European semiconductor company SiPearl is reviving CPU development initiatives for HPC and AI, led by former STMicroelectronics engineer Philippe Notton.
In 2015, while an engineering group leader at STMicroelectronics, Philippe Notton grew enthusiastic about reviving European efforts to develop homegrown CPUs. In 2017 he joined Atos and led efforts to found the European Processor Initiative (EPI), then in 2019 spun off new company SiPearl with EU support and an ARM-based strategy. Europe's first exascale supercomputer, JUPITER at the German national HPC center in Juelich, will feature SiPearl's Rhea1 CPUs alongside NVIDIA and AMD accelerators, with general availability of Rhea1 scheduled for the end of 2026. Europe's second exascale system, Alice Recoque, will be hosted by CEA/DAM south of Paris and will include CPUs based on the company's Rhea2 ARM architecture.
**On SiPearl's founding milestone:**
The vision that led to the creation of the European Processor Initiative consortium became a reality in May when SiPearl received the first Rhea1 units in its laboratories. The company has built the most complex server CPU ever designed in Europe, competing directly with current market leaders—all of whom are non-European.
**On long-term strategy:**
SiPearl's strategy is to be a leading player wherever there is a need for sovereign advanced computing—whether in HPC, AI, defense, or data centers—by offering a high-performance, energy-efficient alternative to market leaders. Sovereignty is important, but the company's ambition is ultimately to compete on merit in terms of computing power and energy efficiency. SiPearl is seeing significant interest from outside Europe in countries seeking alternatives to non-European CPUs.
**On complete hardware solutions:**
SiPearl is working toward a complete hardware solution combining its CPUs with Axelera's AI processors and leveraging Semidynamics' RISC-V-based ASICs for AI inference. The company participates in the AETHER Consortium alongside 2CRSi, a French specialist in high-performance servers, alongside European real estate, construction, services, energy providers and infrastructure operators, to demonstrate that Europe possesses all the expertise required to build, power, operate and scale AI infrastructure. In July, the AETHER consortium announced its application for the European Commission's forthcoming AI Gigafactory call for proposals.
**On the CPU resurgence in AI:**
The rapid rise of agentic AI, combined with rising GPU prices, has reinforced the role of central processing units. The CPU-to-GPU ratio in AI servers is now approaching 1:1, a trend that started just months ago. After several years of stagnation, the CPU market is experiencing impressive growth, with longer delivery times and rising prices for major players. This long-term market trend favors SiPearl's positioning.
**On other computing markets:**
CPUs remain the base engines for HPC, cloud, and enterprise computing. They run workloads directly and orchestrate accelerators and quantum computers. Regarding emerging technologies, co-packaged optics are maturing quickly and should be in the market by 2030; SiPearl intends to be a fast follower rather than a market leader in this space. Optical compute replacing CPUs would require engineering breakthroughs not envisioned in the short to midterm. However, matrix multiplication in the optical domain, which could accelerate inference, is expected to be production-capable within five years. SiPearl's go-to-market strategy is to partner with accelerators and provide the orchestration layer for companies doing optical inference acceleration.
**On the Athena1 defense processor:**
Athena1 is a customized version of Rhea1 with four high-bandwidth memory stacks removed, as these are not required for defense and aerospace applications and help reduce cost and size. Discussions are currently underway with major players in the European defense sector.
**On ARM versus RISC-V:**
As a server CPU company, SiPearl operates in a market where the primary instruction set architecture choices are x86 (Intel, AMD) or ARM (SiPearl, NVIDIA, Fujitsu, Ampere, and hyperscalers). The choice of ISA is governed by software ecosystem requirements; many software elements, both open source and proprietary, must be ported to an ISA for it to be viable as a server CPU. RISC-V is a popular, relatively new ISA gaining market share in IoT and embedded CPU markets, with many AI accelerator companies building on RISC-V. SiPearl interfaces with partner companies building RISC-V accelerators. While most industry experts agree RISC-V will eventually become viable for server CPUs, the timeframe remains debated. SiPearl maintains a technology watch on RISC-V solutions but considers mainstream adoption of RISC-V server CPUs unlikely before 2030.