Google is building a four-partner chip supply chain to challenge Nvidia in AI inference

image via TNW | The heart of techThe Next Web

April 20, 2026, 8:27 PM

• •Google is assembling a diversified custom chip supply chain with four design partners.
• •Ironwood, Google's seventh-generation TPU, is now generally available for Google Cloud customers.
• •Broadcom is handling high-performance chip variants, including the next-generation TPU v8 training chip.
• •MediaTek is designing the cost-optimized inference variant of the TPU v8.
• •Marvell is in talks to develop a memory processing unit and a new inference-focused TPU.
• •Intel is supplying Xeon processors and custom infrastructure processing units for Google's AI data center.
• •TSMC fabricates all of Google's custom silicon.
• •Google's total expected TPU shipments are projected at 4.3 million units in 2026, scaling to more than 35 million by 2028.

Google is building a diversified custom chip supply chain with multiple partners, including Broadcom, MediaTek, Marvell, and Intel. This strategy aims to challenge Nvidia's dominance in AI inference by focusing on cost-effective, purpose-built silicon. The roadmap includes the current Ironwood TPUs and the upcoming TPU v8 chips, targeting TSMC 2nm in late 2027. This multi-partner approach provides optionality and aims to reduce reliance on any single supplier, potentially creating a cost advantage in the long run.

Read original article