AI’s Hidden Bottleneck: Memory Costs Surge as Data Centers Expand

[encrypted]

When people discuss AI infrastructure, the spotlight usually falls on GPUs from Nvidia. But behind the scenes, memory is rapidly becoming just as critical — and far more expensive. As hyperscalers pour billions into new data centers, DRAM chip prices have surged roughly sevenfold over the past year, reshaping the economics of large-scale AI deployment.

Industry experts like semiconductor analyst Doug O’Laughlin and Val Bercovici, chief AI officer at Weka, argue that mastering memory orchestration will define the next phase of AI competition. The ability to move the right data to the right agent at the right time can reduce token usage and dramatically lower inference costs — potentially determining which AI startups survive as infrastructure expenses climb.

[bonk]

AI gold rush but the real money is selling shovels… and apparently RAM.