Google has finalized a landmark $12.2 billion agreement with Marvell Technology to co-develop and procure custom AI accelerators, granting Marvell a substantial equity stake option.

Following initial reports in April hinting at a potential expansion, Alphabet’s Google is cementing its next semiconductor shift by diversifying beyond Broadcom for custom chips. The US$12.2bn agreement with Marvell Technology grants the search giant an option to purchase a major stake in the chipmaker while securing specialised hardware to power its expanding Tensor Processing Unit (TPU) ecosystem. This move underscores Google’s strategic pivot toward custom silicon as demand accelerates among hyperscalers seeking alternatives to general-purpose graphics processing units (GPUs).

Companies are increasingly turning to specialised hardware to lower costs and manage complex workloads, as custom silicon delivers targeted performance for operational demands—particularly for AI inference, which involves running trained AI models. Google’s expanded partnership with Marvell directly targets this specialised hardware market. The collaboration will focus on developing custom semiconductor products integrated into Google’s TPU ecosystem, including AI inference accelerators, storage controllers, network interface controllers, memory interface controllers, and near-memory compute solutions.

According to a regulatory filing, Marvell has issued a warrant allowing Google to purchase up to 58.97 million shares at US$206.58 apiece. If fully exercised, the transaction would be valued at US$12.18bn, making Google the fifth-largest investor in Marvell. However, Google is not committing the full amount upfront. Instead, 1.3 million shares will vest in equal quarterly instalments during the first year following the warrant’s execution. The remaining shares will vest based on discretionary purchases from Marvell’s third quarter of 2027 through the end of 2033, with one tranche vesting for every US$500m in revenue generated from the co-developed custom products. Analysts estimate that if Google meets its revenue targets, the partnership could generate approximately US$120bn in revenue for Marvell through fiscal 2033.

The arrangement reflects a broader industry trend of technology companies taking financial stakes in critical hardware suppliers to secure long-term capacity. Last year, AMD agreed to supply OpenAI with AI chips projected to generate tens of billions of dollars in annual revenue, alongside an option to acquire up to a 10 per cent stake in the business. In March, NVIDIA invested US$2bn in Marvell to advance silicon photonics technology aimed at accelerating data flows within data centres. This investment followed Marvell’s acquisition of Celestial AI to develop photonics infrastructure that interconnects chip clusters powering large-scale systems such as Google’s Gemini, Anthropic’s Claude Code, and OpenAI’s ChatGPT. Additionally, NVIDIA has agreed to provide a financing backstop of up to US$105bn for a data centre project leased by OpenAI in Ohio. These interconnected arrangements highlight how deeply aligned financial and technological partnerships have become across the AI infrastructure sector.