1. Overview
On September 3, 2026, MIPS — a subsidiary of GlobalFoundries — announced three workload-native development platforms designed to bring physical AI capabilities to embedded devices. The three platforms — Acies, Actus, and Aegis — are each optimized for distinct physical AI workloads: AI inference at the edge, real-time control, and safety-critical systems respectively.
The announcement was made in Shanghai, China, and marks a significant strategic shift for MIPS from a pure IP-licensing model to a platform provider targeting the rapidly growing physical AI market. The platforms leverage the open-standard RISC-V architecture, GlobalFoundries' semiconductor manufacturing capabilities, and a software-first design methodology.
Key takeaway: MIPS is positioning itself not as just another CPU IP vendor, but as a workload platform provider — combining processor IP, workload analysis, and deployment-ready solutions for physical AI applications. The core metric is tokens-per-watt: how much AI computational output a system can generate for each watt of power consumed.
2. What Is Physical AI?
Physical AI refers to AI technology that senses and interprets the physical world and takes direct actions in it. Unlike cloud-based generative AI, physical AI operates at the edge — in industrial machines, transportation platforms, and embedded systems.
Applications include:
- Robotic arms — real-time perception and motion control
- Autonomous vehicles — sensor fusion, decision-making, fail-safe operation
- Autonomous factory machinery — predictive maintenance, adaptive control
- Smart appliances — context-aware operation, energy optimization
The key challenge for physical AI is that traditional general-purpose computing architectures cannot efficiently match the diverse workload characteristics of physical-world applications. Each scenario has different requirements for latency, throughput, safety certification, and power consumption.
3. Three Workload-Native Platforms
3.1 MIPS Acies — Inference-Native Platform
Acies is a developer platform for modern edge AI inference workloads. It features:
- An open, standards-based Neural Processing Unit (NPU)
- An open software stack for model deployment and runtime
- An easy-to-adopt form factor for rapid prototyping
Acies targets edge computing scenarios where data is processed on industrial, commercial, and on-site devices rather than being sent to cloud servers. This reduces latency, preserves privacy, and eliminates network dependency for critical inference tasks.
3.2 MIPS Actus — Event-Native Platform
Actus is a real-time developer platform for embedded systems that must respond quickly to external signals and changes. Key applications include:
- Motion control — precise servo and stepper motor coordination
- Sensor fusion — combining signals from multiple sensors for unified perception
- Predictive maintenance — detecting potential equipment failures before they occur
Actus is designed for deterministic, low-latency response to physical events — a critical requirement for industrial automation and process control.
3.3 MIPS Aegis — Safety-Native Platform
Aegis is a development platform for mission-critical physical AI systems where failure-free operation is essential. It targets:
- Autonomous vehicles — where system failure can cause loss of life
- Medical devices — requiring regulatory compliance and ASIL-D level safety
- Aerospace systems — demanding highest reliability standards
Aegis supports low latency and high levels of stability and reliability, with built-in mechanisms for fault detection, isolation, and recovery. It addresses the unique challenges of safety-critical AI deployment where the system must maintain safe operation even in the presence of hardware or software faults.
4. Software-First Design Philosophy
MIPS's approach is fundamentally software-first: hardware architecture is designed and optimized around the software and AI algorithms that will run on it, rather than the traditional model where software is adapted to fit pre-built hardware.
This methodology involves:
- Workload analysis — detailed profiling of target AI workloads to identify performance bottlenecks and optimization opportunities
- Platform modeling — creating accurate models that predict how workloads will perform on candidate architectures before silicon is built
- Iterative optimization — refining both hardware and software in tight feedback loops
Sameer Wasson, MIPS CEO: “As AI expands into machines, vehicles and other physical systems, compute platforms must evolve to meet the characteristics of each workload. MIPS provides open, software-first platforms that allow customers to develop and optimize physical AI on a common foundation, with energy efficiency and tokens-per-watt as the cornerstone of our technology.”
5. Industry Ecosystem Support
The announcement drew support from major semiconductor industry partners:
- AMD — Manuel Uhm, Director of Silicon Marketing for AMD Embedded, stated: “RISC-V's continued growth momentum remains driven by strong industry collaboration and an expanding software ecosystem. We welcome efforts to broaden developer access and encourage the use of open-standard designs.”
- MediaTek — Co-COO and CFO David Ku stated: “RISC-V's continued growth depends on close collaboration between processor technology, software, and the broader developer ecosystem. MediaTek welcomes MIPS' commitment to open standards and their contribution of mature processor and software expertise.”
MIPS is a subsidiary of GlobalFoundries (GF), one of the world's leading semiconductor foundries. Using the open-standard RISC-V architecture, MIPS offers a portfolio integrating AI and processor intellectual property, workload-optimized analysis, and solutions for physical AI platforms. GF's manufacturing capabilities provide a path from design to production silicon.
6. Platform Comparison
| Platform | Focus | Workload Type | Key Applications | Optimization Goal |
|---|---|---|---|---|
| Acies | AI Inference | Edge compute | Industrial vision, smart cameras, edge gateways | Tokens-per-watt throughput |
| Actus | Real-time Control | Event-driven | Motion control, sensor fusion, predictive maintenance | Deterministic low latency |
| Aegis | Safety-Critical | Mission-critical | Autonomous driving, medical, aerospace | ASIL-D reliability + fault tolerance |
7. Implications for RISC-V
The MIPS platform launch signals several important trends for the RISC-V ecosystem:
7.1 From IP to Platforms
The industry is moving beyond selling CPU IP cores toward offering complete workload-native platforms — combining IP, software stacks, development tools, and reference designs. This reduces the integration burden for customers and accelerates time-to-market.
7.2 Physical AI as a RISC-V Stronghold
Physical AI — with its diverse workloads, strict power constraints, and need for customization — is a natural fit for RISC-V's modular, extensible architecture. MIPS's focus on tokens-per-watt efficiency aligns with RISC-V's advantage in domain-specific customization.
7.3 Software-First as the New Paradigm
The software-first approach represents a departure from traditional hardware-led design. By profiling workloads first and designing hardware to match, MIPS can achieve better efficiency than general-purpose architectures for target applications.
7.4 Open Standards Ecosystem Growth
With AMD and MediaTek publicly endorsing the open-standard RISC-V approach, the industry consensus around RISC-V for embedded and physical AI continues to strengthen. MIPS's Premier Membership in RISC-V International and the appointment of CTO Yankin Tanurhan as Board Vice Chairman further solidify this commitment.