Micron Executive: Memory Bandwidth Sets the Ceiling for AI Performance, New Capacity Only After 2028

During the Six Five Summit 2026 on September 15, Sumit Sadana, senior adviser to Micron's CEO, said memory capacity and bandwidth have become the primary factors determining AI system performance. He expects meaningful new memory supply will only begin to come online after 2028, as building new fabs takes years and the technology is highly complex.

Sadana noted that decades of consolidation have left only a handful of DRAM makers, while the number of processor designers has surpassed 20, making the supply chain fragile amid surging AI demand. He emphasized that memory subsystem performance and capacity now set the ceiling for AI performance.

To close the gap, Micron is sharply increasing capital expenditure: over $13 billion in fiscal 2025, doubling in fiscal 2026, and more than $45 billion expected in fiscal 2027. The company also raised its planned U.S. investment from $200 billion to $250 billion and accelerated the timeline, with $50.2 billion expected to be invested by the end of 2027. Around 20 expansion projects are underway in Idaho, New York, Taiwan, Japan and Singapore.

However, supply relief will be slow. Sadana said a memory fab involves roughly 2,000 process steps and takes about five months from production start to customer delivery, while new sites require infrastructure, permits and skilled workers. Thus, effective additional supply is only expected to ramp gradually after 2028.

The executive also highlighted a shift to multiyear strategic customer agreements, with clients integrating memory into 5-7 year product roadmaps. Looking ahead, he said humanoid robots could become the largest incremental market for memory, with each unit requiring hundreds of gigabytes of DRAM and terabytes of NAND, which would significantly boost DRAM and NAND demand.