DRAM makers have been outperforming hyperscalers of late

Chris Wood worris that future commitments on the leasing of data centers like this are liabilities that the hyperscalers are keeping off balance sheet at present.

A growing sense of fatigue with the artificial intelligence trade and the related arms race is prompting one strategist to advise that investors stick with the so-called picks and shovels rather than the spenders.

Jefferies chief global strategist Chris Wood prefers the memory chip makers like Micron $(MU)$, Samsung Electronics (KR:005930) and SK Hynix (KR:000660) rather than the hyperscalers, such as Meta $(META)$, Microsoft $(MSFT)$, Amazon (AMZN) and Alphabet $(GOOG)$ who have committed so much capital towards developing AI, and plan to commit so much more.

The next big chapter in the AI narrative, the next stress test for the industry and its capex plans, is looming as results for the hyperscalers start to appear from July 22.

Wood outlined his recommendations in the weekly "Greed and Fear" strategy note he's been sending to clients for thirty years. In it, he describes the recent correction in the Korean heavyweights, Samsung and SK Hynix, as "both natural and healthy." Wood points out the picks and shovels theme is well-established among investors by now, hence the huge outperformance of the semiconductor stocks versus the hyperscalers.

The memory-chip makers have surged past the hyperscalers

Going forward Wood wants to own the chip makers because he forecasts the demand for compute will continue to grow, even if the cost of tokens - the payment units for AI usage - continues to decline. Moreover, Wood emphasizes that he, like everyone else, "has no idea which, if any, of the hyperscalers are going to be successfully monetizing their AI capex."

He adds: "There's a real possibility that none of them could be."

Despite the signs of AI fatigue, Wood detects no signs of flagging commitment from the hyperscalers.

After the last reporting season, they all received a slight fillip after they upped their proposed capex to a level equivalent to 92% of their cashflow. Wood is concerned that it's not just their cashflow they are consuming so rapidly, but the rate at which they are incurring debt. Those four hyperscalers have issued $169 billion of bonds so far in 2026 and Oracle $(ORCL)$ also sold $25 billion.

The scale of the planned investments worries Wood. In 2027, AI capex could top $1 trillion after around $700 billion in 2026 and on top of this the hyperscalers also have off-balance sheet commitments that the market doesn't quite appreciate yet. These include future data-center lease commitments of $662 billion that sit off balance sheet at present. A recent Moody's report found that off-balance commitments stood at 113% of the published debt outstanding.

Hyperscalers' data center commitments: Undiscounted future lease commitments. Note: Amounts in billion US dollar as of 31 December 2025 SEC filings for each company

So for Wood, these concerns justify recent share-price underperformance of the hyperscalers relative to the S&P 500 SPX and the market's reaction to second-quarter numbers later this month is of paramount importance.

Wood thinks it's also worth noting that the pick and shovel stocks are getting their earnings from the capex boom upfront. The hyperscalers must wait to see for certain if they can generate returns on their investments while for now, the depreciation expenses are rising.

-Jules Rimmer

This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

July 10, 2026 05:31 ET (09:31 GMT)

Copyright (c) 2026 Dow Jones & Company, Inc.