Nvidia‘s semiconductors dominate the market for running the large language models that underlie generative AI programs. Now, a cohort of startups has emerged to take on the chip goliath.
Makers of AI chips are going niche, designing tailored semiconductors for running specific models with less energy and lower costs. While it will be an uphill climb to take on Nvidia, investors think chips with specialized use cases could attract investment.
“The big gorilla in the room is Nvidia,” said Lux Capital partner Grace Isford. “You could see growth by non-Nvidia players for specific specialized use cases.”
Investment in AI chips has not directly corresponded to the explosion of interest in generative AI, according to PitchBook data. In Q3, $810 million was raised across 26 deals, more subdued than the highs of 2021. Deal counts have been trending flat as investors focus more on what AI applications can do as opposed to what powers them.
“AI chip startups are circling the generative AI market, looking for insertion points. They’re looking to finally get commercial traction,” said PitchBook senior analyst Brendan Burke.
Beyond Nvidia, startups must also navigate competition from the Big Tech companies. Amazon, Microsoft, Alphabet and Meta have all either unveiled custom AI chips or have plans to do so soon. These tensor-processing units, or TPUs, are seen as a successor to Nvidia’s GPUs.
One specific entry point is AI cloud computing—the chips needed to power data centers. D-Matrix, based in Santa Clara, CA (where Nvidia is headquartered), raised a $110 million Series B in September for its semiconductors that target AI workloads in data centers. The round was led by Temasek, with Microsoft and Samsung Venture Investment among participating investors.
Several big AI chip startups are based in China, which is looking to beef up its domestic semiconductor industry. Enflame Technology, based in Shanghai, raised a 2 billion yuan (about $237 million) Series D from investors including Tencent and Golden Partners Capital in September. Its chips are designed for low-cost AI training and cloud computing.
While some startups are focused on cloud computing, others hope to move smaller chips to localized devices to hasten processing. Syntiant, which is backed by Microsoft’s M12, is making low-power AI semiconductors designed for consumer devices like hearing aids and smart speakers. The Irvine, CA-based company was valued at $456 million in a $56 million Series C1 earlier this year, according to PitchBook data.
Some pure-play chip startups are also gaining traction. Toronto-based Tenstorrent, which develops chips designed for AI workloads, raised $100 million led by Hyundai and Samsung Catalyst Fund in August. Kneron, another chipmaker designed to handle AI specifically, extended its Series B to $97 million in September. Horizon Ventures and Qualcomm Ventures led the funding.
“Strategic investors competing with Nvidia are willing to support innovative companies,” said Burke. “Generative AI will become embedded in all kinds of devices.”
Featured image by Chloe Ladwig/Pitchbook News