The generative AI boom is driving demand for AI chips that are purpose-built to coach and run generative AI models. And major players, from VCs to startups, are jockeying to get in on the ground floor.
This is the son of Masayoshi SoftBank supposedly wants to boost $100 billion for a chip initiative that will compete with tech giant Nvidia. OpenAI, meanwhile, is he said will hold talks with investment corporations regarding the launch of a project involving the production of integrated circuits based on artificial intelligence.
A startup producing AI systems Akselera he kept a relatively low profile. Nevertheless, it managed to win over supporters, including Samsung, in part by focusing on a area of interest in the burgeoning AI chip market: chips that run AI on edge devices.
“There is no denying that the artificial intelligence industry has the potential to transform many sectors,” Fabrizio Del Maffeo, one of Axelery’s co-founders and its CEO, said in an interview with TechCrunch. “However, to truly realize the value of AI, organizations need a solution that delivers high performance and efficiency while balancing costs.”
Axelera – headquartered in the Netherlands and employing around 180 people with offices in Belgium, Switzerland, Italy and the UK – designs AI-enabled chips and systems for applications such as security, retail, automotive and robotics, which it supplies to partners manufacturing B2B edge computing and IoT products.
Axelera was founded as a results of efforts led by Del Maffeo and a group at Imec, a Belgian technology lab, along with Evangelos Eleftheriou and a group of IBM researchers in Zurich who built a high-performance AI chip architecture. The founding team incubated most of Axelera at Bitfury Group, a blockchain company specializing in Bitcoin hardware.
The distinguishing features of Axelera’s AI hardware stack are the RISC-V instruction set architecture (ISA) and in-memory processing.
ISA is a technical specification that underlies integrated circuits and describes how software controls the hardware of an integrated circuit. Chip designers typically license an existing ISA from a large chipmaker such as Arm or Intel, but RISC-V provides an open alternative with no licensing fees. In terms of in-memory computation, this refers to running computations in the system’s memory. Aries to scale back latency introduced by storage devices.
Axelera is not the first to try its hand at in-memory and/or RISC-V-based architectures for AI chips.
NeuroBlade is developing chips that mix compute and memory into a single hardware block for processing data. MemVerge, GigaSpaces, Hazelcast, and H20.ai also offer in-memory hardware solutions for AI and data analytics applications. Elsewhere, Tenstorrent, backed by Hyundai Motor Group and Samsung, sells AI processors and other related IP built on RISC-V.
Axelera has tried to distinguish itself by providing each chip firmware to administer and deploy AI models on that hardware. And from all appearances, this strategy appears to be working.
On Thursday, Axelera announced it had closed a $68 million Series B funding round, bringing its total funding to $120 million. The round includes funding from the European Innovation Council Fund, the Innovation Industries Strategic Partnership Fund, Invest-NL and Samsung Catalyst Fund.
The new funds will probably be used to expand into new markets ahead of full production of Axelera’s flagship Metis AI platform in the second half of 2024, in response to Del Maffeo. Axelera also has its eye on the data center chip market, with initial plans to fund research and development of chips designed for high-performance computing applications.
“Metis started full production in the second quarter, with mass deliveries in the third quarter,” Del Maffeo said. “Axelera AI is currently developing a new generation of computer vision products, large language models and large multimodal models. This new family of products will be presented later this year, with full production starting in 2025.”
The challenge will probably be shipping AI chips at scale — and competing with countless others in the AI chip race. Many rivals have powerful support; Crunchbase report in June shows that VC-backed chip startups have raised nearly $5.3 billion this 12 months in just 175 deals.
However, the reward may be significant. According to Statista and Market.us data, AI Chip Market Could Reach $67 Billion revenues by 2027. Axelera has little probability of displacing vendors like Nvidia from the market anytime soon, if at all. (Nvidia has estimated (Between 70% and 95% of the AI chip market share, in response to Mizuho Securities.) But capturing even a fraction of the market could be a significant win.
“The funding supports our mission to democratize access to AI, from the edge to the cloud,” Del Maffeo said, adding that Axelera has “dozens” of enterprise customers. “By expanding our product lines beyond the edge computing market, we are positioned to address the industry’s AI inference challenges and support current and future AI computing needs.”