Rebellions lands $124M to develop its new AI Rebel chip with Samsung

Rebellions, a South Korean fabless AI chip startup, said today it has closed $124 million (165 billion KRW) in a Series B round of funding to develop its third AI chip, called Rebel. The startup will also use the new capital, oversubscribed with an initial target of $90 million, to ramp up production of its data center-focused chip, Atom, and for hiring.

This Series B values the three-year-old startup at approximately $658 million (880 billion KRW) post-money, CFO of Rebellions Sungkyue Shin said in an exclusive interview with TechCrunch. This latest capital infusion brings the total raised to around $210 million since Rebellions’ inception in 2020.

KT, the South Korean telecom giant, led this latest round as a strategic investor. Previous backers Temasek’s Pavilion Capital and Korea Development Bank, and new investors including Korelya Capital and DG Daiwa Ventures, also participated.

Rebellions’ fundraise comes at a key moment in the chip industry, specifically around the development and use of AI chips.

Nvidia is the AI chip market leader, its name synonymous with the AI boom that is currently sweeping the technology world. Many have observed how Nvidia has thrived in part because of the moat created around an ecosystem of hardware and software. But it’s far from game-over for the rest of the field. Data processing and related high costs continue to be major issues when it comes to AI applications, so the scramble continues in the search for innovative breakthroughs to improve these.

Developments are coming from multiple fronts. Big Tech titans such as Google, Amazon, Apple and Microsoft develop or have their own chips to integrate AI into their products and services. Open AI chief executive officer Sam Altman reportedly visited South Korea last week to meet the country’s chip industry leaders, Samsung and SK Hynix. Beyond that, Open AI is said to be raising billions of dollars to set up chip fabrication factories, to make its own AI chips. And there are a number of startups beyond Rebellions bringing new concepts to the table to speed up processing while improving efficiency.

Team up with Samsung

This fundraise — which has been rumored for months — comes on the heels of other moves at the startup. Last October, Rebellions announced that it would develop its newest Rebel chip in partnership with Samsung Electronics, building on a relationship it initially forged around its Atoms chips. The two companies aim to complete the development of Rebel by the end of this year and start mass production in 2025, Shin said, adding that the next-generation AI chip will target the generative AI market running large language models (LLMs) and hyperscalers.

Shin told TechCrunch that Rebel will use Samsung Electronics’ 4-nanometer fabrication process, and that its AI chip will be deployed in Samsung’s advanced memory chip technology HBM3E, designed to handle high bandwidth memory, used for building and operating large language models. Rebellions’ unique selling point is a claim that its technology and products have more versatility than customized AI chips, meaning they can support various generative AI models that need AI accelerators.

The company CFO stressed that Rebellions will cooperate with Samsung from co-development and chip design to mass production of Rebel. There is a second motivation for Samsung’s work here: Aside from its efforts in chips, South Korea’s largest memory chip maker has been working on its own generative AI model, Samsung Gauss.

ATOM and ION

It’s also been working with customers using its previous generations of chips. In May 2023, Rebellions’ strategic investor, KT, installed Atom, Rebellions’ data-center targeted AI chip, in its cloud-based neural processing units (NPU) infrastructure. Rebellions says it expects to generate revenues from Atom in the second half of this year and will continue to produce that chip model via Samsung’s 5-nanometer fabrication process. Atom is designed for data centers and language models of up to 7 billion parameters, while Rebel targets larger large language models, Shin noted.

Meanwhile, the startup’s first AI chip, Ion, which was launched in November 2021, is in the process of qualification testing in the U.S. and has yet to sign on any commercial customers. Ion is designed for edge computing and one key use case, the company believes, will be in financial services applications, where larger institutions building their own hardware could use the chips to power stock prediction and trading applications.

Rebellions CEO Sunghyun Park, a former quant developer at Morgan Stanley in New York, and four co-founders set up the AI chip startup in 2020.

source

Leave a Reply

Your email address will not be published. Required fields are marked *