Memory has become one of the most closely watched parts of the AI build-out. It is where the data that a processor works on is stored and transferred, and it now sets the ceiling on how much computing can actually be used. Two developments matter for investors: what recent earnings reveal about pricing power, and the growing effort to engineer around the shortage.
The memory hierarchy. Memory sits in tiers: fast and costly at the top, cheaper and slower below. SRAM is the small, ultra-fast layer inside the processor. DRAM is the main working memory, where data in active use resides, and it spans HBM, DDR5, and LPDDR, all of which now power AI servers. NAND is the high-capacity storage layer beneath, cheaper but slower.
Why it matters. Every tier is growing rapidly, and the shortage shows no sign of easing. With supply tight across the memory hierarchy, pricing and margins are strong throughout, as reflected in recent earnings.
What recent earnings show. The leading producers have reported record results, with gross margins now well above 80 percent, versus the high 30s a year earlier, and high-bandwidth memory output sold out for the year. Several have locked customers into multi-year supply agreements, a clear break from an industry long defined by short, violent cycles. The signal is durable pricing power rather than a typical peak. Even so, forward earnings multiples remain modest, roughly 6 to 12 times next-year earnings, because the market still treats these profits as cyclical and likely to revert. The chart below shows how sharply margins have stepped up.
The push for alternatives. The second development, and arguably the more consequential, is the effort to reduce reliance on scarce memory. The approaches fall into three groups:
- Substituting cheaper media: pairing low-cost flash with software so it behaves like working memory, plus emerging high-bandwidth flash designs.
- Reducing data movement: embedding limited compute inside the memory itself and connecting memory to processors optically rather than electrically.
- System and software efficiency: an interconnect standard called CXL lets servers pool and share memory rather than stranding it; tiering software keeps frequently used data on faster layers; and compression and leaner model designs lower requirements at the source.
Implication. The shortage is structural, not a conventional cycle, given multi-year fab lead times and disciplined capacity expansion, and recent earnings confirm the resulting pricing power. But the same dynamic creates the opening for alternatives. The firms that most effectively lower the cost of memory may ultimately capture value comparable to that of the producers of the constrained components themselves.
A caution on positioning. The case for owning memory is strong; the point is simply to stay disciplined. Today’s margins are unlikely to last, and as supply catches up, they should settle back toward more normal levels. That is not the same as the businesses unraveling: even after margins cool, these producers look structurally larger and more profitable than in past cycles. Because this cycle is less driven by traditional supply swings, the items to watch are the durability of AI capital spending and the pace at which new capacity comes online. Should that spending slow, or new supply arrive faster than demand, the pricing power now lifting earnings could reverse. The real risks here are AI capex and the pace of new supply, not the memory cycle itself, and positions should be sized with that in mind.
DISCLOSURES
The information provided is for educational purposes only. The views expressed here are those of the author and may not represent the views of Leo Wealth. Neither Leo Wealth nor the author makes any warranty or representation as to this information’s accuracy, completeness, or reliability. Please be advised that this content may contain errors, is subject to revision at all times, and should not be relied upon for any purpose. Under no circumstances shall Leo Wealth be liable to you or anyone else for damage stemming from the use or misuse of this information. Neither Leo Wealth nor the author offers legal or tax advice. Please consult the appropriate professional regarding your individual circumstance. Past performance is no guarantee of future results.
This material represents an assessment of the market and economic environment at a specific point in time. It is not intended to be a forecast of future events or a guarantee of future results.
