A major shift is underway in how artificial intelligence gets deployed: instead of sending data to distant data centers, companies are now building AI chips that live inside machines, cars, and factory equipment. MIPS, a processor design company, just unveiled three new platforms specifically engineered for this "Physical AI" movement, marking a turning point in where and how AI actually gets used in the real world.

What Is Physical AI, and Why Does It Matter?

Physical AI refers to artificial intelligence embedded directly into machines, vehicles, and industrial systems rather than running on cloud servers. This approach solves a real problem: sending data to the cloud and waiting for a response can be too slow for applications where split-second decisions matter. Think of a self-driving car that needs to react to a pedestrian, or a factory robot that must respond instantly to equipment failure. These systems can't afford the latency of cloud communication.

The three platforms MIPS launched on September 2, 2026, each target different types of physical AI workloads. They represent a fundamental rethinking of how compute resources should be designed, moving away from one-size-fits-all processors toward specialized chips built for specific tasks.

How Do These New Platforms Work Together?

• Acies Platform: Designed specifically for AI inference at the edge, meaning it runs trained AI models on local devices. It features an open Network Processing Unit and software stack, prioritizing efficiency metrics like tokens-per-watt, a measure of how much AI computation you get per unit of energy consumed.
• Actus Platform: Built for real-time, event-driven embedded systems such as motion control and predictive maintenance. Unlike general-purpose processors, Actus responds instantly to external triggers, critical for applications where delays could have serious consequences.
• Aegis Platform: Focused on mission-critical physical AI systems that demand unwavering reliability and low-latency processing. This platform prioritizes agentic processing, meaning the AI can take autonomous actions without waiting for human input.

All three platforms are built on RISC-V, an open-source processor architecture that allows developers to customize and extend the design for their specific needs.

Why Energy Efficiency Matters More Than Raw Speed

Traditional AI benchmarks focus on how fast a processor can compute. But for devices running on batteries or powered by solar panels in remote locations, speed is less important than efficiency. MIPS is emphasizing tokens-per-watt as the key performance metric, which measures how much useful AI work you can accomplish per unit of energy. This shift reflects a practical reality: embedded AI systems live in a world of power constraints that data center chips never face.

"As intelligence moves into machines, vehicles, and other physical systems, compute platforms need to adapt to the workload," said Sameer Wasson, CEO of MIPS, emphasizing that efficiency and tokens-per-watt are the foundation of the company's technology.

Sameer Wasson, CEO of MIPS

This philosophy extends beyond just the hardware. MIPS is taking a software-first approach, using open standards to guide developers and platform designers. The goal is to make it easier for engineers to build custom AI solutions without reinventing the wheel each time.

Which Industries Will Benefit First?

The immediate applications for these platforms span several sectors where real-time, local AI processing offers clear advantages. Automotive systems need instant decision-making for autonomous driving features. Industrial automation relies on predictive maintenance that can catch equipment failures before they happen. Edge computing applications, from smart city infrastructure to remote monitoring systems, all benefit from processing data locally rather than sending it to the cloud.

The emphasis on workload optimization signals that the era of generic processors is fading. Instead, the future looks like specialized chips designed for specific problems, with software ecosystems built around them. This approach mirrors how the AI industry has evolved more broadly, moving from general-purpose models toward fine-tuned systems optimized for particular tasks.

Why Industry Collaboration Matters for Open Standards

The success of these platforms depends heavily on industry adoption and software ecosystem development. RISC-V, the underlying architecture, only becomes valuable if developers build tools, libraries, and applications around it. Companies like AMD and MediaTek are publicly supporting this effort, recognizing that open standards create larger markets than proprietary approaches.

"The continued growth of RISC-V depends on close collaboration across processor technology, software, and the broader development ecosystem," said David Ku, Co-COO and CFO of MediaTek.

David Ku, Co-COO and CFO of MediaTek

This collaborative approach contrasts with the traditional semiconductor industry, where companies guard their designs closely. By embracing open standards, MIPS and its partners are betting that a rising tide of innovation will lift all boats, creating more opportunities for everyone than a closed, proprietary approach would allow.

The launch of these three platforms represents more than just new products; it signals a fundamental shift in how the AI industry thinks about deployment. As AI becomes embedded in everyday machines and systems, the compute platforms powering that intelligence need to evolve. MIPS is positioning itself at the center of that evolution, offering specialized tools for a world where intelligence is no longer confined to data centers but distributed across the physical systems we interact with every day.

" }