Catenaa, Monday, April 20, 2026- Marvell Technology’s stock jumped by 5% on Monday after reports of developing custom AI chips for Google Cloud.
The Information reported over the weekend that the chipmaker is in talks with Alphabet’s Google. Shares of Alphabet were down about 1%.
The reported talks center on two distinct components. One would be a memory processing unit built to operate alongside Google’s existing tensor processing units, while the other involves a new TPU tailored specifically for AI inference tasks.
The companies could finalize the memory chip design as early as next year, before shifting toward test production.
A deal could see Marvell design new custom AI chips for Google, which could help them better compete with offerings from AI chip leader Nvidia, the report said.
A deal with Google would add to Marvell’s wins this year, after a run-up in its stock.
Marvell’s Monday surge is the stock’s latest boost from partnering with a Magnificent Seven member, after a $2 billion investment and partnership with Nvidia sent shares soaring at the end of last month.
With Monday’s jump, Marvell shares have gained more than 70% since the start of the year. Alphabet has added about 7% over the same period, compared to the broader S&P 500’s close to 4%.
For Google, the collaboration with Marvell represents another step in positioning its in-house TPU hardware as a viable alternative to the graphics processors sold by Nvidia.
Google has recently expanded external sales of its TPU chips and launched a software initiative to make the hardware compatible with PyTorch, a widely adopted AI framework.
Marvell stands to benefit from deeper integration into a major cloud provider’s supply chain. The potential arrangement comes as technology firms continue seeking diversified sources for the advanced silicon required to power data centers and generative AI models.
However, JPMorgan said reports that Marvell had “won” Google’s TPU business were false, arguing that Alphabet is still engaging multiple semiconductor partners, while Broadcom remains the clear incumbent in TPU-related design work.
Wedbush analysts had noted that such a collaboration would strengthen Marvell’s position in custom silicon for hyperscale cloud customers and could contribute meaningfully to revenue starting in 2027.
The conflicting reports highlight ongoing uncertainty around which semiconductor firms will ultimately benefit from the next wave of custom AI silicon spending, even as hyperscalers ramp up investment in specialized chips for both training and inference workloads.
