Particle.news

Musk Says Memory Is the Bottleneck for AI Compute

Markets now view Micron along with a tight HBM supplier base as the likely source of upward pressure on memory prices.

Overview

  • Elon Musk told SpaceX investors on Aug. 4 that “the limiting factor currently is memory,” shifting attention from GPUs and power to high‑bandwidth memory as the main constraint on building larger AI systems.
  • Musk described memory output growing about 20% a year while AI demand rises roughly 200% a year and said that mismatch would push prices higher, but those specific percentages have not been independently verified.
  • The comment focused investor scrutiny on Micron Technology as the only U.S. high‑bandwidth memory (HBM) maker and one of a few global suppliers, a position that could amplify the effect of any supply tightness.
  • Musk also said SpaceX plans to overprovision power and cooling with a target near 15 to 20 gigawatts and warned it expects only a very small share of NVIDIA’s GPUs next year, highlighting trade‑offs in system design and procurement.
  • Analysts are split on the near‑ and long‑term picture: TrendForce projects DRAM price growth will slow in Q3 2026, raising short‑term valuation risk for memory stocks, while Bank of America models large earnings upside for Micron by 2030; HBM production, however, requires advanced fabs and packaging and takes years to scale, which has historically led to boom‑and‑bust price swings that could affect cloud builders, chip buyers, and end users.