Want to connect with CMP (Circuits Multi-Projets)?
Join organizations building the agentic web. Get introductions, share updates, and shape the future of .agent.
Is this your company?
Claim this profile to update your info, add products, and connect with the community.
CMP is relevant to the AI agent ecosystem because it provides the prototyping pathway for the specialized hardware required to run agents locally and efficiently. While cloud-based agents dominate current headlines, the long-term deployment of agents in physical environments—such as robotics, wearables, and industrial IoT—requires custom ASICs and NPUs (Neural Processing Units). CMP enables startups and researchers to design and manufacture these custom chips without the multi-million dollar overhead of dedicated foundry runs.
By acting as an aggregator for Multi-Project Wafer runs, CMP supports the diversification of the hardware layer. This is critical for moving beyond general-purpose GPUs toward domain-specific architectures that can process agentic workflows with minimal latency and power consumption.
The physical substrate of the artificial intelligence boom is often discussed in terms of massive GPU clusters and the electricity required to cool them. However, a significant portion of the ecosystem focuses on the opposite end of the spectrum: efficient hardware designed to run models at the edge. This is the domain of CMP (Circuits Multi-Projets), a service organization headquartered in Grenoble, France, that is a vital intermediary in the semiconductor supply chain. Since its inception, CMP has focused on lowering the barrier to entry for integrated circuit (IC) and micro-electro-mechanical systems (MEMS) production.
At the heart of the CMP model is the Multi-Project Wafer (MPW) service. In standard semiconductor manufacturing, the cost of creating a set of masks for a specific design is immense, often reaching millions of dollars for advanced process nodes. This cost effectively bars universities, research laboratories, and small startups from experimenting with physical silicon. CMP solves this by aggregating multiple different designs from various customers onto a single mask set and wafer. By sharing the fixed costs of manufacturing across dozens of projects, CMP allows a single researcher or a small engineering team to tape out a design for a fraction of the standard cost.
The organization maintains relationships with the world’s leading foundries, including STMicroelectronics, TSMC, and GlobalFoundries. These partnerships enable CMP to offer a wide range of technology nodes, from legacy processes used in power electronics to the advanced nodes required for high-performance computing. For the AI agent ecosystem, this access is increasingly important. As the industry moves away from a one-size-fits-all approach to compute, there is a growing demand for custom ASICs (Application-Specific Integrated Circuits) that are optimized for specific inference tasks. These chips are the foundation for the next generation of autonomous devices and edge agents that cannot rely on a constant cloud connection.
Located in Grenoble—a city often referred to as the silicon valley of Europe due to the presence of CEA-Leti and major semiconductor manufacturers—CMP occupies a unique strategic position. It is not a commercial foundry itself, but rather a service that facilitates the transfer of intellectual property from a Verilog file to a physical chip. This role is particularly relevant as the 'long tail' of the AI hardware market begins to grow. While large-scale model training is dominated by a few players, the deployment of specialized agents in robotics, medical devices, and automotive systems requires a diversity of hardware that can only be developed through iterative prototyping.
In an era where hardware sovereignty and supply chain resilience are central to the tech narrative, services like CMP provide a necessary on-ramp for innovation. They ensure that the design of the physical machine remains as accessible as the software that runs upon it. For those building the future of agentic AI, understanding the path from algorithm to silicon is essential, and CMP remains one of the few organizations making that path navigable for those without a billion-dollar R&D budget.
CMP (Circuits Multi-Projets) is hiring
You've explored CMP (Circuits Multi-Projets).
Join organizations building the agentic web.