Nvidia (NVDA +1.33%) has been the undisputed winner of the artificial intelligence boom.
Its graphics processing units (GPUs) became the foundation of modern AI infrastructure, powering everything from ChatGPT to autonomous driving systems. And Nvidia's data center revenue has exploded over the past few years as cloud infrastructure giants raced to build massive new AI clusters packed with thousands of GPUs.
But the AI infrastructure market is evolving, and the next phase may not be dominated solely by ever-larger GPU clusters.
Instead, some of the world's biggest technology companies are increasingly turning to custom-built AI chips known as application-specific integrated circuits, or ASICs.And that shift could create major opportunities for two semiconductor companies in particular: Marvell Technology (MRVL +13.29%) and Broadcom (AVGO +3.24%).
Why AI infrastructure is evolving
If you're new to the chip space, think of a GPU as a Swiss Army knife.
It's designed to handle a variety of tasks that require heavy parallel processing power, and it's particularly well suited for training large language models and supporting a wide range of AI applications.
An ASIC is different. It's designed to perform a specific type of task as efficiently as possible. Components that are extraneous to that task are not part of its design.
Image source: Getty Images.
For years, companies relied heavily on Nvidia GPUs because AI was still a developing technology, and flexibility mattered most. But as AI workloads become more predictable, many technology giants are designing chips tailored specifically for their own systems and the workloads they're expecting to see commonly, as this provides them with significant cost and efficiency benefits.
You see, custom chips consume less power and deliver better performance for their specific workloads. Some analysts estimate that custom AI chips can reduce ownership costs by up to 65% for large-scale inference workloads. Meanwhile, according to a study by TrendForce, shipments of AI ASICs are expected to grow about 44.6% in 2026. Shipments of GPUs, on the other hand, are expected to grow by roughly 16.1%.
Why Marvell is positioned to benefit
Marvell has quickly become one of the semiconductor industry's most important custom AI chip designers.
It works directly with hyperscalers to design custom AI accelerators and networking chips optimized for their specific needs. And that business is growing rapidly.
In its fiscal 2027 first quarter, which ended May 2, Marvell reported record revenue of $2.42 billion, up 28% year over year. Data center revenue climbed to $1.83 billion, and non-GAAP (generally accepted accounting principles) earnings came in at $0.80 per share, while operating cash flow reached a record $639 million.
NASDAQ: MRVL
Key Data Points
The company now expects fiscal Q2 revenue of approximately $2.7 billion, representing roughly 35% year-over-year growth at the midpoint.
Marvell's role in AI infrastructure has become so significant that Nvidia actually invested $2 billion in Marvell and partnered with the company on future AI technologies.
Why Broadcom may be an even bigger winner
Broadcom is already one of the dominant players in custom AI silicon.
The company has spent years helping hyperscalers develop AI accelerators and networking hardware. Broadcom has also worked closely with Alphabet (GOOG 1.41%) (GOOGL 1.32%) on its Tensor Processing Units (TPUs), which are among the most successful custom AI chips.
In its most recent fiscal quarter, Broadcom generated $10.8 billion in AI semiconductor revenue, up 143% from a year earlier, as demand for its custom AI accelerators and AI networking products continued to surge.
NASDAQ: AVGO
Key Data Points
Broadcom's opportunity extends beyond AI chips, as the company also sells the networking hardware that allows processors to communicate efficiently inside massive AI clusters.
As AI infrastructure continues scaling up, networking could become just as important as the processors doing the calculations.
Nvidia isn't losing, but the market is expanding
To be sure, investors shouldn't view the rise of ASICs as a threat to Nvidia.
GPUs remain the preferred solution for training cutting-edge AI models and for handling many advanced AI workloads. Nvidia still controls the majority of the AI accelerator market.
NASDAQ: NVDA
Key Data Points
What's changing is that parts of the AI infrastructure space are becoming more specialized.
Some workloads will continue running on Nvidia GPUs. Others will increasingly shift to custom chips designed specifically for a company's own applications.
That's why Marvell and Broadcom are attracting so much attention. Both companies help the world's largest technology firms build the custom silicon and networking infrastructure that they will need for the next phase of the AI revolution.
Nvidia helped launch that revolution. But as companies focus more on running AI efficiently and at lower cost, Marvell and Broadcom could emerge as two of the biggest winners in its next chapter.
