SiFive To Integrate NVIDIA NVLink Fusion For Next-Gen AI Data Center Chips

In a move that could have major implications for future AI datacenters, SiFive, a leader in RISC-V processor IP, has announced a strategic partnership with NVIDIA to integrate NVIDIA’s NVLink Fusion interconnect technology into its forthcoming silicon platforms. This collaboration marks a significant shift for both companies and the broader AI ecosystem, which brings together SiFive’s customizable, power-efficient RISC-V processor IP with NVIDIA’s industry-leading NVLink scale-up interconnect and OCP MGX rack-scale architecture. NVLink Fusion, which was made available for third-party licensing in Q2 2025, enables high-bandwidth, low-latency, scale-up capabilities at the rack scale, with seamless integration of NVIDIA AI accelerators and the broader NVLink Fusion ecosystem. SiFive now joins other leaders in the space that have also adopted NVLink Fusion, but the partnership also paves the way for the RISC-V open source compute architecture as a whole.

SiFive Joins The NVIDIA NVLink Fusion Ecosystem

SiFive’s current business model centers on delivering highly customizable, scalable, and energy-efficient processor IP that can be tailored to the unique needs of its customers, whether it be AI, machine learning, and high-performance computing workloads or low-power IoT applications. SiFive is a clear leader in the RISC-V space and the company’s commitment to open-source software, which includes collaborations with Red Hat and Canonical, has accelerated the maturation of the RISC-V software stack, making it a powerful, viable alternative to x86 and Arm architectures for demanding AI and HPC applications. Foreshadowing today’s release, NVIDIA announce back in July that it was porting CUDA to RISC-V as well.

“With SiFive, we’re bringing our coherent, high-bandwidth NVLink interconnect to the RISC-V ecosystem,” said Jensen Huang, founder and CEO of NVIDIA. “This provides the flexibility to combine customizable RISC-V CPUs with NVIDIA accelerators to build scalable, power efficient and specialized AI infrastructure – and move from innovation to deployment faster than ever.”

NVIDIA’s NVLink technology is the backbone of high-performance NVIDIA-powered AI and HPC systems. It enables ultra-fast, low-latency communication between NVIDIA’s GPUs, CPUs, networking, and other technologies. With the introduction of NVLink Fusion, NVIDIA opened this once-proprietary interconnect up to select third-party partners, allowing custom CPUs and accelerators to be more seamlessly integrated into NVIDIA’s rack-scale architecture. The platform is built on the Open Compute Project (OCP) MGX rack architecture, ensuring compatibility and rapid deployment for hyperscalers and cloud providers.

Integrating NVLink Fusion With SiFive’s RISC-V IP

The integration of NVLink Fusion into SiFive’s RISC-V platforms could be a major disruptor moving forward. For SiFive, this means future RISC-V-based processor platforms can participate as first-class citizens in NVIDIA’s rack-scale technologies. The NVLink interface will allow future silicon based on SiFive’s customizable IP to communicate directly with NVIDIA’s (and partner) chips, unlocking access to the full CUDA-X software stack, including libraries like NCCL for collective communication and deep learning frameworks such as PyTorch and TensorFlow. This kind of integration will enable seamless orchestration of large-scale AI workloads, with processors based on SiFive’s technology managing application logic and system control while NVIDIA GPUs handle AI acceleration.

MORE FOR YOU

One of the key considerations of this partnership is customization, and co-design and co-optimization of future platforms. The modularity of RISC-V and SiFive’s design philosophy allows for tailored implementations optimized for power, area, and performance. By combining SiFive’s technologies with NVLink Fusion, SiFive customers can build highly efficient, scalable systems that address the unique demands of AI datacenters, while targeting specific power and performance levels.

Transforming Next-Generation AI Datacenters

SiFive’s RISC-V processor IP is renowned for its power efficiency, a critical concern as hyperscale datacenters deal with soaring energy demand and cost.

Custom silicon with NVLink Fusion, enabled by SiFive’s flexible IP portfolio, will allow SiFive’s customers to optimize for specific workloads, whether it’s LLM inference, recommendation systems, or real-time analytics. This bespoke approach stands in stark contrast to using off-the-shelf designs, offering a path to differentiated, workload and power optimized, high-margin AI services.

While SiFive and NVIDIA have not specified an exact timeline for the integration of NVLink Fusion into SiFive’s platforms, it will most likely be a couple of years before actual silicon is deployed. The nature of SiFive’s business model requires customers to begin working on processor designs many months or years in advance of commercialization. However, as the AI hardware market continues to shift toward bespoke, workload-optimized silicon, SiFive’s ability to deliver customizable, power-efficient RISC-V processor IP—now with access to NVIDIA’s interconnect and software stack—positions it well moving forward.