Google and Marvell announced a partnership to design two new AI chips, one a memory‑processing unit that will work alongside Google’s existing Tensor Processing Units (TPUs) and another a dedicated inference TPU aimed at accelerating real‑time AI workloads.
The memory‑processing unit is slated for design completion next year, after which it will enter test production. The inference TPU will target inference‑heavy applications, potentially lowering the cost of running trained models for Google Cloud and other services.
Google has earmarked $75 billion for AI infrastructure in 2025 and plans $175‑185 billion for 2026, underscoring its commitment to expanding custom silicon. By partnering with Marvell, Google seeks to reduce reliance on NVIDIA GPUs and broaden its own chip portfolio.
Marvell brings deep expertise in custom ASICs for data‑center workloads, with a broad IP portfolio that includes high‑speed SerDes and advanced packaging. The company’s Q1 FY2026 results showed record revenue driven by AI demand, highlighting its fit as a partner for this initiative.
The new chips could lower inference costs for Google Cloud and other AI services, improving margins and strengthening Google’s competitive edge in the rapidly growing AI chip market, where NVIDIA currently dominates. The partnership also signals a broader trend among hyperscalers to develop custom silicon to secure supply‑chain control and performance advantages.
The collaboration reflects Google’s strategic push to secure a more efficient, cost‑effective AI infrastructure that can better serve enterprise customers seeking scalable, low‑latency AI solutions.
The content on EveryTicker is for informational purposes only and should not be construed as financial or investment advice. We are not financial advisors. Consult with a qualified professional before making any investment decisions. Any actions you take based on information from this site are solely at your own risk.