FuriosaAI raises $620m to scale its RNGD AI chips for global inference

A major funding milestone

FuriosaAI, a Seoul-based company specializing in AI inference chips, recently closed a $620m Series D funding round. This investment brings the company's valuation to approximately $2.9bn, a significant jump from its $735m valuation last year. The capital injection marks a major step in the company's effort to scale its business globally.

South Korea’s National Growth Fund acted as the lead investor, contributing $280m to the round. Other notable participants included the US firm Infinitum, which provided $110m, and several international investors from Singapore, Abu Dhabi, and Europe. Existing shareholders, including seed investors NAVER, DSC Investment, and KDB Capital, also contributed $125m.

The Export-Import Bank of Korea also participated, marking its first-ever direct investment in a startup. This diverse group of investors highlights the growing interest in hardware that can provide alternatives to the chips currently dominating AI data centers. The company plans to use this capital to increase RNGD chip production, enhance its software, and begin work on its next-generation hardware.

The RNGD inference chip

The company's primary product is the RNGD chip, which is pronounced as "renegade." This hardware is specifically designed for AI inference, which is the work of running trained AI models rather than the initial training phase. Because inference is a high-volume task for many applications, efficiency in this area is a major priority for hardware manufacturers.

The RNGD card is designed to be highly efficient, drawing only 180 watts and requiring only air cooling. FuriosaAI claims that the chip matches the throughput of Nvidia’s RTX PRO 6000 card while serving twice as many users per kilowatt. These specifications are intended to appeal to operators looking for ways to reduce energy consumption in large-scale deployments.

Manufacturing the RNGD chip is a significant logistical challenge. TSMC is currently producing 20,000 units for the company this year. However, FuriosaAI says it has already secured the necessary memory and factory capacity to produce more than 40,000 units by 2027, despite a looming memory shortage in the broader industry.

Market adoption and partners

Several major corporations have already begun integrating RNGD chips into their infrastructure. Samsung SDS launched a cloud service powered by the chips in July, and LG Group has utilized them to run its EXAONE models. Additionally, the Korean AI firm Upstage has deployed the hardware to power its search summaries.

FuriosaAI is also expanding its geographic reach through strategic partnerships. The company brought RNGD to the European market in July in collaboration with Equinix. Chief executive June Paik noted that the company is currently in discussions with many potential customers across the Asia-Pacific region and the Middle East.

The company's leadership also includes high-profile expertise from the industry. Charlie Foo, the founder of QUDA Capital Partners, serves as an advisor to FuriosaAI and is a former executive at Nvidia. This connection provides the startup with deep insights into the competitive landscape of AI hardware and the requirements of large-scale data centers.

The road to 2nm

While the RNGD chip is the current focus, FuriosaAI is already looking toward the future of AI hardware. The company is currently developing its third-generation chip in collaboration with Broadcom and other partners. This upcoming hardware will utilize a 2nm multi-die design, which involves connecting multiple smaller chips to act as one larger unit.

The next-generation chip will also incorporate HBM4 memory, a high-bandwidth memory technology essential for high-performance computing. By moving to a 2nm process, FuriosaAI aims to stay at the cutting edge of manufacturing capabilities as the industry moves toward smaller and more efficient transistors. This development is intended to maintain their competitive edge against established rivals.

The company's growth reflects a broader trend in the semiconductor space where investors are seeking alternatives to the current status quo. As the demand for AI continues to rise, the ability to move from design to volume production successfully is a key differentiator. FuriosaAI aims to prove it can meet this demand while providing a more energy-efficient path for global AI infrastructure.