Login
Sign Up
Woofun AI notes that Gavin Baker, founder and CIO of Atreides Management, identified DRAM as the most critical bottleneck for AI performance during a recent All In podcast appearance. Baker emphasized that memory capacity and bandwidth are fundamental constraints for all AI models.
Baker stated that while SK Hynix may offer cost-effective alternatives for consumer-grade demand, only Samsung, Micron, and Hynix possess the capability to produce DRAM for AI servers. He projected that DRAM will constitute 30% to 40% of total capital expenditures in hyperscale data centers by 2027.