The Coming AI Compute Crunch - Martin Alderson
My saying: Enjo the time when AI consumption is so affordable:
“There has been so much written about the "unsustainable" AI capex recently. Thinking through this recently, it occurred to me that this is potentially the wrong way to think about it, and it's actually more likely, from my research, that we're going to experience a very significant crunch on compute in the coming years.”
Martin Alderson's blog post discusses the impending AI compute crunch driven by DRAM shortages, not capital, and its impact on AI infrastructure growth through 2027. The post highlights the exponential increase in token consumption by users, especially software engineers, and the resulting surge in infrastructure spending by hyperscalers. However, it argues that the ability to deploy this capital is constrained by the limited supply of DRAM, which will support only 15GW of AI infrastructure. This shortage will likely lead to increased dynamic inference pricing and reduced free plan generosity as providers attempt to manage demand.
0 Comments