AI Data Centers Are Driving Up Memory Chip and Laptop Prices, Analysts Say

A $350 hard drive from two years ago now costs $800, if you can find one in stock at all. That's the anecdote The Atlantic used to illustrate a real supply squeeze: AI companies are reportedly buying up as much as 70 percent of the world's supply of high-end computer memory to feed data centers running ChatGPT, Claude, and similar large language models.

The ripple effects are hitting ordinary consumers. Laptop prices have risen as much as 50 percent on some models, according to The Atlantic, with low-cost, entry-level computers taking the biggest hit. One forecast cited by the magazine suggests affordable computers could effectively disappear from the market by 2028 if the trend continues. The memory shortage is expected to persist for years, not months.

When a handful of companies chase the same scarce component, the market clears at a higher price, and everyone else pays for it. Microsoft, Google, Amazon, Meta, and OpenAI all need the same chips at the same time.

Why AI Is Different From Past Tech Booms

Previous tech booms didn't do this. Streaming video and music now moves unfathomable amounts of data every day. Smartphones put billions of new devices online. Whole industries shifted their computing to the cloud. None of it caused a global memory shortage or a spike in electricity demand on this scale.

The difference, per The Atlantic's reporting, is that generative AI doesn't scale efficiently. In normal tech economics, the cost of serving each additional user drops as a company grows, which is exactly what venture capitalists look for before writing a check. Efficient engineering is supposed to make it cheaper to add the millionth user than the thousandth.

Generative AI hasn't cracked that problem. Model sizes have exploded, from an estimated 175 billion parameters in 2020 to more than a trillion parameters today, according to independent estimates cited by the magazine. Companies like OpenAI and Anthropic keep the exact size of their production models secret, which makes it hard for outsiders to verify these numbers independently, but the trend line is not in dispute: bigger models, more compute, more electricity, more memory.

Data-center capacity in the U.S. is reportedly on pace to multiply eightfold in the coming years, and demand for power at these sites has gotten so extreme that some companies are repurposing jet engines just to keep the lights on. This reflects a scramble rather than normal infrastructure planning.

The Fair Counterargument

There's a real case that this is what breakthrough technology always looks like in its early, expensive phase. Railroads, the power grid, and the internet itself all required massive up-front capital spending before costs came down. AI companies would argue that today's inefficiency is a temporary cost of a genuinely new computing paradigm, and that chip designers, including Nvidia and its competitors, are already racing to build more efficient hardware specifically for AI workloads. If that hardware race succeeds, the per-user cost problem could shrink the way it has in past technology cycles.

That argument deserves to be taken seriously. But it's a bet on the future, not a description of where things stand today. Right now, the inefficiency is real, it's measurable, and consumers are the ones eating the cost increase at Best Buy and on Amazon.

What Happens Next

None of this is illegal, and no regulator has announced an investigation into AI companies' memory purchases. This is a market outcome, not a scandal. But it is a trillion-dollar bet that the current approach to building these models—throwing more parameters and more chips at the problem—will eventually pay off in efficiency gains the way past computing revolutions did.

If it doesn't, consumers keep paying more for basic hardware, entry-level computers keep getting squeezed out of the market, and power grids keep straining to feed data centers that, according to The Atlantic's reporting, still haven't figured out how to do more with less. The open question is whether chipmakers and AI labs can engineer their way out of this before the shortage The Atlantic describes becomes permanent rather than temporary.