Venture capitalist Tomasz Tunguz pointed out that the narrative of AI infrastructure is like a slow relay race, with bottlenecks sequentially transmitting from GPUs to memory, CPUs, and storage. Each link in the chain can freeze the supply chain for the next link for years and lock in higher benchmark costs. At the beginning of 2023, the GPU shock caused H100 rental prices to exceed $9 per hour, and server shipments fell by 22%; subsequently, manufacturers shifted capacity to HBM, causing enterprise SSD prices to rise by 80% in a single quarter, and DRAM prices increased by over 60%.

By the end of 2025, the workload of intelligent agents will push the CPU to GPU ratio to about 1:1, with the average price of server CPUs rising by 27% year-on-year; by 2026, nearline HDD annual capacity will be sold out. The construction cost of data centers has risen to about $20 billion per gigawatt, and orders for long-cycle equipment such as transformers and turbines have been scheduled until 2029 to 2031.

Tunguz referred to this as the long tail effect in the hardware sector: years of manufacturing delays amplify downstream demand shocks upstream. When pressure is relieved at a certain bottleneck, it will be delayed in transmitting to the next link. Transformers scheduled for delivery in 2027 to 2028, NAND wafer fabs, and turbine production lines may face the risk of overcapacity.