Semiconductor company Marvell Technology might not get as much love as industry peers like Nvidia, Advanced Micro Devices, or Broadcom — a factor which only becomes more obvious when you compare its stock performance with other industry peers. Its stock is down 18% year-to-date, leaving it largely behind prevailing names in the AI demand.
It might simply be that Marvell — which creates networking and storage chips, server processors, and other products — has been seen as comparatively boring. And while the company counts AI accelerators among the products it builds for the data center, it remains a modest business compared to some other semiconductors.
But by all accounts, the tides might finally be turning. While GPU names like Nvidia and AMD have seized the economy of traders’ attention thanks to hyperscaler spending, a new approach is gaining momentum, promising to be more efficient and cost-effective. In recent weeks, traders have demonstrated greater interest in Alphabet‘s Tensor Processing Units (TPUs), which are co-developed by Broadcom, a company building custom AI accelerators called XPUs.
And if that trend away from GPUs is to continue, Marvell — which makes XPUs like Amazon’s Trainium and Inferentia, as well as Microsoft’s Maia chips — might finally shine through.
Bad, Worse, Great
It seemed that might not be the case after the company’s earnings, though. Despite besting analyst’s earnings expectations in its latest quarterly earnings report and announcing a multi-billion-dollar acquisition of chip startup Celestial AI, the firm’s strong growth wasn’t sufficiently compelling for traders. The company’s stock fell up to 8% after its Tuesday evening report, even as its revenue hit an all-time high of $2.075 billion, growing 37% year-over-year. Non-GAAP diluted income per share was $0.76, rising 77% year-over-year; also a beat.
But as soon as the initial negative reaction came, it was gone — from down 8% to up 8%, and then 11%, and then 15% after hours. A few sentences in the company’s commentary might have been the thing that tripped high-frequency traders, an opportunistic algorithm, or a budding money manager to press ‘buy.’
What Happened?
Marvell noted that expectations for its data center business have dramatically grown since an investor call on Sept. 24, when the company projected data center growth potential of 18% year-over-year next fiscal year. In the weeks since, expectations have nearly doubled, with the company now guiding for 30% year-over-year growth.
What’s more, the company says its custom business — which counts Amazon as a customer, among others — about a quarter of the company’s data center revenue, would grow “at least 20%” next year, which was “higher than prior expectations.”
Rounding out positive news, the company said that it already has purchase orders for a next-generation program, which could be anchored by a large customer making a “transition to a next-generation XPU.”
That would seem to align with the trend in the industry. While GPUs have scored on the backs of massive hyperscaler capex, many tech companies seem to be embracing cheaper alternatives, including chips of their own. In many cases, they no longer see GPUs as ‘the only game in town’ and are embracing alternatives like XPUs.
War of the XPUs
While GPUs remain the industry’s go-to, there’s reasons to believe the future of the AI industry could belong to the XPUs. And while Broadcom might be squarely in the lead by virtue of co-developing Alphabet’s all-powerful TPUs, plus striking partnerships with OpenAI, Meta, and ByteDance to build XPUs, Marvell could muscle its way into the space with competitive leaps and an emphasis on faster ‘optical connections.’