Mega-cap CEOs Warn Memory Chip Supply Constrained; NVDA, MSFT, META Lead AI Capex

The AI infrastructure build-out has hit a painful reality: memory is the choke point. In rapid succession, the largest tech firms disclosed that memory supply remains constrained and will stay that way. MSFT, META, GOOGL, AMZN and AAPL each raised the same concern on their earnings calls, signaling that despite massive capex deployment, the memory supply chain cannot keep pace with demand. This is no longer a whisper; it is a unified warning from the companies driving the bulk of AI spending.

The gap between CEO candour and equity pricing is striking. Memory chipmakers, particularly MU, are valued at just 7 times earnings despite being the linchpin of the entire AI capex cycle. If NVDA, MSFT, META and others cannot source enough memory for their data centers and AI accelerators, the ROI on their capex investments erodes. Institutions have been rotating into Mag 7 dip-buys (as evidenced by large single-leg call premium being accumulated in NVDA, TSLA and AAPL), but they may be blind to the upstream supply risk that the biggest tech CEOs are now flagging in public.

The implication cuts across multiple verticals: memory makers stand to gain pricing power, while AI infrastructure plays face margin pressure if they cannot secure supply at the cadence needed. Energy and rare earth materials also feature prominently in the capex mix, but memory is the gating factor that no CEO can circumvent. This narrative could shift valuations in the semiconductor complex if investors begin to price in the supply constraint premium that memory producers deserve.

Sceptics point out that capex has historically proven resilient and that supply eventually catches up. However, if memory supply remains tight through 2026 and 2027, the cost of AI infrastructure will rise, potentially dampening demand from smaller players and reducing the addressable market for hyperscalers' AI services.