Google is cooking up new AI chips with Marvell

New chips, same AI arms race

Google apparently isn’t done tinkering with its AI stack. The company is in talks with Marvell Technology to develop two new chips designed to run AI models more efficiently, according to The Information.

For Alphabet, this is classic Big Tech behavior: if the hardware bottleneck is slowing down the software party, just build better hardware. That matters because AI inference — the part where models actually answer your questions, generate images, or summarize your inbox — can get expensive fast.

Why investors should care

If Google leans more on custom silicon, it could:

• lower the cost of running Gemini and other AI products
• improve performance for search, cloud, and consumer AI features
• keep Google less dependent on outside chip suppliers

For Marvell, a deal like this would be a nice proof point that its custom chip business has real legs. These designs aren’t one-line item headlines; they can turn into sticky, high-value relationships if they make it past the “talks” phase.

The big picture

This is still early-stage, so nobody’s popping champagne yet. But in AI, the companies with the better chips often get the better economics — and that’s the kind of edge Wall Street likes to reward.

Big picture: if Google wants AI to feel faster and cheaper, it may need to keep building its own picks and shovels instead of renting them.