There's no denying that chip-making giant Nvidia (NVDA +4.30%) has been the shining star of the artificial intelligence revolution thus far. After all, its processors are the digital "brains" of most of the world's artificial intelligence platforms.
As the business matures, however, AI data center operators increasingly realize there's a significant cost -- figuratively as well as literally -- to using a third party's off-the-shelf solutions. Not only is using Nvidia's technology simply expensive, but this tech doesn't necessarily do exactly everything its users want it to do.
The solution? Bypass the costly middleman that's limiting your performance anyway, and instead make your own artificial intelligence processors.
That's easier said than done, of course. But that's quickly changing thanks to a relatively small up-and-coming company called Marvell Technology (MRVL 0.76%) that could be the Nvidia of artificial intelligence's next chapter.
Image source: Getty Images.
From support...
It's not a household name. In fact, there's a good chance you've never even heard of it. Its $75 billion market cap isn't exactly head-turning these days.
Give it time, though. Marvell's day in the sun is coming, driven by raw growth.
In simplest terms, Marvell Technology makes much of the connectivity hardware you'll find inside a data center that turns hundreds of computer processors (like those manufactured by Nvidia) into a true neural network -- hardware like ethernet switches or digital signal processors.
Its Orion digital signal processor, for example, is capable of delivering data from one processor's circuit board to another at a pace of up to 800 gigabits per second. That's roughly 800 times faster than the typical fiberoptic broadband connection speed consumers have at home, for perspective. Marvell also makes solid-state hard drive controllers specifically for data centers, where speed and reliable performance are everything.
NASDAQ: MRVL
Key Data Points
And this is nothing trivial. As it turns out, everything else besides data centers' core computing processors has become the industry's performance bottleneck.
That's changing though. As Marvell's Sr. Vice President and General Manager of Connectivity Xi Wang penned just last month, "the optical interconnect global market has doubled since 2020 to nearly 20 billion dollars in 2025, and it is expected to double again by 2030, with an industry CAGR of about 18%." Marvell stands ready to deliver these fiberoptic interconnectivity solutions the AI industry needs to take itself to the next proverbial level.
But there's a kicker.
...to centerpiece
Marvell has primarily focused on creating and providing connectivity solutions that work with artificial intelligence processors from companies such as Nvidia. In AI's infancy, the use of whatever processing platforms were available was good enough; data centers found a way to make it work.
It was only a matter of time before the AI industry learned enough about building platforms to realize it may be better served by making its own custom processors.
Then in 2020, a fairly small outfit called Groq did exactly that, tapping Marvell Technology to design and manufacture a customized application-specific integrated circuit (or ASIC) that would end up being the industry's very first peta-operations-per-second (POP/s) AI accelerator-on-a-chip.
Since then, more and bigger names have followed suit, including Microsoft, Alphabet's Google, and Amazon. Specifically, Amazon is using Marvell's know-how in its Trainium and Inferentia chips, which are being used to train (among others) Anthropic's AI platform (called Claude) at a price that's significantly lower than what it would cost to use comparable tech from Nvidia.
These major players aren't overwhelmingly using Marvell's semiconductors in their artificial intelligence data centers just yet, to be clear. The company would likely have a tough time supplying a massive number of custom ASIC chips at this point in time anyway. This top-down solution is the direction the AI business is moving, though. The proof lies in the company's second-quarter revenue growth of 58% (year over year), largely driven by the 69% improvement in data center sales, which now accounts for three-fourths of its total top line.
This is just a taste of what's likely to come, however. An outlook from Credence Research suggests the worldwide artificial intelligence ASIC market is poised to grow at an average annualized pace of nearly 19% through 2032, jibing with Wang's growth expectation for the interconnect technology that unites all of these custom-made processors into a unified computing platform.
The key to the bullish argument
Marvell Technology isn't the only big name in the customized artificial intelligence chip business. Broadcom (AVGO +0.62%), a larger company, also makes customized AI chips. There's also nothing to prevent another newcomer from becoming a major name in the industry. That would-be rival would simply need to tap Taiwan Semiconductor to manufacture its designs, as Marvell has.
As we've learned from computer processor companies like Intel and even Apple, however, simply building a chip foundry and making your own -- or others' -- silicon is neither easy nor cheap. It's also time-consuming. Don't look for a serious rival to tiptoe onto Broadcom's/Marvell's turf now and then catch up.
Even among just these two names though, Marvell Technology arguably has an edge. As Klover AI's Dany Kitishian recently pointed out, Marvell's custom ASICs combined with its interconnect solutions paired with its ethernet switching tech used in conjunction with its data processing units that ease the burden on a data center's primary processors "creates a comprehensive platform solution that is difficult for component-focused competitors to replicate."
Given this, Marvell Technology's target of growing its 13% share of a market currently worth $33 billion to a 20% share of a market that should be worth around $94 billion by 2028 is not only believable, but arguably conservative.
Analysts are on board anyway. Despite this year's subpar performance, the vast majority of them currently rate Marvell stock as a strong buy.
