Frictionless

Logan Jastremski × Vikram Sekar

Memory—not raw compute—is AI’s deepest systems bottleneck, making lower-memory architectures a potential lever across bandwidth and networking.As 144-GPU systems push power and cooling higher and copper fails at 400 gigabits, optics, DRAM-on-logic and specialized inference architectures may gain relevance, while Anthropic’s S1 points to more than $500 billion in obligated capital and compute and possible over-ordering.

Logan Jastremski × P Equity Research

AI infrastructure is constrained less by GPUs alone than by a shifting stack of memory, advanced packaging, power, construction, and networking, with memory potentially absorbing roughly half of hyperscaler capex through 2028.Long-term agreements and higher utilization may support a higher earnings floor, but cannot repeal cyclicality if ROI and free cash flow disappoint.Watch ABF, optics, and China’s potential 20-25% global memory share.

Logan Jastremski × Bubble boi

AI’s scaling bottleneck is shifting from transistor shrinkage to packaging, interconnect, and memory, as tens or hundreds of billions of dollars may buy only 15–20% better density.NVIDIA’s roadmap increasingly depends on suppliers and technologies it does not control, while rack-scale architectures, flash offload, and smarter caching could determine whether the same hardware serves two or three times more users.