Phone-grade DRAM chips cost $7 at the start of 2025. By mid-November, the same chip cost over $30. That fourfold jump in a single year is not a market anomaly. It is a policy outcome. And nobody is framing it that way.
The story being told is a supply chain story. Memory is scarce, costs go up, manufacturers pass them along, consumers pay more. Inevitable, impersonal, nobody's fault. Counterpoint Research now projects average smartphone prices will rise 6.9% in 2026, nearly double their earlier forecast. Gartner goes further, estimating smartphone prices up 13% by year's end if the DRAM surge continues. Budget phones, the under-$200 devices that billions of people in emerging markets rely on, face production cost increases of 20 to 30%. Meanwhile, U.S. nominal wages grew 4.3% over the same period. The math does not work out in your favor, and it works out considerably worse if you are poor.
This is not a supply chain story. It is a power story.
Who Decided the Data Center Gets the Memory First
Data centers will consume over 70% of the world's high-end memory chips produced in 2026. That is not a natural law. That is a series of capital allocation decisions made by Samsung, SK Hynix, and Micron, the three companies that control more than 90% of global DRAM production. They looked at their order books, saw that hyperscalers like Microsoft, Meta, Google, and Amazon were paying premium margins for high-bandwidth memory to power AI infrastructure, and did exactly what companies do: they followed the money.
IDC describes this as a "permanent reallocation" of supply capacity toward AI. Every wafer allocated to a data center GPU stack is a wafer denied to a mid-range smartphone or a budget laptop. This is a zero-sum game with a clear winner. The winner is not the person in Lagos or Manila trying to buy their first reliable smartphone.
Follow the incentives. Samsung earns roughly 60% margins on the high-bandwidth memory going into AI servers, versus 40% on commodity DRAM going into consumer devices. Micron announced late 2025 it was discontinuing its consumer-facing Crucial brand entirely to focus on AI supply. The market is doing exactly what the market was designed to do. The question is whether anyone with regulatory authority thinks that outcome requires a response.
The People Paying for AI's Infrastructure Don't Get to Use It
Consider a family in Nairobi or Jakarta who budgeted $180 for a new Android device this year. That price point is gone. Manufacturers cannot sell below cost, and budget bill-of-materials costs are up 25 to 30%. The options are: pay more for a phone, accept a device with downgraded specs (industry analysts are already warning that budget phones may revert from 8GB to 4GB of RAM in 2026), or hold onto a deteriorating handset longer. IDC expects basic smartphone buyers to exit the market five times faster than premium buyers this year.
Apple and Samsung will weather this. Their margins are thick enough to absorb the shock, their supply contracts long enough to buffer the spikes, their customer bases wealthy enough to pay whatever the new price is. Tim Cook told investors in January that the memory cost increases had "minimal impact" on Apple's gross margin. He is not the one being priced out.
The structural irony here is precise. The communities with the least to gain from AI tools, the people who are not using ChatGPT to write investment memos or generate code, are bearing a disproportionate share of the cost of building AI's infrastructure. IDC says even when the shortage resolves, memory prices are not expected to return to 2025 levels. This is not a temporary disruption. It is a permanent transfer.
Nobody is asking the obvious question: if the AI buildout imposes documented, measurable costs on consumer hardware markets, on the budgets of the world's most price-sensitive technology buyers, should the companies driving that buildout bear any of that cost? The CHIPS Act funds new semiconductor manufacturing. The EU AI Act concerns itself with model outputs. Somewhere in between, billions of dollars in consumer harm is occurring, attributable to specific corporate decisions, with no policy framework even attempting to address it.
New memory fabs won't come online at scale until 2027 or 2028. Until then, the question is not whether this technology works. It is who it works for, and who is quietly paying the bill while being told this is just how supply chains go.
