Marvell Ships 1.6T DSP, Broadcom Qualifies Rival Chip

Marvell Technology has started volume shipments of its 1.6-terabit Ara DSP, built on TSMC's 3-nanometer process. The chip targets the PAM4 optical interconnect market that connects servers inside AI data centers.

Broadcom is qualifying a competing 1.6T DSP. The company holds the dominant share in the current 800-gigabit generation. Design wins at the largest cloud operators–Microsoft, Google, Amazon, Meta–will determine who captures the higher-margin share of the transition.

Data-center operators face power and space constraints. A more efficient optical layer buys them time before the next architecture refresh. Marvell says its Ara DSP cuts power per bit by 30% versus the prior generation.

Broadcom's counterargument rests on scale and integration. Its Tomahawk switch silicon and DSP portfolio share a common software stack, which simplifies qualification for hyperscalers running at massive port counts. Marvell counters with a merchant-silicon model that lets customers mix and match vendors, avoiding lock-in to a single switch ecosystem.

The design-win cycle for 1.6T is already underway. Cloud buyers typically qualify two to three DSP vendors per generation to maintain supply optionality. Both Marvell and Broadcom can win. The revenue split matters for margin trajectory.

Marvell's Alpha Score sits at 70/100, reflecting moderate positioning in a competitive end market that is growing. Broadcom scores 57/100, weighed by a broader product mix that includes slower-growing legacy businesses.

The next catalyst is qualification at the largest hyperscalers. Marvell's first-mover status on 3nm gives it an edge in power efficiency. Broadcom's existing 800G relationships shorten the qualification path. Cloud operators typically finalize DSP qualifications in the second half of the year. Both companies are positioned to win a share of the socket.