The High Cost of AI Memory

**Shawn Kim** (0:00)
Welcome to Thoughts on the Market. I'm Shawn Kim, head of Morgan Stanley's Europe and Asia Technology Team. Today, we're talking about chipflation, when memory chips stop getting cheaper over time, and become more expensive, and even harder to find. It's Monday, June 8th at 3 p.m. in London.
Memory chips are easy to ignore until your laptop slows down, your phone costs more, or your cloud build jumps. Memory is the computer's workspace. It holds whatever the machine needs at that moment, whether that is a web search, a video, a spreadsheet, or an AI model answering a question.
DRAM is a fast memory inside servers, PCs, and phones. NAND is what stores file in a solid state drive. An HPM, or High Bandwidth Memory, is the high performance version sitting right next to the AI chip, helping them move huge amount of data quickly. The last one, HPM, is key, because AI has become intensely memory hungry.
Memory prices have risen more than sixfold over the last year, a sharp break from decades when the cost of DRAM generally kept falling. The pressure is coming from AI infrastructure build outs. We see server accounting for about 59 percent of DRAM demand by 2028 up from 37 percent in 2023 We also see enterprise solid-state drive reaching 65 percent of NAND demand up from 18 percent. And simply put, data centers are taking a much bigger share of the memory pie.
AI memory use is climbing fast at every scale. A newer AI chips uses seven times more HPM than earlier generations. A full system use about 65 times more. Across an entire AI data center build-out, the jump gets even bigger. HPM has gone roughly from 10 terabytes in 2020 to about 18 petabytes in 2026, orders of magnitude more. This demand is running into supply chain that cannot respond quickly. New memory capacity takes years to build, qualify and ramp up. Supply relief is a process, not a switch. And that creates a two-tier market. Large AI and cloud buyers can sign long-term agreements, prepay and secure priority access. Traditional buyers, including PC makers, smartphone makers and industrial hardware companies must compete for what remains. This impacts everyday products. In 2027, we see PC memory demand potentially facing a 15% shortfall equivalent to about 58 million PCs. Smartphones could face a 12% shortfall equivalent to about 134 million units. Companies may have to raise prices, cut specifications, delay launches, and accept lower profits.
The dollar number is striking. We see the memory market growing from about 220 billion US dollars in 2025 to about 890 billion dollars in 2026 Expectations for 2026 memory revenues rose 71% in just three months. That implies roughly 600 billion US dollars of incremental memory revenues in 2026, more than the annual market for smartphones, PCs, or servers, each taken on its own.
The broader economy may not see a significant direct inflation shock. We estimate the direct impact on headline CPI at about 0.1% in 2026, but the pressure is showing up in producer prices, in corporate margins, cloud costs, capital spending plans and delayed technology upgrades. AI has turned memory from the cheapest part of the digital economy into one of its most contested resources. These tiny chips most people never think of may now decide what's get built or delayed and how much we all end up paying.
Thanks for listening. If you enjoyed the show, please leave us a review wherever you listen and share Thoughts on the Market with a friend or colleague today.

**SPEAKER_2** (4:15)
The preceding content is informational only and based on information available when created. It is not an offer or solicitation, nor is it tax or legal advice. It does not consider your financial circumstances and objectives and may not be suitable for you.