This article first appeared on GuruFocus.
Alphabet (GOOGL, Financials) is looking to deepen its control over AI infrastructure, holding talks with Marvell Technology to develop a new set of chips designed to run AI models more efficiently.
The discussions center around two pieces of hardware. One is a memory-focused chip that would work alongside Google’s existing tensor processing units. The other is a new version of its TPU, built specifically for handling AI workloads once models are already trained.
That second part is where things are heading. Running AI models, not just training them, is quickly becoming the bigger challenge. It is also where costs can add up fast, which is why companies are starting to rethink chip design.
For Google, this is not just about performance. It is also about control. The company has been pushing its in-house chips as an alternative to Nvidia’s GPUs, especially inside its cloud business.
Nothing is finalized yet, and the report has not been independently confirmed. But the direction is clear. Big tech companies are trying to rely less on outside suppliers and build more of their own AI stack.
If the plans move forward, the next step would be finalizing the chip design before moving into testing and production, potentially as soon as next year.
