As semiconductor earnings season approaches, UBS analyst Timothy Arcuri has simultaneously highlighted Semtech (SMTC.US) and Marvell Technology (MRVL.US), reiterating “Buy” ratings on both stocks. He believes the two companies are well-positioned in the long-term expansion cycle of artificial intelligence (AI) infrastructure, with data center and optical communications businesses poised to continue accelerating.
Semtech is scheduled to report quarterly results on August 25, followed by Marvell on August 27. In his report, Arcuri noted that while both stocks appear “somewhat expensive” relative to historical valuation multiples, their long-term growth trajectories are clear, and investors should not overlook the structural tailwinds from ramping AI deployments.
For Semtech, Arcuri expects the data center business to deliver another strong ramp, with overall revenue likely approaching the high end of guidance and profitability exceeding the midpoint. He projects October-quarter revenue guidance in the range of $370 million to $375 million (approximately NT$12 billion), with actual results potentially landing at the upper end, driving earnings per share (EPS) close to $0.70 (approximately NT$22).
Arcuri further noted that management has been consistently conservative in its data center growth commentary, highlighting second-quarter expectations of 85% year-over-year growth and accelerating demand through the remainder of the year. However, he believes that with hyperscaler AI deployments scaling up and expanding content-per-unit opportunities, fiscal 2027 growth could exceed 100%, with his model projecting 124% year-over-year growth. He is also bullish on gross margin upside, noting that as the company increasingly focuses on high-value connectivity products, the divestiture of the module business will provide a favorable gross margin tailwind in fiscal 2028. Arcuri maintains his price target on Semtech at $225 (approximately NT$7,100).
For Marvell, UBS believes the company stands to benefit from strong demand for optical communications and networking products. Arcuri noted that recent commentary from hyperscalers including Google, Meta, and Amazon continues to point toward higher AI infrastructure spending. Combined with Nvidia’s (NVDA) sustained Blackwell platform strength and accelerating Rubin platform ramp, both should provide positive read-throughs given Marvell’s optical communications content within the Nvidia ecosystem.
The analyst particularly emphasized that investors may be underestimating Marvell’s custom ASIC business, especially its partnership with Microsoft (MSFT). There are already indications that Microsoft’s Maia custom AI chip procurement volumes could exceed 1 million units, which could layer an additional $1 billion to $2 billion in revenue on top of the approximately $2 billion (approximately NT$64 billion) revenue base Marvell has already guided for.
Arcuri also mentioned that Marvell’s CXL attach solutions at Google are gaining momentum, with average selling prices exceeding $1,000 per chip (approximately NT$32,000), representing another substantial tailwind, with the magnitude depending on assumptions for CXL attach rates. He has also become more positive on the switching chip business, now projecting calendar year 2027 revenue approaching $1.2 billion (approximately NT$38 billion), above management’s prior expectation of approximately $1 billion (approximately NT$32 billion).
Market sources indicate that GF Securities analyst Alicia Yap also addressed Marvell’s optical communications business prospects in her report. She cited Nvidia’s announcement last Friday that its Spectrum-X CPO (co-packaged optics) switch entered volume production the same day, noting that while Marvell appears to have no specific content in that product, its ability to connect hyperscale customers’ custom ASICs to Nvidia’s NVLink technology positions it to benefit from the industry’s broader transition to co-packaged optics. Marvell also counts Amazon (AMZN) Trainium among its custom ASIC customers and received an investment from Nvidia in May to link custom ASICs with NVLink technology.
Marvell shares rose 6.2% on Monday. Beyond the bullish analyst reports, OpenAI’s announcement of a 4.25-gigawatt (GW) data center investment in Ohio also lifted the broader semiconductor sector. By some estimates, 1 GW of compute investment costs between $38 billion and $60 billion (approximately NT$1.2 trillion to NT$1.9 trillion), with the majority flowing to chips, memory, and networking equipment, once again raising market expectations for AI compute demand.
From a valuation perspective, Marvell currently trades at approximately 59 times earnings estimates for the fiscal year ending January 2027, leaving limited margin of safety. However, analysts broadly believe that with AI buildout continuing to advance, the company has the conditions to sustain strong earnings growth.
UBS’s view echoes a recent research report from Morgan Stanley, which argued that while open-weight models compress profits at the model layer, lower costs will drive AI application expansion and inference demand growth. The key determinant of AI commercial value is shifting from model capability to compute efficiency, with GPUs, custom silicon, and cloud infrastructure becoming core competitive advantages. Amazon, Google, and Meta are well-positioned to continue benefiting from the AI infrastructure wave given their cloud service ecosystems, customer bases, and hardware advantages.
Overall, Semtech and Marvell occupy critical positions in data center connectivity and optical communications networking, respectively. UBS’s bullish stance reflects market confidence in the long-term growth of the AI infrastructure supply chain. As both companies prepare to report earnings, investors will closely monitor the actual ramp trajectory of data center businesses, custom ASIC order momentum, and management’s tone on next-quarter guidance—all of which will serve as important litmus tests for the AI investment thesis.