Jensen Huang Said Nvidia Will Be the First Customer for HBM4. Here's the AI Memory Stock That Has Reportedly Locked Up 70% of Those Orders.

The market is still treating AI as a compute story, but the scarce asset is now qualified memory allocation. That shifts value from the most visible platform owner to the supplier with the deepest design-in relationship: if one HBM vendor secures disproportionate share, the payoff is not just revenue growth but better pricing, tighter working capital turns, and a higher-quality earnings stream that can justify a rerating versus broader semiconductor peers.

NVDA is not a clean beneficiary in the near term; memory tightness is effectively a bottleneck tax on its shipment cadence. Over the next 1-3 quarters, the key risk is that Rubin-related demand outruns HBM supply, which would delay cluster deployments and push some revenue recognition rightward even if end-demand remains intact. The second-order losers are the fast followers trying to buy share with capex: if yields lag, MU can spend into a lower-return ramp while Samsung risks being relegated to residual allocation and lower strategic relevance.

Contrarian view: consensus is underestimating how long the memory constraint can persist. If HBM4 qualification is sticky, the cycle lasts 6-18 months and the winners are the suppliers with locked-in sockets, not the names most exposed to AI headlines. What would falsify the bullish supplier thesis is evidence that HBM4 output ramps faster than expected, or that NVDA shifts volume away from the favored vendor due to yield, reliability, or pricing issues.