Make the CPU great again! RISC-V chip IP firm SiFive’s fundraising is red-hot—NVIDIA and Gavin Baker participated in the investment.

RISC-V chip IP company SiFive today announced it has completed a $400 million Series G funding round. The most notable name on the investor list is NVIDIA. NVIDIA’s participation not only boosted SiFive’s valuation to $3.65 billion, but also sent a signal flare across the semiconductor industry: the open instruction set architecture has officially entered the mainstream competition track in data centers.

So-called agentic AI refers to AI systems that can make multi-step decisions and perform task planning. As AI evolves from single-model inference to multi-agent collaboration, the system-level scheduling and control requirements rise significantly, bringing renewed attention to the CPU’s role in the data center.

SiFive said that in these workloads, the CPU coordinates the operation between different models and accelerators, handling complex logic and workflow control—tasks that GPUs are not particularly good at—making the CPU’s performance and power consumption key factors limiting overall computing capability.

The existing x86 server CPU architecture was designed before these needs emerged. It has high power consumption and limited room for customization. Meanwhile, RISC-V is an open architecture, allowing cloud providers to deeply tailor it to their own workloads, extracting higher performance at the same power budget. Against this backdrop, the RISC-V architecture is seen as an important option for replacing traditional CPU designs. And SiFive has an important position within the RISC-V architecture camp.

SiFive G round funding overview

Funding amount: $400 million (Series G, oversubscribed)

Post-funding valuation: $3.65 billion

Lead investors: Atreides Management (Gavin Baker)

Co-investors: NVIDIA, Apollo, Point72 Turion, T. Rowe Price

Use of funds: Development of high-performance RISC-V CPU IP, software ecosystem, expansion of global engineering team

Why the CPU suddenly matters

Over the past few years, data center investment focus has been almost monopolized by GPUs and AI accelerators, leaving CPUs as the supporting actor. But as AI applications evolve from simple model inference into “Agentic AI” that can autonomously plan, call tools, and coordinate multi-step tasks, the CPU’s workload is quietly expanding. It needs to handle control-plane work in agent workflows such as scheduling, memory management, and coordinating API calls—tasks that GPU architectures are not suited for.

SiFive’s argument is this: the current x86 server CPU architecture was designed before these needs emerged, with high power consumption and limited room for customization. As an open architecture, RISC-V enables cloud providers to deeply tailor it to their workloads, delivering higher performance at the same power budget. In the industry, some ultra-large-scale and chip design cases have already adopted RISC-V as the core for specific control or management tasks.

The subtext of NVIDIA’s investment

The most worth closely reading detail in this round is the integration plan between SiFive and NVIDIA NVLink Fusion. NVLink is NVIDIA’s high-speed interconnect technology for GPU clusters; in the past, it was mainly paired with NVIDIA’s own Grace CPU (based on the Arm architecture).

If a RISC-V CPU core can connect to NVLink Fusion, it means that ultra-large-scale customers in the future may have the opportunity to build their own Host CPUs using SiFive IP, paired with NVIDIA GPUs—while escaping reliance on Intel or Arm authorization without sacrificing interconnect bandwidth.

For NVIDIA, investing in SiFive is also a step in defense. If the RISC-V ecosystem grows strong, securing a foothold early can ensure its GPUs can integrate seamlessly across various CPU environments, avoiding displacement caused by ecosystem lock-in to specific CPU vendors.

SiFive CEO Patrick Little said, “Ultra-large-scale customers have clearly stated that it is time to accelerate the adoption of data center open-standard alternatives. They have consistently requested CPU solution options in customizable IP forms so they can substantially enhance the differentiated advantage of their data center computing solutions. RISC-V is the only architecture that can truly meet these needs. As the industry accelerates toward intelligent AI, SiFive is building significant momentum in the data center domain.”

Challenges remain: software ecosystem is the real moat

However, between technical feasibility and market success is still a deep trench called “software ecosystem.” x86 and Arm have accumulated decades in the data center—device drivers, compiler optimizations, operating system support, and enterprise software certifications—which is a distance that cannot be closed in a single round of funding. Even though SiFive has claimed completion of CUDA, RedHat, and Ubuntu porting, there is still a considerable gap between porting completion and stable, reliable deployment in production environments.

SiFive’s current IP has been adopted in more than 500 designs, with total shipments exceeding 10 billion cores, but these are mainly concentrated in embedded, edge computing, and consumer electronics—data center-scale deployments are still in the early stage. The view of industry analyst Dan Newman may be more in line with reality: “CPUs suddenly became exciting, especially for data center applications. SiFive saw this trend early, and as the industry evolves, it is positioned favorably.”

This article makes the CPU great again! RISC-V chip IP company SiFive’s fundraising is extremely hot; NVIDIA and Gavin Baker joined as investors. First appeared on Chain News ABMedia.