AIToday

AI chip memory shortage fuels semiconductor ETF opportunity

Yahoo Finance AI1h ago
AI chip memory shortage fuels semiconductor ETF opportunity

Key takeaway

Artificial intelligence accelerators from Nvidia and AMD are demanding unprecedented amounts of high bandwidth memory—Nvidia's Rubin GPU needs 288 gigabytes of HBM4 and AMD's MI400 series up to 432 gigabytes—creating supply shortages expected to persist beyond 2027. The iShares Semiconductor ETF (SOXX) gives investors diversified exposure to memory producers like Micron and equipment suppliers driving the expansion of manufacturing capacity, though it trades at a premium valuation of more than 66 times earnings as of July 21.

Summaries like this, in your inbox every morning.

Sign up free →

3 Key Points

  • What happened

    Nvidia's new Rubin GPU requires up to 288 gigabytes of high bandwidth memory (HBM4)—nearly three times what its predecessor Blackwell used—while AMD's MI400 accelerators need up to 432 gigabytes, creating acute shortages across the memory market. The iShares Semiconductor ETF (SOXX) offers exposure to memory producers like Micron and equipment suppliers like Lam Research and Applied Materials that stand to benefit from the need to expand production.

  • Why it matters

    Micron management expects supplies of DRAM and NAND to remain tight beyond 2027, and expanding output requires major investments in new chip foundries—a process slowed by construction timelines, worker shortages, permitting, and energy needs. The ETF's nearly 21.4% allocation to semiconductor manufacturing equipment makers positions investors to capture the broader memory opportunity without betting on a single producer.

  • What to watch

    The iShares Semiconductor ETF closed at $552.69 on July 21, allowing $100 investors to buy roughly 0.18 shares via fractional trading. The ETF trades about 16% below its June 22 peak and at more than 66 times earnings, raising downside risk if AI spending slows, memory prices weaken, or new supply arrives faster than expected.

In Depth

High bandwidth memory (HBM) has become a critical constraint in the AI accelerator market. Nvidia's new Rubin GPU architecture requires as much as 288 gigabytes of HBM4—the fourth generation of high bandwidth memory—compared to nearly one-third that amount in Blackwell, its immediate predecessor. Rival Advanced Micro Devices' MI400 series accelerators demand even more, with designs calling for up to 432 gigabytes of HBM4. These escalating requirements have created shortages across the entire memory market, extending beyond HBM itself.

The bottleneck exists because expanding memory supply requires the few major players in that niche to build large and complex new chip foundries—a process far slower than demand would prefer. Micron Technology's management has provided guidance that supplies of DRAM and NAND will remain tight beyond 2027, citing constraints from long construction times, shortages of skilled workers, permitting requirements, and the need for additional energy infrastructure. Growth in HBM production further pressures the broader memory market, as each new HBM generation requires more wafer supply compared to conventional memory, dividing scarce manufacturing capacity between HBM and non-HBM applications.

The iShares Semiconductor ETF (SOXX) offers exposure across the memory supply chain without concentrating bets on a single producer. As of July 21, the fund held 30 stocks with Micron at 8.33% of assets, directly capturing the HBM, DRAM, and NAND opportunity. Nvidia and Advanced Micro Devices together accounted for another 17% of the portfolio—companies that are increasing the amount of memory they embed in their AI accelerators. Nearly 21.4% of the ETF's assets sit in semiconductor manufacturing equipment makers including Lam Research, which supplies etching and deposition tools for advanced memory production; Applied Materials, which provides equipment for DRAM manufacturing and HBM packaging; and KLA, whose inspection tools help chipmakers detect defects and improve yields. Additional holdings in Broadcom and Taiwan Semiconductor Manufacturing provide wider exposure to the AI chip ecosystem.

However, the ETF is not a pure-play memory fund. It does not hold stakes in SK Hynix and Samsung Electronics, two of the three major HBM suppliers, both of which are positioned to benefit from the supply shortage. The fund's top 10 holdings accounted for about 60.8% of assets as of July 21, making it considerably more concentrated than a broad market index. On that same date, the ETF closed at $552.69, offering fractional share access—an investor with $100 could buy roughly 0.18 shares. The recent semiconductor sell-off has improved the entry point, with the ETF trading about 16% below its June 22 peak, though it remains up sharply for 2026 overall. As of July 21, the ETF traded at more than 66 times earnings, a premium valuation that increases downside risk if AI spending slows, memory prices weaken, or new memory supply reaches the market faster than expected. Despite those risks, the fund offers a diversified path to the memory supply opportunity for investors who can tolerate semiconductor volatility and plan to hold for several years.

Context & Analysis

The semiconductor industry faces a genuine supply constraint driven by the architecture of modern AI chips. As AI accelerators embed larger amounts of high bandwidth memory—Nvidia moving from Blackwell to Rubin with a near-tripling of HBM capacity, and AMD's MI400 requiring up to 432 gigabytes—the memory supply chain has become a bottleneck. Expanding production capacity is not a quick fix; Micron's guidance that DRAM and NAND will remain tight beyond 2027 reflects the capital intensity and long lead times required to build new foundries.

The iShares Semiconductor ETF captures this opportunity across multiple points in the supply chain. Its 21.4% weighting in equipment makers (Lam Research, Applied Materials, KLA) directly benefits from foundry expansion, while its 8.33% stake in Micron provides exposure to the memory producer itself. However, the ETF carries structural limitations: it excludes SK Hynix and Samsung Electronics, two of the three major HBM suppliers, and its top 10 holdings represent 60.8% of assets, making it considerably concentrated. At a valuation of more than 66 times earnings as of July 21 and a recent entry point 16% below its June 22 peak, the fund reflects both opportunity and premium pricing—particularly exposed to downside if AI spending moderates or new supply reaches the market ahead of schedule.

FAQ

How much HBM does Nvidia's new Rubin GPU require compared to Blackwell?
Nvidia's Rubin GPU is designed to use as much as 288 gigabytes of HBM4, nearly three times the memory used in Blackwell, its predecessor architecture.
How long will the memory supply shortage last?
Micron management expects supplies of DRAM and NAND to remain tight beyond 2027, constrained by long construction times, worker shortages, permitting requirements, and energy infrastructure needs.
What is the iShares Semiconductor ETF's expense ratio and top holding?
The fund charges a 0.34% expense ratio and owns 30 stocks, with Micron accounting for 8.33% of the ETF's value as of July 21.

Get the latest AI Stocks & Markets news every morning

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Free · takes 30 seconds · unsubscribe anytime

Discussion

No comments yet. Be the first to share your thoughts!

Log in to join the discussion

Related Articles

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime