Investing Legend John Templeton Has a Warning for Micron and SK Hynix Investors

The market is still treating AI-memory as a higher-quality version of the old cycle, but the core issue remains the same: capacity is fixed in the short run and then brutally elastic with a lag. That creates a classic peak-earnings trap where gross margins look durable right until depreciation, new lines, and incremental competition hit the same demand pool; once that happens, EPS can fall much faster than revenue. The biggest second-order loser is not just MU or SKHY, but also AI-system buyers whose bill of materials is becoming more memory-intensive, which can quietly squeeze accelerator economics and capex ROI.

Near term, the crowded long is the vulnerability. After a strong rerating, even a modest shift in commentary toward "balanced supply" or higher capex can compress the multiple before fundamentals roll over; the market does not need a demand collapse to punish these names, only a lower implied terminal margin. Over 1-3 months, the key falsifier for the bear case is continued contract price firmness into the next print; over 6-18 months, the risk is that the industry builds too much capacity just as hyperscaler growth normalizes, recreating the exact oversupply dynamic investors think AI has abolished.

The cleanest relative-value expression is to own the capex beneficiaries rather than the commodity itself. Equipment suppliers should capture the early-cycle spend even if memory ASPs later mean-revert, while MU/SKHY carry the more direct downside if utilization slips. The contrarian nuance is that longer-dated supply agreements and hyperscaler concentration may indeed extend the cycle versus consumer memory, so this is more a valuation/cycle timing trade than a structural short on AI memory itself.