Chip startup SiFive raised $400 million in a funding round led by Atreides Management on April 9. Nvidia, Apollo Global Management, and Point72 also joined. The round valued SiFive at $3.65 billion and was oversubscribed. Investors offered more capital than the company ultimately accepted. SiFive plans to use the funding to expand its presence in AI data centers, where it sees growing demand for processor architectures that complement the GPU-centred compute stack.
SiFive develops processors based on the RISC-V architecture, an open-source chip instruction set that has gained traction across the semiconductor industry as an alternative to proprietary designs from Arm and Intel. RISC-V carries no licensing fees and lets chipmakers customise designs freely for specific applications. Data center operators find that flexibility attractive for workloads that do not need peak GPU throughput. Networking, inference orchestration, storage management, and edge compute are all strong candidates for RISC-V based silicon.
Why Nvidia’s Participation Stands Out
Nvidia’s involvement is the most strategically significant element of the round. By backing SiFive, Nvidia signals that it sees value in the RISC-V ecosystem for data center roles outside its own GPU product lines. As AI infrastructure scales and hardware stacks diversify, RISC-V becomes a strong candidate for custom silicon across multiple layers of the data center. Nvidia appears to be hedging toward a future where its GPUs pair with purpose-built RISC-V processors across a heterogeneous compute architecture. Apollo’s participation reflects similar logic from the asset management side. These investors want exposure to the AI buildout without absorbing the capital commitments that GPU manufacturers and hyperscalers carry. Point72’s involvement adds further signal that institutional capital considers the RISC-V opportunity investable at this valuation.
What the Oversubscription Signals
The oversubscription at $3.65 billion reflects market confidence in purpose-built silicon. The first phase of the AI buildout concentrated on GPU capacity. The next phase is about optimising the full compute stack, including the networking, storage, and orchestration layers that make GPU clusters economically viable in production. SiFive is targeting that optimisation layer with processor designs it can customise for specific performance and efficiency requirements. Enterprise inference workloads are also diversifying fast. Operators now need processors built around latency, memory efficiency, and workload-specific throughput. SiFive’s RISC-V platform gives it the flexibility to serve those needs across applications where established vendors with proprietary architectures struggle to compete.
