When the CEO of the world’s most valuable company says a component has become “unsustainable” and announces price hikes, it’s worth paying attention. Tim Cook said it about RAM, and it isn’t an Apple-only problem: it’s the symptom of a crisis affecting the entire computing industry.
What’s Happening with Memory
Memory—both conventional DRAM and the high-bandwidth memory (HBM) that powers AI accelerators—has become one of the most contested resources on the planet. The reason is straightforward: training and serving AI models consumes colossal amounts of memory, and manufacturers have redirected much of their capacity toward the most expensive, most profitable chips.
The result is a domino effect. The production capacity that once went to RAM for phones, laptops, and consoles is now being diverted to data centers. Less supply for general consumption, with demand intact, means just one thing: rising prices.
Why AI Monopolizes Memory
Large models don’t just need compute power; they need to feed that power with data at extremely high speed. That’s where HBM comes in, a memory stacked in three dimensions that offers far greater bandwidth than traditional DRAM. Each high-end AI accelerator consumes several HBM modules, and manufacturing it is complex and costly.
Because HBM and DRAM share fabrication lines and materials, the boom in the former chokes the supply of the latter. Manufacturers logically prioritize their highest-margin products. And the margins, today, are in AI.
The Consumer, at the End of the Line
The end user sits last in the priority queue. That translates into pricier laptops, phones that climb in price between generations, and device lifespans that stretch out of necessity rather than choice. When a company like Apple, with enormous negotiating power, admits the cost is squeezing it, the rest of the market will feel it even harder.
What It Means for Those Who Build Software
For those of us who develop, there are practical implications. If hardware gets more expensive and memory grows scarce, efficiency stops being a luxury and becomes a competitive advantage. Optimizing memory usage, avoiding infrastructure over-provisioning, and choosing right-sized models for each task suddenly have a direct impact on costs.
The era of “throw more RAM at the problem” is giving way to one where every gigabyte counts, both in the user’s wallet and on the data center bill.
No Short-Term Relief
Building new memory fabrication capacity takes years and billions in investment. As long as AI demand keeps growing at this pace, there’s no reason to expect prices to ease soon. The paradox is elegant in its cruelty: the technology that promises to make everything cheaper and more efficient is, for now, driving up the cost of the most basic component in any computer.
