Key Takeaways:
Goldman Sachs trader Shawn Tuteja says the AI trade is splitting, with optical networking, data centers, and neocloud leading while memory and power lag. The divergence follows two weeks of theme-based differentiation after April-to-June gains and July's selloff moved AI sub-themes almost in lockstep.
"One of the most important trends that we're seeing in the market today is the shift from AI training, driving most of the compute, to AI inference, driving most of the compute," said Sung Cho, co-head of public technology investing at Goldman Sachs Asset Management. "Underneath that architecture is a completely different set of chips, optical equipment."
The rotation comes as 90 percent of S&P 500 companies reported second-quarter earnings, with 64 percent beating consensus by at least one standard deviation — among the highest on record — while the market's reaction stayed muted. Stocks that beat by one standard deviation returned a median 33 basis points over the next session, versus a 95-basis-point historical median since 2010, and tech names that beat actually trailed the S&P 500 by 99 basis points.
Tuteja kept his year-end S&P 500 target at 8,000 but expects the index to trade in a narrow range over the next six weeks as the market's shift from rate-hike fear to complacency removes the asymmetric upside that drove earlier gains. He recommends selective AI positioning rather than broad high-beta exposure.
The physical footprint changes with inference. "As you go into inference, you need a lot more data centers that are closer to the customers that they're serving," Cho said. "What's happening is that compute speeds, the processor speeds are no longer the bottleneck. What the bottleneck is, is actually the ability to be able to have chip to chip communication, server to server communication."
The supply-side setup gives the trade duration. "One of the unique aspects of optical is that it's extremely hard to bring new capacity online," Cho said. "The ability for the industry to bring capacity online is going to be somewhat limited and keep that duration of that trade."
Goldman's picks in the space — Lumentum and Coherent — have backed the story. Lumentum's fiscal Q3 2026 revenue reached $808.4 million, up 90.1 percent year over year, with non-GAAP operating margin expanding 700 basis points sequentially to 32.2 percent. Shares closed at $813.51 on Aug. 10, up 120.71 percent year to date and 599.67 percent over one year. Coherent posted Q3 FY2026 revenue of $1.81 billion, up 20.5 percent, with Datacenter & Communications contributing $1.36 billion, up 40.6 percent and now 75 percent of total revenue.
On the technical side, U.S. semiconductor levered ETF assets under management have shrunk to about $99 billion from a $157 billion peak. During July's selloff, investors added roughly $15 billion in excess buying, but August's rebound has seen redemptions as net asset value recovers — the only source of AUM growth.
Since July 29, the weighted average implied volatility of Goldman's AI leaders basket (GSTMTAIP) has fallen more than 10 volatility points, a 14.3 percent drop, versus an 11.8 percent decline for the S&P 500 ex-AI basket (SPXXAI). Still, AI implied volatility premium over the broader market remains at its highest since early 2023, making options-based upside protection expensive.
Memory (DRAM, NAND, HDD) investment logic is shifting from average-price and margin-driven earnings upgrades toward stability, long-term agreements, and capital-return-driven valuation expansion, Goldman tech analyst Peter Callahan noted. Power themes face additional pressure from midterm-election policy uncertainty and ERCOT regulatory questions.
Tuteja's call is not bearish but a call to lower expectations on pace. Optical networking, data centers, and neocloud remain attractive, while memory and power require more careful selection. As the market moves from fearing the Fed to complacency, the asymmetric opportunity has narrowed, and investors need to reassess AI exposure under finer theme selection and more disciplined risk management.
This article is for informational purposes only and does not constitute investment advice.
