Artificial intelligence stocks have taken a beating over the past month. While AI-fueled enthusiasm caused several stocks to boom in 2024, investors now appear to be experiencing angst over valuations in the face of potential tariffs, trade restrictions to China, or a digestion period in AI spend.

Two 2024 winners were Broadcom (AVGO -5.39%) and Marvell (MRVL -7.30%). Their portfolios are somewhat similar, so much so that Marvell is sometimes called "baby Broadcom."

However, when each stock reported last week, Marvell plunged, while Broadcom soared. It was an interesting dichotomy and may have reflected a few changing dynamics in the AI races.

Marvell fails to reassure investors

While there are differences between the two, both Marvell's and Broadcom's portfolios are weighed toward infrastructure chips for networking and communications, along with custom ASIC IP that cloud hyperscalers use in their custom AI accelerators, called XPUs.

All eyes in recent years have been on custom ASICs, which have seen hypergrowth as the AI buildout has accelerated. Marvell CEO Matt Murphy noted that two years ago, Marvell projected about $200 million in ASIC revenue in calendar 2023 and $400 million in 2024. But Marvell wound up exceeding $1.5 billion in 2024, with management projecting more than $2.5 billion in 2025!

However, the topic of ASICs is also where things got dicey for Marvell during its conference call.

You see, Marvell's ASIC revenue is highly concentrated in a single customer: Amazon (AMZN -2.36%) Marvell also also works with Alphabet on Alphabet's new Arm-based CPU, but not its AI accelerator, as we'll see.

Although Murphy noted that it has great visibility into ASIC revenue this year and next , one analyst asked about rumors of potential competition for the business of its all-important customer. Murphy answered:

[T]o be crystal clear, we do expect revenue from these custom XPUs with this customer, not only to grow in fiscal '26, which is the year we're in, but to grow in fiscal '27 and beyond. And I just want to also note that I'm not including revenue here that would be from other products, access networking connectivity that would be incremental to custom. And certainly, we've got a great setup, I think, overall with this account relative to the opportunity set, but it's really on both. And then with respect to the question that our customer may be working with someone other than Marvell on a next-generation XPU, it's just something we can't comment on.

That last line was likely the one that spooked investors, especially in a nervous market in which investors seem to be looking for reasons to sell. A lack of a definitive "no," to the rumors about Amazon switching all or part of its custom chipmaking off Marvell's IP may have worried investors over the sustainability of Marvell's ASIC growth.

Image source: Getty Images.

Meanwhile, Broadcom reported more potential customers

On the other hand, not only did Broadcom report blowout AI numbers of its own, but CEO Hock Tan also noted Broadcom had engaged with two additional potential ASIC customers.

Currently, Broadcom has three major ASIC customers. Its biggest is Alphabet for its tensor processing units (TPUs), along with what many believe to be Meta Platforms and China's Bytedance, the owner to Tik Tok. On its previous earnings call in December, Broadcom noted that it had engaged an additional two customers for custom ASICs.

But on last week's earnings call, Broadcom said it has now engaged with an additional two potential custom ASIC customers on top of the previous two. While we don't know who those two new customers of interest might be, it could suggest a potential engagement with Amazon. Of course that's highly speculative at this point.

Before Marvell investors freak out, some industry sources believe the original two new customers announced in December were Apple and OpenAI, while the two new customers alluded to last week may be Oracle and Elon Musk's xAI. But again, those four potential customers have not been confirmed by Broadcom.

In any case, the prospect of having Broadcom add even more customers to its already-larger roster of custom ASIC customers bolstered its stock. Meanwhile, doubts over Marvell hanging on to its single biggest customer longer-term likely dented sentiment.

And don't forget about valuation

In addition to the differing narratives over custom ASIC customers, valuation likely played a role in both Broadcom's and Marvell's recent fortunes. While Broadcom has more custom ASIC customers and a more diversified business with its high-margin software segment, it still traded at a lower valuation than Marvell heading into last week.

Both companies' valuations look extremely high on a GAAP basis, since each has a lot of amortization of intangible assets on its income statements, because of both companies' history of acquisitions. However, unlike the depreciation of property, plant, and equipment, intangibles amortization isn't really an ongoing cost that has to be replaced by capital expenditures, as depreciation is. So in these cases, looking at both companies valuations on a non-GAAP basis is probably appropriate.

Marvell made a little over $1.3 billion, or $1.57 per share on a non-GAAP basis in its fiscal year ended in January, which makes its current P/E ratio on adjusted earnings about 46.8 times. Prior to its report earlier this week, the stock traded at 57.3 times. And at its January highs, Marvell traded at 81.2 times trailing earnings.

Meanwhile, Broadcom currently trades around 36.3 times trailing-12-month adjusted EPS. Heading into earnings, that figure for Broadcom was just 33.5.

Marvell may have been valued higher since it's a smaller company, and most investors seem to think that the smaller a company is, the larger its growth prospects. However, Broadcom has been able to defy those limitations of large numbers. After all, Broadcom's 25% growth last quarter was nearly as fast as Marvell's 27% growth, despite Broadcom being about nine times as large.

Lessons to draw

When comparing two companies, two big elements to consider are customer concentration, as well as valuation. Even though Marvell and Broadcom reported quite similar results, Broadcom's lower valuation and more diversified customer base inspired more confidence relative to Marvell's sky-high valuation and dependence on a large single customer.

As we've seen from the excellent results across the Magnificent Seven stocks, big companies can still find ways to grow, especially since only large companies can credibly participate in capital-intensive megatrends such as cloud computing and AI. So in this new world of transformative technology megatrends, sometimes bigger is in fact better.