Marvell Technology reported strong growth in its data center and artificial intelligence (AI) businesses in its fiscal Q3 2024 earnings call, even as other segments like carrier infrastructure and enterprise networking declined.

One of the leading networking silicon vendors, Marvell Technology has grown in recent years in large part via a series of acquisitions. In 2021, Marvell acquired Innovium for $1.1 billion to boost Marvell's enterprise capabilities. That acquisition was on the small side compared to the $10 billion acquisition of data center chip vendor Inphi in 2020 and Marvell's $6 billion acquisition of enterprise chip vendor Cavium in 2017.

Even with all its technology assets, Marvell is facing some challenging markets for its solutions.

For the third quarter of its fiscal 2024, Marvell reported quarterly revenue of $1.42 billion, which is an 8% year-over-year decline. Marvell's results were led by its data center segment, which accounted for 39% of total revenue. Within the data center, AI solutions saw robust demand, which is where the company is hoping to see future growth.

However, Marvell is cautioning that near-term economic headwinds in carriers and enterprises will impact upcoming quarters.

"As expected, revenue from the enterprise on-premises portion of our data center end-market declined sequentially in the third quarter, reflecting weakening demand," Marvell CEO Matt Murphy said during his company's earnings call. "Demand for data center storage also remains depressed, and industry expectations for recovery have continued to push out.

Accelerated computing, cloud and genAI are driving Marvell networking chip prospects

With the rapid growth and excitement around generative AI in 2023, Marvell is increasingly capitalizing on strong traction for its high-speed optical interconnects, which are used in AI accelerators.

"In cloud, revenue from both AI and standard cloud infrastructure grew sequentially, with AI growing significantly faster," Murphy said. "Growth was broad-based, led by our PAM4 optical products, Teralynx Ethernet switches, as well as our data center interconnect (DCI) products."

The PAM4 is an optical networking digital signal processing (DSP) technology. Murphy noted that Marvell is seeing strong demand for 800 Gb/s PAM4 electrical-optical products, which is tightly correlated to the growth in deployment of AI accelerators.

Marvell is also seeing growth in the cloud with the development of customized silicon for cloud providers.

"Cloud customers remain focused on enhancing their AI offerings by building custom compute solutions of their own, and we have already won a number of these designs," he said.

It's not just about AI; ethernet switching is still a solid market, too

Marvell's switching portfolio is gaining adoption, including its 51.2 Tb/s Teralynx cloud switching platform, which is based on open standards and was first announced in March.

"Customers have started development on our 51.2 Tb/s solution, and we look forward to ramping this platform into production," said Murphy.

Marvell is also joining the Ultra Ethernet Consortium (UEC), which is a multivendor organization tasked with the goal of advancing Ethernet networking for high-performance computing (HPC) and AI workloads.

"This is another step in our commitment to driving continuous innovation on an open Ethernet-based cloud fabric, which can deliver the scale and performance required for next-generation workloads including generative AI," Murphy said.

Continued decline in enterprise networking for Marvell Technology

While cloud and AI are going well for Marvell Technology, the enterprise networking segment is not. The enterprise networking market accounted for 19% of Marvell's Q3 revenue but disappointed with a 17% sequential drop. Weakness in enterprise is expected to persist in Q4.

"As we have been signaling, we see weak demand in this enterprise market," commented Marvell Technology CFO Willem Meintjes. "As a result, for the fourth quarter of fiscal 2024, we project enterprise networking revenue to decline in the mid-single digits sequentially on a percentage basis."