Goldman Sachs delivered a striking verdict in its latest US semiconductor third-quarter earnings preview: most segments of the semiconductor ecosystem are set for upward earnings revisions, and the risk-reward ratio has improved markedly.

Led by analyst James Schneider, the team pointed out that over the past two months the Philadelphia Semiconductor Index (SOX) tumbled 11% amid a large-scale risk-off selloff, while the S&P 500 actually gained 4% over the same period. It is precisely this extreme divergence that has shifted pre-earnings positioning from "tactical caution" in the second quarter to a decidedly "more constructive" stance.

Equipment and Manufacturing: WFE Spending Visibility Extends to 2028

Goldman's clearest upward revision was reserved for semiconductor equipment. The report noted that wafer fab equipment (WFE) spending is being locked in early by customers, with industry order visibility now stretching to 2028. Applied Materials (AMAT.US) was named one of the top picks, with a $670 price target: Goldman expects the company to raise its long-term margin targets at the October 13 SEMICON West event and to lay out a path toward a $300 billion WFE market, followed by a strong mid-November earnings report driven by DRAM/HBM and advanced logic demand. Lam Research (LRCX.US), after a roughly 20% share price pullback, now offers a better risk-reward setup, with threefold drivers in its GAA gate-all-around and backside power delivery-related deposition/etch exposure, rising DRAM share, and NAND upgrade demand; Goldman assigned it a "Buy" rating and a $380 price target. By contrast, although KLA Corporation (KLAC.US) is expected to beat on the quarter, it was flagged as a tactical underperformer because WFE spending is skewed toward DRAM and its process control exposure is relatively insufficient.

Computing Power: Agentic AI Is Repricing the Value of CPUs and ASICs

Against the backdrop of continually upward-revised cloud capex, Goldman believes both server CPUs (benefiting from rising Agentic AI penetration) and key ASIC projects have upside. AMD (AMD.US) is the name with the largest upward revision this round, with its price target raised from $640 to $700 while maintaining a "Buy" rating. Goldman expects both third-quarter results and guidance to beat, with 2027 earnings per share forecast at $17.10, 7% above Wall Street, with core catalysts being the Helios platform ramping from the fourth quarter, Agentic AI-driven server CPU demand, and improved pricing. Cadence (CDNS.US) is viewed as the name "least afraid of AI disruption," as Agentic tools are driving monetization of EDA software; Goldman expects it to raise its 2026 revenue guidance to roughly 21% year-over-year growth and assigned it a "Buy" rating with a latest price target of $470. Notably, although Qualcomm (QCOM.US) may slightly beat, Goldman believes market expectations for Agentic AI are too high and sees tactical pullback risk.

Analog Chips: The Recovery Is Underestimated, with Industrial and Data Center as Twin Engines

Goldman said bluntly that "Wall Street is still systematically underestimating the slope of the analog chip recovery," and it favors companies with the greatest industrial, aerospace/defense, and data center exposure. Microchip Technology (MCHP.US) is one of its most favored names, with industrial restocking combined with a data center revenue push toward the $1 billion target and a $115 price target. NXP Semiconductors (NXPI.US) focuses on automotive demand (especially in the China market) and a plan to double data center revenue, with a $325 price target; SiTime (SITM.US), leveraging the logic of MEMS clocks replacing quartz and the consolidation of Renesas' clock business, is expected to beat quarterly revenue by about 5%, with a $900 price target. In contrast, although Texas Instruments (TXN.US) has strong results, Goldman believes high inventory and high depreciation will weigh on its gross margin relative to peers, maintaining a Sell rating.

Memory: A "Money Printer" Under Tight Supply, with a Preference for HDD and NAND

In the memory sector, SanDisk (SNDK.US) received a "Buy" rating with a latest price target as high as $2,200, the highest on the entire list: with no new NAND supply in the near term, combined with enterprise SSD penetration at cloud providers and expanding long-term agreement (LTA) coverage, Goldman expects its quarterly revenue to beat the midpoint of guidance by about 6%. Seagate (STX.US) received a "Buy" rating with a $960 price target, leading peers in the HAMR technology transition, and combined with capital return room after deleveraging, it is poised to continue capturing HDD share; by comparison, Western Digital received only a "Neutral" rating, with Goldman explicitly stating it "prefers Seagate."

The "Buy" List

Overall, Goldman issued 12 Buy ratings this earnings season, forming the core basket of its semiconductor allocation: in computing power and EDA, AMD (AMD.US) and Cadence (CDNS.US); in analog chips: Microchip Technology (MCHP.US), NXP Semiconductors (NXPI.US), and SiTime (SITM.US); in equipment and materials: Applied Materials (AMAT.US) on the conviction buy list, along with Lam Research (LRCX.US), Onto Innovation (ONTO.US), Qnity (Q.US), and Teradyne (TER.US); and in memory, SanDisk (SNDK.US) and Seagate (STX.US). Tactically, Goldman is explicitly bullish on Applied Materials, Cadence, and Microchip Technology, while flagging downside risk in Qualcomm, KLA Corporation, and Western Digital; it also maintained "Sell" ratings on Arm (ARM.US), Texas Instruments (TXN.US), Entegris (ENTG.US), and MKS Instruments (MKSI.US), citing stretched valuations, gross margin underperformance, or growth lagging the broader WFE market. Goldman's core conclusion is that the two-month deep pullback has already washed out crowded long positions, while fundamentals—WFE spending pulled forward, Agentic AI driving computing power, analog restocking, and zero new memory supply—are all turning upward together. The third-quarter earnings season will be a window where short covering and upward earnings revisions converge.