# Morgan Stanley Warns Storage Stocks Face Cycle-Top Valuation Risk

By Enna Lee

Canonical URL: https://www.tokenpost.com/news/investing/30942
Published: 2026-10-13T08:30:26.000Z
Updated: 2026-10-13T08:30:26.000Z
Section: Investing

> Samsung and SK Hynix trade at roughly three to four times forward earnings, but supply growth and changing AI infrastructure spending could pressure future returns.

Morgan Stanley warned Oct. 13 that apparently cheap valuations for Samsung and SK Hynix may signal a cyclical peak rather than a durable buying opportunity, as new supply and shifting AI infrastructure economics threaten future returns.

The two South Korean memory-chip makers trade at roughly three to four times forward earnings. Morgan Stanley said that valuation can look attractive when profits are near their highest point, but memory stocks have historically peaked before earnings did — by about four months in 2008 and 2012 and by roughly eight months in 2018.

The memory market remains supported by AI demand, tight supply, higher prices and improving profitability. However, Morgan Stanley said those favorable conditions may already be reflected in share prices after Asian memory stocks rose about ninefold from September 2025. Average earnings-per-share expectations have increased 432%, concentrating optimistic assumptions in current valuations.

The report said DRAM contract prices would need to rise another roughly 70% from the previous quarter to support the upside implied by current stock prices. That would be a difficult hurdle, it said, as high profits are likely to attract additional production.

Supply is expected to begin returning gradually in the second half of 2027, potentially creating a new turning point for the memory cycle. The industry’s strongest fundamentals may therefore make the cycle-top risk harder to recognize, Morgan Stanley said.

AI infrastructure spending may also distribute benefits differently across the semiconductor supply chain. In Nvidia’s NVL72 configuration, memory accounts for about 29% of the value of a single rack. When the system expands to eight interconnected racks, memory’s share falls to about 19%, while networking equipment rises from 10% to 20%.

That shift means memory manufacturers may not benefit in proportion to overall growth in AI infrastructure investment. The outlook for Samsung and SK Hynix will therefore depend not only on AI demand, but also on pricing, capacity additions and how future data-center architectures allocate spending.