As cloud giants increasingly design custom AI chips to lessen reliance on Nvidia, companies like Broadcom and Marvell Technology are quietly capitalizing on this trend. Broadcom has seen explosive AI-driven revenue growth from co-designing accelerators for clients like Google and OpenAI, while Marvell Technology, with Amazon’s cloud arm as a major customer, projects significant expansion in its custom silicon business. Despite high valuations and customer concentration risks, these firms represent key players in the evolving AI landscape, with Broadcom potentially offering a more diversified investment.
While Nvidia remains at the core of the artificial intelligence (AI) surge, providing the bulk of chips for training and operating today's most advanced AI models, a significant shift is underway. Major cloud providers are increasingly developing their own custom processors for specific AI tasks. This strategic move aims to reduce reliance on a single supplier and optimize hardware to perfectly match their unique software ecosystems.
A prime illustration of this trend is Alphabet's Google unit, which has engineered its own tensor processing units (TPUs) to power its AI services. Notably, Google hasn't embarked on this complex design journey alone, hinting at a less obvious yet highly lucrative avenue in the evolving AI landscape.
This shift reveals a quieter, yet incredibly impactful, play: the specialized chip designers assisting these cloud titans in crafting their bespoke silicon. Two such companies are already securing a rapidly expanding share of the colossal AI spending.
Broadcom: Nvidia's Formidable Challenger in Custom AI
Broadcom stands out as a powerful testament to the overwhelming demand for custom AI chip solutions, emerging as potentially Nvidia's most significant competitor. In its fiscal second quarter of 2026 (ending May 3, 2026), Broadcom reported a remarkable 48% increase in revenue, reaching an all-time high of $22.2 billion.
The catalyst for this explosive growth was AI. Revenue attributed to AI chips and networking soared by 143%, hitting $10.8 billion, which constitutes nearly half of its total sales. This figure represents a substantial jump from approximately $8.4 billion in the previous quarter.
A significant portion of this success stems from the custom accelerators Broadcom co-designs for a select group of cloud customers, a roster that now includes industry giants like Google and OpenAI. Broadcom CEO Hock Tan encapsulated the situation perfectly during the company's fiscal second-quarter earnings call, stating, "Demand for XPUs and networking is simply insatiable."
Broadcom has recorded over $30 billion in AI orders during the quarter and has projected AI revenue to more than triple year-over-year in the current quarter. Furthermore, the company has boldly predicted its AI business could reach an astounding $100 billion in annual revenue by fiscal 2027.
Marvell Technology: A Focused Bet on Custom Silicon
Marvell Technology offers a more concentrated investment opportunity in this same burgeoning trend. The company unveiled its fiscal first-quarter 2027 results (period ended May 2, 2026) last month, reporting a 28% surge in revenue to a record $2.4 billion.
Its data center segment spearheaded this growth, climbing 27% to $1.83 billion, accounting for roughly three-quarters of its total revenue. For the full fiscal year ending in January, revenue expanded by 42% to $8.2 billion. Management has forecasted around 40% growth this year, targeting approximately $11.5 billion, with further expansion towards $16.5 billion anticipated the following year.
"We are seeing exceptional AI-related bookings," affirmed Chairman and CEO Matt Murphy in Marvell's fiscal first-quarter earnings release.
Marvell plays a crucial role in designing custom chips for some of the largest cloud operators, with Amazon's cloud division reportedly being its largest customer in this domain. The company expects its custom silicon business to exceed $10 billion in annual revenue by fiscal 2029. Notably, in March, Marvell also expanded its partnership with Nvidia, aiming to enhance the interoperability of its custom processors with Nvidia's systems – a significant collaboration given that Marvell's chips often compete directly with Nvidia's offerings.
Understanding the Market: Valuation and Risks
However, investors should note that both stocks have already experienced substantial appreciation. Broadcom currently trades at a price-to-earnings (P/E) ratio in the low 60s, even after a recent dip from its early June peak of $495. Marvell, which has more than tripled in value in 2026 and is slated to join the S&P 500 by June 22, commands an even higher P/E ratio, nearing 100.
Both companies are not without significant risks. Their custom-chip revenue is heavily dependent on a limited number of cloud customers, leading to customer concentration risk. Moreover, neither company holds a monopoly; Broadcom has openly stated that a major customer might engage multiple suppliers. Additionally, Broadcom's rapidly growing AI sales reportedly yield lower margins compared to its software business.
Given these factors, where do these two stocks stand? Personally, I lean towards Broadcom. Its larger operational scale and diversified business portfolio may offer a greater buffer against potential slowdowns in AI spending, and its valuation appears less demanding in comparison.
