Google in Talks With Marvell to Build New AI Chips for Inference
Google's building new inference chips with Marvell. Here's why everyone else should care.
Why it matters
Google is moving beyond reliance on its own TPU line by partnering with Marvell to develop specialized inference chips—a signal that the inference bottleneck is becoming a competitive battleground as AI agents scale into production. This mirrors Nvidia's $20B Groq acquisition and suggests chip customization, not GPU monoliths, will define AI infrastructure economics.
The key facts
6 to knowGoogle in talks with Marvell Technology for two new AI chips
One chip is a memory processing unit to pair with TPU
Second chip is new TPU optimized specifically for inference
Focus on inference efficiency for commercial AI products and autonomous agents
Nvidia released LPU (language processing unit) at GTC March 2026, built on $20B Groq technology license
Inference demand surge driving chip specialization across industry
Go to the source
The Informationtheinformation.com
Publisher excerpt: Google is in talks with Marvell Technology to develop two new chips aimed at running AI models more efficiently, according to two people with direct knowledge of the discussions. One is a memory processing unit designed to work alongside Google’s tensor processing unit. The other is a new TPU built…