Nvidia (NVDA) continues to dominate the AI chip market, with its graphics processing units (GPUs) serving as the preferred hardware for training and running the majority of AI applications. However, because GPUs are designed for general-purpose computing, a significant portion of their computing capacity can go underutilized when handling highly repetitive and specialized AI workloads.
This efficiency gap has prompted major AI hyperscalers to invest heavily in custom AI chips tailored to their specific needs. As a result, Marvell Technology (MRVL) and Broadcom (AVGO) have emerged as two of the most closely watched semiconductor companies in the custom-silicon space, as both are playing key roles in helping large technology firms develop specialized AI hardware.
Neither hyperscaler builds these custom chips alone. Both Marvell and Broadcom specialize in helping clients design application-specific integrated circuits (ASICs), along with the networking infrastructure required to connect large clusters of them together. Eidellux‘s research team compared both names across customer base, growth trajectory, and business mix to determine which looks like the stronger buy right now.
AI is the growth driver, but not the only business
Both companies maintain significant businesses beyond artificial intelligence. Broadcom (AVGO) generates revenue from a diverse portfolio that includes mainframe hardware and software, virtual desktop infrastructure, and cybersecurity solutions.
Meanwhile, Marvell Technology (MRVL) remains more focused on semiconductors, with networking chips for mobile devices and controller chips for solid-state drives (SSDs) contributing to its non-AI revenue streams.
However, custom AI silicon has become the primary focus for both management teams and investors. The market opportunity for custom application-specific integrated circuits (ASICs) is substantial and continues to grow, and both companies are leveraging their established chip-design expertise to secure a larger share of this expanding market.
Customer rosters: Broadcom holds the stronger hand
The most significant distinction between the two companies lies in their customer base. Marvell Technology (MRVL) primarily develops custom AI chips for Amazon and Microsoft, supporting their Trainium and Maia silicon initiatives.
In contrast, Broadcom (AVGO) counts Alphabet, Meta Platforms, OpenAI, and Anthropic among its customers—a group that includes some of the largest purchasers of AI computing infrastructure in the world.
From a customer-success perspective, Broadcom currently appears to have the advantage. Among the custom AI chip programs currently in deployment, Alphabet’s Tensor Processing Unit (TPU) is widely regarded as the most commercially proven and successful custom-silicon platform. Meanwhile, Amazon’s Trainium and Microsoft’s Maia chips are still working to reach a similar level of adoption and performance.
As a result, Broadcom’s close involvement with the industry’s most established custom AI chip program gives it a notable edge in this area, at least until competing platforms demonstrate comparable commercial success.
Growth estimates also favor Broadcom
Revenue growth expectations provide another advantage for Broadcom (AVGO), and the difference becomes even more noticeable when looking beyond the near term. For fiscal 2026 (ending October 2026), Wall Street analysts currently forecast 66% revenue growth for Broadcom, followed by an additional 62% growth in fiscal 2027.
By comparison, Marvell Technology (MRVL) is expected to grow at a slower pace, with analysts projecting 41% revenue growth in fiscal 2027 (ending January 2027) and 45% growth in fiscal 2028.
While the two companies operate on different fiscal-year schedules, the gap in projected growth remains substantial. As a result, Wall Street appears considerably more optimistic about Broadcom’s near- and medium-term growth prospects, reinforcing its position as the stronger contender in the custom AI chip market at present.
Outlook: Broadcom’s diversified AI exposure looks stronger today
Putting everything together, Broadcom (AVGO) currently appears to have the advantage on the two factors that matter most in this comparison: customer quality and projected revenue growth.
The company benefits from a customer base that includes the industry’s most proven custom AI chip program—Alphabet’s TPU—as well as several of the largest AI infrastructure spenders, while also delivering meaningfully stronger expected revenue growth over the next two fiscal years.
That does not diminish Marvell Technology’s (MRVL) position as a credible way to gain exposure to the growing custom-silicon (ASIC) market. Its relationships with Amazon and Microsoft, both of which continue to invest heavily in AI infrastructure, provide substantial long-term opportunities.
However, Marvell’s current growth outlook and customer-success metrics lag Broadcom’s by a wide enough margin that Broadcom stands out as the stronger investment candidate today.
Investors should remember that the custom AI chip market is evolving rapidly. If Amazon’s Trainium or Microsoft’s Maia chips begin to narrow the performance and adoption gap with Alphabet’s TPU, the competitive landscape could shift significantly in Marvell’s favor.
For now, one of the most important metrics to monitor is quarterly capital expenditure (capex) guidance from major hyperscalers, as their AI infrastructure spending remains the leading indicator of future ASIC demand and order growth for both companies.