The investment thesis is straightforward: every large language model training run and every inference query requires substantial memory bandwidth. As AI deployments scale from pilot to production, hardware procurement volumes grow accordingly. DRAM suppliers sit at a chokepoint in that supply chain.
The flip side falls on the buyers. Cloud AI providers and startups with heavy compute dependencies will face materially higher infrastructure costs in H2 2026.
Earnings risk is concentrated among AI SaaS companies with large GPU and memory footprints. Those that locked in supply contracts or hedged hardware costs earlier in the cycle are better insulated. Those that did not may miss estimates as capex assumptions embedded in their models prove too conservative.
Corporate investment strategies are adjusting. Hyperscalers with balance sheet capacity are accelerating procurement to avoid peak-cycle pricing. Smaller AI companies face a harder choice: pay elevated spot prices, delay scaling, or raise capital to secure supply.
For semiconductor investors, the near-term read is constructive for suppliers. Micron and SK Hynix hold pricing power in a demand-constrained market. The longer-term question is whether AI infrastructure spending sustains at current levels or whether a capex pause — once major deployments stabilize — reverses the memory cycle.
For now, the market is betting on sustained demand.
