Venture capital slowdown hasn’t hit AI chip designers, yet • The Register
Analysis Some AI chip startups are managing to raise capital from investors despite operating in a crowded market of competitors where venture capital funding has plummeted in the past year.
Venture-backed AI chip designers that have raised funding rounds in recent months include Israel-based NeuReality, Netherlands-based Axelera as well as SiMa.ai, Quadric, and EnCharge AI, three US firms that are headquartered in Silicon Valley.
These startups convinced investors to part with tens of millions of dollars in an investment environment that is much more conservative due to this year’s shaky economy, jostled by inflation and rising interest rates.
As The Register recently reported, global VC funding for semiconductor startups in 2022 declined 46 percent to $7.8 billion, as of December 5, reflecting increased scrutiny for these capital-intensive firms.
While funding took a deep plunge for semiconductor startups this year, the decline in the total number of known funding rounds has not been as pronounced, dipping 20 percent in 2022 to 618 deals.
Out of the five startups that raised recent funding rounds, four of them — Axelera, SiMa.ai, Quadric, and EnCharge AI — are focused on AI chips for running inference on edge devices, where processing data at high speeds while using the least energy possible in constrained form factors is paramount.
That may not be a complete coincidence. As of October 31, VC funding for inference-based AI chip designers had exceeded money raised by startups focused on processors for training, the first step in developing AI applications before turning to inference for real-world deployment. That’s according to a November research note written by PitchBook Senior Analyst Brendan Burke.
Burke said this is a reversal of a trend seen in the last four years, where funding for chip startups focused on training or both training and inference exceeded capital for those only working on inference.
“There is a cyclical component to this pattern, given the limited need for training companies to raise funding each year, yet we observe that inference-focused companies are achieving significant commercial partnerships during the economic recession,” he said.
This increased need for inference chips is coinciding with a projected growth in spending on AI chips for edge cases, a rather nebulous term that, in the case of Pitchbook’s report, includes the PC, automotive, and industrial markets. Burke said the PC and automotive markets for AI chips grew more than 22 percent in 2022, faster than spending on such silicon for datacenters.
“Automotive and edge computing demands are driving more commercial agreements for inference-focused chips than for cloud training chips,” the analyst added.
Not all startups working on AI chips have been lucky this year.
Mythic, a Texas-based startup that was developing analog chips for edge AI use cases, ran out of funding from investors before it was able to generate revenue, a top executive said in November.
Then there’s Graphcore, a British startup that has commercialized datacenter chips for training and inference. The company reportedly had its private valuation slashed by $1 billion after losing a key deal with Microsoft, among other financial woes.
Ruta Belwalkar, a private investor and chip designer, previously told us she wouldn’t be surprised if more chip design startups ended up getting acquired or shut down in the next year because they failed to transition fast enough from research and development to commercialization.
It’s not just investors who are finding some AI chip startups still palatable in a weaker economy. Some industry veterans are also making the leap.
For instance, Lightelligence, a Boston-based optical AI chip startup, recently hired Weifeng Zhang, Alibaba’s former chief scientist of heterogeneous computing, along with Wayne Wu, the previous head of AMD’s PCIe design team, and Hal Conklin, who was most recently Arm’s vice president of worldwide channel sales.
To them, we say good luck. ®
