Skip to content
News

AI Company FuriosaAI Secures $125 Million Funding to Expand Manufacturing of Advanced AI Inference Chip Prototypes

South Korean semiconductor firm, FuriosaAI, unveiled that they have successfully concluded a funding round worth $125 million, aiming to construct a novel base for artificial intelligence computation.

,   and  Administrator
2 min read
Next-Generation AI Inference Chip Production Growth Funded by $125M Investment from FuriosaAI
Next-Generation AI Inference Chip Production Growth Funded by $125M Investment from FuriosaAI

AI Company FuriosaAI Secures $125 Million Funding to Expand Manufacturing of Advanced AI Inference Chip Prototypes

FuriosaAI, a semiconductor company specializing in building a new type of AI chip for data centers and enterprise customers, has announced the completion of a $125 million Series C bridge funding round. The investment was supported by a number of new and existing investors, including the Korea Development Bank, Industrial Bank of Korea, Keistone Partners, and Kakao Investment.

The new funding will be used to accelerate mass production of Furiosa's RNGD AI accelerator to support global enterprise customers and lay the groundwork for the development of Furiosa's next-generation chip. FuriosaAI's mission is to create a sustainable hardware and software stack for AI computing, providing global access to powerful AI.

Jeehoon Kang, a renowned expert in parallel systems from KAIST, has joined Furiosa as Chief Research Officer. Kang's expertise will undoubtedly contribute to the company's mission, as Furiosa was founded to solve the problem of unsustainable power consumption, operating costs, and infrastructure limitations of legacy GPU-based hardware for AI compute.

Furiosa's RNGD AI accelerator achieves 2.25x better LLM inference performance per watt compared to GPUs primarily through its innovative Tensor Contraction Processor (TCP) architecture. This architecture is co-designed with its software stack specifically to optimize tensor contraction operations, the core computations in deep learning. By providing the compiler a high-level view of these tensor operations, RNGD creates an efficient, custom dataflow for each computation that minimizes wasteful data movement, significantly improving power efficiency and performance in real-world AI models.

The key details of the RNGD AI accelerator include its efficiency in handling diverse AI models and deployment scenarios, its co-designed hardware-software stack that optimizes data flow tailored to tensor contractions, and its design that leads to dramatically lower total cost of ownership while enabling faster scaling and more flexible deployments. This innovation has been validated in enterprise adoption, notably by LG AI Research using RNGD to run EXAONE foundation models.

Youngjin Cho, a silicon and SoC expert who was previously a Corporate Vice President at Samsung Electronics, has been appointed Vice President of Hardware at Furiosa. Since the company's founding in 2017, Furiosa has raised a total of $246 million from investors and is currently valued at $735 million.

For more information about FuriosaAI, please visit their website at furiosa.ai. Oliver Libaw is a contact for press inquiries at [email protected].

[1] FuriosaAI, "FuriosaAI's RNGD AI Accelerator Achieves Breakthrough Efficiency for AI Inference," [Accessed 20 April 2023]. [2] LG AI Research, "LG AI Research Adopts Furiosa's RNGD AI Accelerator for EXAONE Foundation Models," [Accessed 20 April 2023]. [3] AnandTech, "Furiosa's RNGD AI Accelerator: A New Approach to AI Inference," [Accessed 20 April 2023].

Read also:

Related

Latest