Samsung Memory Warning Shows AI Chip Shortage May Last Into 2028

Samsung has added a new warning to the AI infrastructure story: the memory shortage may not be a short cycle. It could get worse next year and remain tight into 2028, which means the AI data centre boom is now pulling on one of the most basic parts of modern computing.

The warning follows Samsung record second-quarter results, where its semiconductor business was lifted by strong demand from AI servers. In its earnings release, Samsung said its Memory Business achieved another record quarter by prioritising high-value AI server demand. But the more important signal came from the outlook. Samsung expects supply constraints to become more severe in 2027 and persist through 2028.

That matters because memory is not optional. AI servers need GPUs and accelerators, but they also need high-bandwidth memory, server DRAM and storage to keep models running. If memory makers prioritize HBM and server products, consumer devices and smaller hardware makers may face tighter supply or higher prices. The shortage can therefore spread from data centres into laptops, phones, cars and industrial devices.

The industry has seen chip shortages before, but this one has a different shape. During the pandemic, supply chains were shocked by lockdowns and demand swings. This time, demand is being pulled structurally by AI workloads. Cloud companies are signing long-term agreements because they know that model training and inference need large, predictable memory supply. Samsung is responding to where the money is.

This also helps explain why the recent earnings season has felt so infrastructure-heavy. Microsoft, Amazon and Google are all showing stronger cloud demand, while Meta is showing how expensive the buildout can become. Amazon latest numbers showed AWS growth accelerating, but also showed free cash flow pressure from AI capital spending. Samsung memory warning tells us why that spending will remain difficult to control.

The winners are clear for now. Samsung, SK Hynix and Micron sit at the centre of a market where AI demand is outpacing available supply. But even winners have to spend heavily to expand capacity. We saw the same tension when SK Hynix posted record AI memory profit and still faced investor questions about expectations and future investment.

For consumers, the risk is higher device prices or fewer deals on memory-heavy products. For startups and enterprises, the risk is that AI projects become more expensive because cloud providers pass infrastructure costs into pricing. For governments, the risk is that access to AI compute becomes even more concentrated among companies that can secure long-term supply.

The broader lesson is that AI is no longer just a model race. It is a supply-chain race. Whoever controls memory, power, cooling, networking and data centre capacity has a real advantage. Samsung warning is therefore not a small component story. It is a reminder that the AI boom will be limited by physical infrastructure, and memory may be one of the tightest chokepoints for the next two years.