Developing and deploying machine learning (ML) models is already a time-consuming and expensive affair. As AI use has accelerated over the past year, so has demand for related resources as companies have scrambled to access the computing power required to train ML models. This has helped fuel a scarcity of graphics processing units (GPUs), the specialized chips at the core of generative AI programs.
“What we’ve done is we get way more juice out of the available hardware for customers.”
– Gennady Pekhimenko, CentML
Enter CentML, which has secured $37 million CAD ($27 million USD) in seed funding and support from a host of big-name investors as the startup looks to help companies figure out what hardware they can use to boost the performance and reduce the cost of their ML models.
“Essentially, what we’ve done is we get way more juice out of the available hardware for customers,” CentML co-founder and CEO Gennady Pekhimenko told BetaKit in an interview.
CentML’s seed round, which closed in early September, was led by Gradient Ventures, Google’s AI-focused venture fund. It saw support from strategic investors including chip giant Nvidia, Deloitte Ventures, and Thomson Reuters Ventures, as well as existing backer Radical Ventures. Pekhimenko noted that Nvidia, Deloitte, and Thomson Reuters have also become partners, adding that engineering teams at the latter two firms will be using CentML’s product “very soon.”
The round consisted of an undisclosed combination of equity and simple agreement for future equity (SAFE) funding.
CentML isn’t the only Canadian AI firm to recently secure convertible funding from strategic investors; Toronto-based AI chip manufacturer Tenstorrent recently did the same from the likes of Samsung, Kia, and Hyundai. While Tenstorrent is looking to produce a new generation of AI chips, CentML is looking to get the most out of current hardware by optimizing ML models.
Pekhimenko, an ML systems expert, Vector Institute for AI faculty member, and University of Toronto associate professor, founded CentML in 2022 alongside a group with past experience working in AI, compilers, and ML hardware at Amazon, Google, Nvidia, and IBM. Today, the startup counts Amazon and Snowflake among its customers.
“Based on the quality of the founders at CentML … and the level of experience the whole team has, we believe that CentML is charting a unique course within the AI model optimization space,” Gradient Ventures general partner Zach Bratun-Glennon told BetaKit
CentML’s latest financing, which marks a particularly large seed round during any market, comes during a challenging time for Canadian tech. As the broader venture capital (VC) funding market has cooled, AI investment has been heating up since OpenAI’s release of ChatGPT late last year made the state of the technology and its promise more widely understood.
Radical Ventures led CentML’s initial $4.8 million CAD ($3.5 million USD) pre-seed round in June 2022. According to Radical Ventures partner David Katz, the AI-focused VC firm was betting that AI innovation and usage would become constrained by inefficiencies in the market for accelerated computing hardware and AI chip performance.
That prediction has come to fruition: since then, the chip market has struggled to keep pace with demand as companies have raced to develop and deploy AI applications as quickly as possible, resulting in a bottleneck in the global supply of chips capable of powering AI solutions. As Pekhimenko put it, the AI world is in a “starvation period” currently as a result.
“The rush for adoption of generative AI has significantly pulled forward demand for state-of-the art chips, exposing significant weaknesses in supply chains,” Katz told BetaKit. “This, exacerbated by geo-political factors, has made compute access and optimization for cloud providers and their customers of critical importance. Access to chips has become a huge problem, even for the largest tech companies and indeed entire countries.”
According to Katz, there are few teams with the expertise required to optimize the selection and performance of a variety of chip types and fix this issue at scale. He described Pekhimenko and the CentML team as “amongst the top people globally capable of solving the most complex problems associated with this challenge.”
Per Pekhimenko, this broader shift in the AI market has been reflected in his conversations with investors, which have changed dramatically since June 2022. While fundraising for CentML’s pre-seed round back then, many VCs asked the CEO whether the market for the startup’s solution was big enough. This time around, he said, “Everyone realizes that the cost is there and the use case is there.”
“We believe that AI will facilitate all sorts of new use cases and capabilities for technology, and AI runs on the massive parallel compute capacity of GPUs,” said Bratun-Glennon. “With CentML, companies and developers can utilize different GPUs to greater performance than previously possible. It’s an exceptional technology and capability that helps to relieve one of the greatest constraints in AI now and for at least the medium-term.”
CentML plans to use its latest funding to open a new office in the San Francisco Bay Area and double its 32-person team by the end of 2024 with hires in engineering, sales, and marketing across Toronto and San Francisco. The startup intends to make its product more widely available by the end of 2023.
Feature image courtesy CentML.