By Lyle Daly
Publication Date: 2026-05-04 12:05:00
No company has benefited from the artificial intelligence (AI) boom more than Nvidia (NVDA 0.48%). In 2020, the chipmaker’s market cap was less than $200 billion. In April 2026, it surpassed $5 trillion. And even with all that growth behind it, Nvidia continues to perform well. The stock rose by 7% over the first four months of 2026, outpacing the S&P 500.
But this year, Nvidia’s returns have paled in comparison to those of Marvell Technology (MRVL 0.12%), a fellow chipmaker that’s up a staggering 95%. That difference in their performances reflects a key shift in how major data center operators are spending on AI chips.
Image source: The Motley Fool.
Hyperscalers are turning to custom chips
In the early stages of the AI build-out, Nvidia graphics processing units (GPUs) were the most popular parallel processors when it came to handling the workloads of training and running large language models. They’re still in high demand, but hyperscalers have been developing their own custom AI chips, too, for a few reasons.
These custom chips — broadly called application-specific integrated circuits (ASICs) — are designed to be narrowly suited to a company’s specific processing needs, in contrast to the general-purpose chips from Nvidia. That means ASICs can be made less expensively and operated at lower costs. Moreover, using these chips also allows tech companies to reduce their dependence on a single supplier.
Since 2016, Alphabet has been partnering with the largest ASIC…
