Broadcom Outpaces Marvell as Big Tech Shifts Toward Custom AI Silicon

Investors are increasingly viewing Broadcom (AVGO) and Marvell Technology (MRVL) as the primary alternatives to Nvidia for AI infrastructure. As Big Tech firms seek to reduce their dependence on Nvidia's GPUs, the industry is shifting toward custom AI accelerators and specialized networking hardware.

Broadcom has established a commanding lead in this transition. In Q3 2025, the company's AI semiconductor revenue surged 63% year-over-year to $5.2 billion. This growth is underpinned by Broadcom's ability to secure massive, high-margin custom design wins. Most notably, the company secured a $10 billion order for custom AI chips from an unnamed client, which multiple reports and analysts have identified as OpenAI.

The Push for Custom Silicon

This trend is driven by a fundamental shift in how data centers are built. While Nvidia provides general-purpose AI hardware, the largest cloud providers are now designing their own custom silicon—often referred to as XPUs—to optimize for specific workloads and reduce costs. Broadcom and Marvell are the two primary fabless semiconductor firms competing for these high-margin contracts, acting as the bridge between the tech giants' designs and the actual manufacturing process.

Market Implications

Broadcom's current trajectory suggests a significantly higher immediate ceiling for growth compared to its peers. Its ability to secure these high-margin custom AI contracts positions it as a top-tier investment in the semiconductor boom. The scale of the OpenAI deal demonstrates that Broadcom is the preferred partner for the most ambitious AI deployments currently in development.

In contrast, Marvell is positioned as a longer-term growth prospect. While it remains a key player in the AI silicon market, it has not yet matched Broadcom's scale of immediate revenue acceleration. This disparity highlights the inherent volatility and risk associated with scaling in a fragmented AI market, where a few massive contracts can create a wide gap between competitors.

Future Outlook

Moving forward, the industry will watch whether Marvell can secure a similarly scaled custom chip contract to close the gap with Broadcom. While Broadcom currently holds the near-term advantage, the total addressable market for custom AI accelerators is expanding as more enterprises move beyond general-purpose GPUs. The critical metric for both firms will be their ability to maintain high margins while scaling production to meet the demands of the next generation of AI data centers.