The first wave of the artificial intelligence infrastructure buildout was a race for capacity: AI model training fueled the earliest growth stages, supported by more powerful xPUs including GPUs and AI ASICs that enabled the rapid expansion of hyperscale data centers.
That initial investment cycle transformed AI into a global computing platform. Today's AI infrastructure has entered a new phase where workloads shift from training models on extremely large datasets to AI inference, which emphasizes continuous, real-time application processing. As model inference becomes an increasingly dominant workload, infrastructure requirements are evolving beyond raw compute capacity.
The emergence of agentic AI and physical AI, which extends autonomous systems capable of reasoning, planning, and executing complex tasks into the real world, is spawning new business use cases across multiple markets and applications. In automotive, for example, AI is enabling software-defined vehicles, while AI-enhanced humanoid robotic deployments could approach one billion by the middle of the century.
As a result, xPUs are expected to share the stage with a host of enabling technologies that will define the future of data center design and performance. In this environment, CPUs and xPUs, but also power, memory interfaces, and control systems, will drive AI market expansion and create a broader and more durable infrastructure opportunity than many had initially anticipated. Amplified by inference workloads, AI's burgeoning power demands are triggering a fundamental redesign of infrastructure architectures. Renesas is uniquely positioned to capitalize on this opportunity. As I shared in an AI infrastructure and compute update at our recent Renesas Capital Market Day investor event, this is due to our extensive portfolio of digital power, power semiconductor, memory interface, control, and analog solutions.