The generative AI boom is driving demand for AI chips specialized for training and running generative AI models, and major players, from venture capital firms to startups, are rushing to get in on the action.
SoftBank's Masayoshi Son is reportedly looking to raise $100 billion for a chip business to rival tech giant Nvidia, while OpenAI is said to be in talks with investment firms to launch an AI chip manufacturing venture.
AI chip startup Axelera has kept a relatively low profile, but it has managed to garner the support of backers, including Samsung, by focusing on a niche area in the burgeoning AI chip market: chips that run AI on edge devices.
“There's no denying the AI industry has the potential to be transformative across many sectors,” Axelera co-founder and CEO Fabrizio Del Maffeo told TechCrunch in an interview. “But to truly harness the value of AI, organizations need solutions that deliver high performance and efficiency while balancing costs.”
Headquartered in the Netherlands with around 180 employees across offices in Belgium, Switzerland, Italy and the UK, Axelera designs and supplies AI-powered chips and systems for applications in security, retail, automotive and robotics to partners manufacturing B2B edge computing and IoT products.
Axelera was born out of an effort to build a highly efficient AI chip architecture, led by Del Maffeo and group at Belgian technology research institute Imec and IBM researcher Evangelos Eleftherio and group in Zurich. The founding team nurtured much of Axelera within Bitfury Group, a blockchain company specializing in Bitcoin hardware.
Axelera's AI hardware stack is characterized by the instruction set architecture (ISA) RISC-V and in-memory computing.
An ISA is a chip's underlying technical specification that describes how software controls the chip's hardware. Chip designers typically license existing ISAs from major chip manufacturers such as Arm or Intel, but RISC-V provides an open, royalty-free alternative. In-memory computing refers to performing calculations in a system's RAM to reduce the latency introduced by storage devices.
Axelera isn't the first company to venture into in-memory and/or RISC-V-based architectures for AI chips.
NeuroBlade is developing chips that integrate both compute and memory into a single hardware block for data processing. MemVerge, GigaSpaces, Hazelcast and H20.ai also offer in-memory hardware solutions for AI and data analytics applications, while Hyundai Motor Group and Samsung-backed Tenstorrent sells AI processors built on RISC-V and other related IP.
One of Axelera's accelerator cards. Image courtesy of Axelera
Axelera has sought to differentiate itself by offering both the chip hardware and the software that manages and deploys AI models on that hardware, and by all appearances, this strategy seems to be working.
Accellera announced on Thursday that it had closed a $68 million Series B funding round, bringing its total funding to $120 million. Investors in the round included the European Innovation Council Fund, the Innovation Industrial Strategy Partnership Fund, Invest-NL and the Samsung Catalyst Fund.
Del Maffeo said the new funding will be used to expand into new markets ahead of full production of Axelera's flagship Metis AI platform, scheduled for the second half of 2024. Axelera is also eyeing the data center chip market, with preliminary plans to fund research and development of chips aimed at high-performance computing use cases.
“Metis is expected to enter full production in the second quarter, with mass production starting in the third quarter,” Del Maffeo said. “Axelera AI is currently developing a new generation of products for computer vision, large language models and large multimodal models. This new product family will be launched later this year and is expected to enter full production in 2025.”
The challenge will be shipping AI chips at scale, and competing in the AI chip race against countless other competitors, many of whom are heavily backed: VC-backed chip startups have raised about $5.3 billion so far this year in just 175 deals, according to a Crunchbase report from June.
But the payoff could be substantial: The AI chip market could generate $67 billion in revenue by 2027, according to data from Statista and Market.us. It's possible that Axelera could overtake incumbents like Nvidia soon, or even someday. (Nvidia is estimated to have a 70% to 95% share of the AI chip market, according to Mizuho Securities.) But capturing even a slice of the market would be a meaningful win.
“This funding supports our mission to democratize access to AI from the edge to the cloud,” Del Maffeo said, adding that Accellera has “dozens” of enterprise customers. “Expanding our product line beyond the edge computing market will enable us to address industry challenges in AI inference and support current and future AI processing needs.”