Memory Stocks Are Plummetting. Jensen Huang Says the Bears Have It Wrong.

Nvidia CEO Jensen Huang said he does not expect memory oversupply until sometime after 2029, while the article notes memory shares are down roughly a third from their July peak despite strong reported results. Micron reported an 86.8% gross margin and 80.7% operating margin in its latest quarter, but forecast gross margin to ease to 85.95% next quarter. The author argues that Micron’s forward P/E of 5 makes memory stocks attractive for long-term investors, while acknowledging uncertainty about how long tight supply and peak-like margins will persist.

Analysis

The key distinction is memory mix, not simply “more memory.” AI accelerators are pulling high-bandwidth memory capacity and investment toward higher-value products; that can support supplier economics while tightening or displacing investment in other DRAM and NAND segments. But capacity additions, customer qualification and yields eventually convert scarcity rents into supply—often before end-demand visibly rolls over. Huang is a credible demand signal, not an independent supply forecast: NVIDIA benefits from reassuring investors about a critical input, and its claim that more memory can be used does not establish that customers will pay enough to sustain today’s margins.

For Micron, the reported exceptionally high margins make a low forward P/E unusually sensitive to earnings normalization. Treat it as a peak-cycle multiple until free cash flow, product mix and customer commitments demonstrate otherwise; buybacks help only if cash generation persists and repurchases are not made at cycle-high earnings. Near term, sentiment and memory pricing can keep the group volatile. Over 1–3 months, track contract prices, supplier capex, customer HBM qualification and guidance revisions. Over 6–18 months, the key risk is announced capacity arriving alongside slower AI infrastructure spending. Chinese competition is a watch item, but product capability, yields and customer qualification need verification before pricing it as imminent share loss. A sustained fall in memory pricing, inventory buildup, or supplier capacity plans accelerating would falsify the shortage-duration thesis.

Key Decisions for Investors

• Consider a staged, moderate long in MU rather than treating its forward P/E as a normalized valuation. Add only if subsequent results show resilient free cash flow and pricing; reduce if gross-margin guidance starts falling materially or memory contract prices weaken across more than one product category.
• Prefer exposure to the memory supplier with the clearest evidence of HBM execution and durable customer commitments; verify product mix and capex disclosures across MU, SK hynix and Samsung rather than assuming all memory names share the same economics. SNDK’s NAND exposure should not be treated as a direct proxy for HBM scarcity.
• Do not use Huang’s post-2029 supply view as a stand-alone catalyst. Monitor supplier capex and construction, customer qualification, memory contract pricing, and AI customer spending over the next 1–3 months; any combination of rising supply plans and weaker pricing argues for waiting or trimming.
• Keep CRWV separate from the memory thesis: its debt-yield sensitivity is a financing and duration risk, not evidence that memory demand has turned. Avoid a simple CRWV-versus-MU pair unless financing conditions and relative valuation support it; a broad AI spending slowdown could hurt both.