AMD Ventures points its AI investing past the data center

AMD Ventures is putting more attention on physical AI, the term being used for AI systems embedded in robotics, industrial automation and other machines outside cloud data centers. Sagi Paz, head of AMD Ventures at Advanced Micro Devices Inc., said the corporate venture arm can invest from $1 million to tens of millions of dollars, but did not disclose a new fund, specific deal or valuation target.

The comments, made during SiliconANGLE Media’s theCUBE coverage of AMD Advancing AI 2026, show how AMD is thinking about AI as more than a data center accelerator market. Paz described AMD’s AI infrastructure strategy as a mix of internal development, acquisitions and startup investing, with the venture group focused on companies where AMD believes it can contribute more than capital.

That is a familiar corporate venture filter, but the category call is the useful part. AMD has been competing for position in AI infrastructure, and Paz said the next opening may come as AI workloads move into physical systems such as industrial robots and embedded devices. He said no company is yet deploying physical AI at scale, while predicting that the category will eventually have a breakout moment comparable to ChatGPT’s effect on generative AI adoption.

How AMD Ventures says it invests

Paz said AMD Ventures is not tied to a single stage. The size of an investment depends on how closely AMD expects to work with the startup and how much capital the company is raising. The lower bound he gave was $1 million, with larger checks reaching into the tens of millions.

He also said the group takes observer seats rather than board seats in about half of its portfolio. Paz framed that as a way to keep regular contact with founders while avoiding fiduciary conflicts that can come with formal board roles. He said each investment still has to qualify on conventional venture criteria: the team, product differentiation and market size need to stand on their own.

AMD was not presented as investing only where it can sell chips. Paz’s argument was broader: AI companies need support across multiple layers of the stack, and AMD Ventures looks for startups where AMD has a specific reason to be useful. He did not name current physical AI portfolio companies in the discussion.

The physical AI bet

Physical AI remains an imprecise label, and Paz’s comments did not include revenue forecasts, customer deployments or technical benchmarks. The use cases named were robotics, industrial automation and embedded systems operating in real-world environments. That makes the category different from the current AI infrastructure cycle, where much of the capital spending has centered on cloud data centers and GPU capacity.

The investment signal is still notable. Corporate venture arms often reveal where a platform company sees demand forming before product lines or acquisitions make the strategy obvious. For AMD, physical AI would extend the AI discussion from training and inference infrastructure into systems that require compute closer to sensors, machines and industrial workflows.

TheCUBE disclosed that it was a paid media partner for AMD Advancing AI 2026 and said event sponsors did not have editorial control over its coverage.

This story draws on original reporting from SiliconANGLE.