Marvell Technology is best known for designing custom chips, the kind of bespoke silicon that powers everything from cloud infrastructure to enterprise storage. But a KeyBanc analyst just made a compelling case that the real prize isn’t the chips themselves. It’s the connective tissue between them.
Analyst John Vinh identified a $30 billion total addressable market in scale-up networking by 2030, split roughly between $20 billion in optical technologies and $10 billion in switching capabilities. He raised Marvell’s price target by 48% to $385, a vote of confidence that sent a clear signal: optical networking isn’t a side hustle for Marvell anymore, it’s potentially the main event.
The numbers behind the networking bet
Marvell’s financial trajectory has shifted dramatically. The company now targets approximately $12 billion in revenue for fiscal year 2027 and $18 billion for fiscal year 2028. Combined, that’s roughly $30 billion across two fiscal years, up from previous guidance of about $20 billion.
The data center interconnect segment is doing much of the heavy lifting, with expected year-over-year growth exceeding 70% in FY2027. The driver is straightforward: as AI accelerators get more powerful, the bottleneck shifts from compute to connectivity. Training a large language model across thousands of GPUs requires moving absurd amounts of data between chips, and traditional electrical connections are running into physical limits.
Marvell is pursuing this through two complementary approaches: near-packaged optics and co-packaged optics. Near-packaged optics places optical components close to the processor but on a separate substrate. Co-packaged optics goes a step further, integrating optical transceivers directly onto the chip package itself. Both reduce latency and power draw compared to pluggable optics modules sitting at the edge of a server rack.
Acquisitions building the moat
Marvell hasn’t been content to build these capabilities organically. In December 2025, the company acquired Celestial AI for approximately $3.25 billion, a deal that brought advanced silicon photonics technology in-house. Silicon photonics is the engineering discipline of building optical components using standard semiconductor fabrication, which makes them cheaper and easier to scale than traditional discrete optics.
A separate acquisition of XConn for roughly $540 million added chiplet interconnect technology to Marvell’s toolkit. Chiplets are smaller, modular pieces of silicon that get assembled into a larger package, and connecting them efficiently is one of the harder unsolved problems in modern chip design.
Marvell has expanded its partnership with NVIDIA to cover custom silicon, optics, and NVLink, NVIDIA’s proprietary high-speed interconnect used to link GPUs in AI training clusters. The company also secured a multi-year supply agreement with Google, another major buyer of custom networking silicon.
Why optical networking matters now
Every major hyperscaler is building out clusters with tens of thousands of accelerators that need to communicate at terabit-per-second speeds. Electrical copper connections degrade over distance and generate substantial heat at high data rates. Optical connections don’t have these problems, at least not at the distances relevant inside a data center.
Marvell’s stock has reflected this narrative. Shares have climbed more than 180% in 2026, buoyed by multiple analyst upgrades and the company’s improved financial outlook.