worse.

What began as a supply constraint inside data centres is now influencing hardware costs, cloud spending and corporate technology budgets, according to the brokerage.

The idea sits at the intersection of three major themes highlighted in recent reports from Morgan Stanley and TS Lombard: the massive financing requirements of AI infrastructure, the sustainability of the current AI investment cycle and the growing shortage of advanced memory chips.

Morgan Stanley estimates that the AI era could ultimately require around $10 trillion of infrastructure investment, spanning data centres, chips, networking equipment, power generation and electricity grids. The bank argues that funding itself is unlikely to be the main obstacle because global capital markets, private credit firms, pension funds and sovereign wealth funds collectively have access to far larger pools of capital. The challenge, it says, is building enough infrastructure fast enough.

At the same time, AI spending continues to accelerate. TS Lombard estimates global AI capital expenditure could approach $800 billion in 2026, with the United States accounting for more than 80% of the total. Spending by major technology companies on AI infrastructure alone is expected to exceed $700 billion next year.

That investment wave has created extraordinary demand for semiconductors, particularly memory chips.

Why Memory has Become the New Bottleneck

Much of the discussion around AI has focused on graphics processing units (GPUs), particularly those supplied by Nvidia. Morgan Stanley argues, however, that memory is emerging as an equally important constraint.

Modern AI systems require large quantities of High Bandwidth Memory (HBM), a specialised form of memory that enables AI chips to process vast amounts of data at high speed. As AI models become larger and more complex, memory requirements increase dramatically across individual chips, servers and entire computing clusters.

The problem is that HBM consumes a disproportionate share of manufacturing capacity. According to Morgan Stanley, HBM could account for roughly 34% of leading-edge memory wafer capacity by 2028, up from about 6% in 2023. As a result, capacity that might otherwise have been used for conventional memory products is increasingly being redirected toward AI applications.

This shift has helped drive a sharp increase in memory prices, as memtioned above, memory chip prices have risen roughly six-fold over the past year as manufacturers prioritise higher-margin AI-related products.

The Emergence of a Two-Tier Market

One of Morgan Stanley’s central arguments is that AI is creating a two-tier memory market.

Large cloud providers and AI companies are increasingly securing supply through long-term agreements. These contracts allow them to lock in future production capacity years in advance. Meanwhile, companies that do not have such arrangements must compete for a shrinking pool of available supply.

The beneficiaries include memory manufacturers such as Samsung Electronics, SK hynix and Micron Technology, which together dominate global memory production. Suppliers are increasingly allocating capacity toward AI customers because those buyers offer stronger pricing, larger volumes and longer-term visibility.

For other industries, the consequences could be significant.

Morgan Stanley’s modelling suggests that prioritisation of AI-related demand could result in a 15% shortfall in PC memory supply and a 12% shortfall in smartphone memory supply by 2027. The estimated gaps are equivalent to memory requirements for roughly 58 million PCs and 134 million smartphones.

The effects of chipflation are not confined to consumer electronics.

Morgan Stanley says higher memory costs are increasingly affecting sectors ranging from networking equipment and industrial systems to automobiles and healthcare technology. The bank also argues that enterprises may experience the impact indirectly through higher cloud-computing costs as data-centre operators absorb rising component expenses.

Reuters reported that the brokerage views the phenomenon as more than a temporary supply squeeze, describing it as a potentially durable shift in the balance between semiconductor supply and demand.

The Larger Question for AI

While Morgan Stanley focuses on infrastructure constraints and memory shortages, TS Lombard highlights a different issue of whether AI investment will ultimately generate sufficient economic returns.

The research firm argues that much of today’s AI-related revenue remains concentrated within the technology ecosystem itself, with spending by cloud providers, chipmakers and AI developers circulating through the same network of companies. Long-term sustainability, it says, will depend on broader adoption by businesses and consumers and the ability of AI applications to generate measurable productivity gains.

The AI investment cycle is expanding rapidly, creating unprecedented demand for infrastructure and semiconductors. As that demand intensifies, memory chips have emerged as one of the industry's most constrained resources, and the source of what Morgan Stanley calls chipflation.