FuriosaAI funding, valuation and investors

Product & use cases

RNGD is a 180W PCIe inference accelerator with 48GB of HBM3. Eight cards form a 4U NXT server. Furiosa sells hardware for on-premises use and makes capacity available through Samsung SDS; its software exposes familiar model-serving interfaces, including an OpenAI-compatible endpoint in the Upstage deployment.

• Enterprise on-premises inference in existing air-cooled data centers.
• Cloud inference through one-to-eight-card Samsung SDS instances.
• High-concurrency API and search-summary workloads for large open models.

Key facts

• Closed a $620 million Series D at a company-reported $2.9 billion post-money valuation in October 2026.
• RNGD entered mass production in January 2026; Furiosa says it plans 20,000 cards in 2026 and more than twice that volume in 2027.
• Named production relationships include Samsung SDS, LG Group and Upstage.

Competitive landscape

Edge: Furiosa combines a purpose-built tensor-contraction architecture with a production software stack and low-power, air-cooled servers that can fit existing enterprise racks. Its advantage remains company-claimed until wider independent benchmarks and repeat deployments are available.

Positron AI is the closest funded inference-silicon comparison. Etched is more specialized around transformer inference, Tenstorrent spans broader AI compute and licensing, and Nvidia remains the incumbent platform with the deepest software ecosystem.

• Positron AI direct

  Purpose-built inference silicon focused on memory efficiency and lower serving cost.

• Etched direct

  Transformer-specific inference accelerators backed by substantially higher private valuations.

• Tenstorrent adjacent

  AI processors, chiplets and IP licensing across a broader compute stack.

• Nvidia incumbent

  Dominant GPU platform with the largest software and developer ecosystem.

Industries

Market / IPO context