SANTA CLARA, California, April 19, 2026, 08:33 PDT
Alphabet’s Google is reportedly negotiating with Marvell Technology on a pair of new chips designed for more efficient AI model performance, according to The Information. The move could push Marvell deeper into the competitive world of custom silicon for major cloud clients. Requests for comment from both Google and Marvell went unanswered, and Reuters was unable to independently confirm the story.
Timing is key here. Alphabet projects capital spending could climb as high as $185 billion in 2026, ramping up its investment in servers, data centers and networking to fuel AI efforts. Google Cloud’s revenue in the December quarter jumped 48%, hitting $17.7 billion. Zacks strategist Ethan Feller called Google a “legitimate hyperscaler,” crediting AI-driven enterprise demand. Over at Marvell, the company in March guided toward almost $15 billion in fiscal 2028 revenue, with the AI surge boosting its forecast. Reuters
The chips in question include a memory processing unit intended to operate with Google’s own tensor processing unit (TPU), which powers its AI efforts. There’s also a fresh TPU on the table, this one built for inference—the phase where a model responds to prompts instead of training. According to the report, Google and Marvell are pushing to lock in the memory-chip design by next year, setting the stage for test production.
Google’s relationship with Broadcom runs deep. Just this month, Broadcom announced a long-term deal to work on and deliver future versions of Google’s custom AI chips and related parts, with the agreement stretching through 2031. Reuters has also reported that both Broadcom and Marvell assist cloud clients in developing custom data center chips, as more buyers seek out options other than Nvidia’s general-purpose AI processors.
Marvell has kept its tone positive. President and COO Chris Koopmans described spending from major cloud customers as “still growing massively.” Back in March, CEO Matt Murphy pointed to bookings “continuing to grow at a record pace,” following a string of design wins that reached a new peak in fiscal 2026. Reuters
Nvidia has tightened its relationship with Marvell. The two firms, in a March 31 press release, revealed a $2 billion investment by Nvidia, while CEO Jensen Huang declared, “The inference inflection has arrived.” Reuters later cited Emarketer analyst Jacob Bourne, who noted the tie-up expands Nvidia’s access to more specialized silicon, along with Marvell’s optical interconnect tech. Marvell Technology, Inc.
Marvell finished Friday at $139.69, a gain of $6.37 ahead of Sunday’s report. With Nasdaq only trading Monday to Friday, investors won’t get the market’s reaction until Monday’s open.
Still, discussions don’t guarantee a deal. Reuters wasn’t able to confirm The Information’s story. The two companies aren’t looking to wrap up design on the memory-processing unit until sometime next year, with test production to follow. Broadcom’s current agreement with Google runs through 2031 and hasn’t changed.
Custom chips make up about 10% to 15% of Marvell’s revenue, according to Koopmans last month. Then there’s Meta, which just last week extended its custom-chip deal with Broadcom through 2029—a sign that the biggest AI buyers are racing to secure supply as inference workloads take off.
