Microsoft's Azure hardware chief Rani Borkar said building massive new memory chip capacity will not resolve the supply chain bottlenecks holding back the AI infrastructure buildout.
Her remarks were delivered at SEMICON Taiwan in Taipei, Nikkei Asia reported, and come from the executive who leads design and procurement for Microsoft's server and AI hardware.
Memory wall
Borkar's diagnosis maps to a wider industry problem often called the "memory wall", where memory bandwidth and system integration, not wafer output alone, leave accelerators idle while waiting for data.
That constraint is concentrated in high‑bandwidth memory and the advanced packaging needed to attach it to AI chips, plus the qualification time for finished modules, creating a system‑level bottleneck rather than a simple chip shortage.
Cloud buyers
Large cloud buyers have absorbed much of the available HBM and packaging capacity, shifting suppliers to tight allocation and higher prices and making marginal wafer increases less effective.
Borkar urged innovation across system design, packaging and bandwidth management as the place to relieve AI infrastructure pressure, a shift that redirects investment away from pure memory capacity expansion.
