April 20, 2026 – Google is in advanced talks with Marvell Technology to co-develop two new custom AI chips focused on inference workloads, according to multiple reports. This move further diversifies Google’s AI silicon supply chain beyond Broadcom and signals intensifying competition in the custom AI chip market.

Google in Talks with Marvell for New Custom AI Inference Chips

Alphabet’s Google is negotiating with Marvell Technology to design two specialized AI chips aimed at running AI models more efficiently. The discussions, first reported by The Information, highlight Google’s aggressive push to reduce reliance on any single supplier while optimizing costs and performance for large-scale AI inference.

Deal Details and Chip Plans

According to sources, the partnership would involve:

• A new **memory processing unit** designed to work alongside Google’s existing Tensor Processing Units (TPUs)
• A dedicated **inference-optimized TPU** focused on serving AI models to users (as opposed to training)

Marvell would serve in a design-services role, similar to its work with other hyperscalers and MediaTek’s involvement on Google’s latest Ironwood TPU. No contract has been signed yet, but talks are progressing toward finalizing designs by next year.

Google’s Custom Silicon Strategy

Google has long invested heavily in its own TPUs to power Google Cloud and internal AI workloads. By adding Marvell as another design partner alongside Broadcom, Google is building a more resilient and diversified supply chain. This approach helps control costs, improve performance for specific workloads, and reduce dependency on any single vendor — especially Nvidia, which still dominates the high-end AI GPU market.

Impact on Nvidia and the AI Chip Market

The news sent Marvell shares up nearly 6% in pre-market trading, while Broadcom shares dipped slightly. It underscores the growing trend among hyperscalers (Google, Amazon, Meta, Microsoft) to develop custom AI silicon to challenge Nvidia’s dominance and lower long-term infrastructure costs. The custom ASIC market is projected to grow rapidly in the coming years.

What This Means for the Industry

This potential partnership reflects the accelerating arms race in AI infrastructure. More companies are moving toward custom chips to achieve better efficiency, lower power consumption, and cost advantages in the inference phase — where most real-world AI usage happens. For developers and enterprises, it could eventually lead to more competitive pricing and specialized performance on Google Cloud.

Final Verdict

Google’s talks with Marvell represent another significant step in the hyperscaler shift toward custom AI silicon. While details are still emerging and no final contract has been signed, the move strengthens Google’s position in the AI infrastructure race and adds pressure on traditional GPU leaders like Nvidia.

Shop Mac Mini & AI Development Gear at Gzmato