Vulnerability Disclosure at FuriosaAI

Job description

The full description lives on the employer’s site. Open the listing.

Culture: “First principles” engineering approach – solving core problems in AI acceleration. Emphasis on full‑stack innovation spanning hardware, software, and algorithms.
Locations: Seoul (HQ), Santa Clara CA, Lisbon, Singapore, Germany. Remote roles available in the US and Europe (non‑Portugal).
Team: 140+ employees with a technical majority (94/140 in technical roles). Notable talent from Samsung, SK Hynix, Google, Naver, and KAIST.
Open roles (54 active postings, growing ~15% monthly): Machine Learning Engineer (Agentic AI), AI Research Engineer, Compiler Engineer, Hardware Architect, Design Verification, Field Application Engineer, Solutions Architect, and more.
Perks: Not explicitly listed, but the global, distributed setup and cutting‑edge AI chip work are key draws. Contact: recruit@furiosa.ai.

One-liner: FuriosaAI designs and develops data center AI accelerators purpose-built for efficient inference of large language models (LLMs) and multimodal models.
Entity Type: Private (Series C)
Headquarters: Seoul, South Korea
Founded: 2017
Founders: Three co‑founders (names not publicly disclosed) with backgrounds in hardware, software, and algorithms, previously at AMD, Qualcomm, and Samsung.

Primary industry: Semiconductor Manufacturing / AI Hardware
Target customers: B2B – hyperscalers, enterprise data centers, cloud providers, and AI‑focused companies deploying LLM and agentic workloads.
Mission: “Make AI computing sustainable so everyone on Earth has access to powerful AI.”

RNGD (Renegade): Second‑generation data center accelerator (TSMC 5nm) for LLM and multimodal inference. Features a Tensor Contraction Processor (TCP) architecture (ISCA 2024) instead of fixed‑size matmul primitives, delivering high throughput and low latency. 4 petaFLOPS, 12 TB/s memory bandwidth, 384 GB HBM3 capacity.
NXT RNGD Server: Multi‑card inference appliance (3 kW, air‑cooled) for agentic systems. Contains 8× RNGD cards, supports containerization, SR‑IOV, Kubernetes. Designed for on‑premises, colocation, or managed environments.
Vision NPU: First‑generation neural processing unit (Samsung 14nm, later Samsung Foundry) for computer vision, deployed with Kakao and other partners.
Furiosa Software: Comprehensive toolchain for LLM inference and agentic workloads – compilation, optimization, and production deployment. PyTorch 2.x integration, Hugging Face ecosystem support.

Valuation / Market Cap: Not disclosed.
Key Metric: Total funding > $100 M (includes a Series C of $123.1 M in August 2025; earlier rounds include a $80 M Series C in 2023 and seed funding in 2017 – some reports show an $80 B figure, likely a data error).
Notable Investors/Partners: DSC Investment, Korea Development Bank, Naver (largest Korean internet provider). Partners: TSMC, ASUS, SK Hynix, Samsung, GUC, Hugging Face, LG AI Research.
Growth Signals: 140 employees (+43.4% YoY). Revenue of ~$1 M (early stage). Expanded to 5 countries (US, Portugal, Germany, Singapore). RNGD sampling began in 2025 with global technology companies; mass production scheduled for 2026. Established a European flagship office in Lisbon.

Tensor Contraction Processor (TCP) architecture: Designed for tensor contraction (the fundamental operation of modern deep learning) rather than rigid matrix‑multiply instructions, enabling higher efficiency and programmability.
Full‑stack approach: Hardware + software + compiler optimizations, ensuring seamless model deployment from PyTorch to production.
Power efficiency: 3 kW NXT RNGD Server delivers up to 4× more inference capacity per rack compared to alternatives, and works with standard air‑cooled data centers (15 kW/rack limit).
Programmability: Compiler‑based flexibility to adapt to evolving AI models without hardware changes.

Inference‑first: Prioritizing inference (not training) for LLMs and agentic AI, where data center power density is a growing bottleneck.
Cost‑efficient scaling: Lowering total cost of ownership through reduced energy draw, fewer racks, and standard cooling.
Global expansion: Building presence in North America, Europe, and Asia‑Pacific to capture enterprise and cloud customers.
Ecosystem development: Through the Furiosa Access Program, enabling customers to evaluate and integrate RNGD hardware and software.

Sources