Google and Marvell Technology Negotiate Development of Custom Artificial Intelligence Processors

2026-04-19 semicon

Eindhoven, Sunday 19 April 2026
Google is negotiating with Marvell Technology to design custom artificial intelligence processors, a technology that could soon power highly classified military infrastructure at the Pentagon.

Accelerating the Custom Silicon Race

Alphabet’s Google is currently engaged in advanced discussions with Marvell Technology to co-develop two new custom chips designed to execute artificial intelligence (AI) models with greater efficiency [2]. According to reports circulating in mid-April 2026, the collaboration focuses on a new tensor processing unit (TPU) built specifically for AI, alongside a complementary memory processing unit [2]. The companies aim to finalise the design of this memory unit as early as next year, 2027, before transitioning to test production [2]. This move is a strategic effort by Google to position its TPUs as a compelling alternative to Nvidia’s widely used graphics processing units (GPUs), as TPU sales have increasingly become a primary growth engine for Google’s cloud revenue [2].

Securing Classified Military Infrastructure

Beyond commercial data centres, these custom TPUs are central to Google’s renewed push into the defence sector. The technology giant is reportedly in talks with the United States Department of Defense to deploy its Gemini AI models within highly classified environments [5]. To overcome the current computing limitations of Google’s Distributed Cloud for classified tasks, the proposed agreement includes the installation of GPU racks and, notably, custom TPUs directly inside secure military facilities [5]. This deployment would enable the Pentagon to process AI workloads at an unprecedented scale [5].

Implications for European Strategic Autonomy

For the broader semiconductor value chain, the accelerating shift by hyperscalers toward custom silicon carries profound implications, particularly for European strategic autonomy. While Alphabet generates 29.6 percent of its net sales from the Europe, Middle East, and Africa (EMEA) region [1], the physical realisation of these custom chips relies inextricably on European manufacturing equipment [GPT]. Advanced lithography systems from Dutch firm ASML and atomic layer deposition technologies from ASM are critical in fabricating the high-performance logic and memory chips required for next-generation TPUs [alert! ‘Specific equipment utilisation by Marvell/Google foundries is proprietary, but ASML/ASM are industry standards for advanced nodes’] [GPT]. As tech giants design more specialised silicon, the demand for cutting-edge European equipment intensifies, reinforcing the continent’s pivotal role in the global supply chain [GPT].

Sources & Ecosystem Partners

Custom silicon Inference processors