VCs take aim at semiconductor innovation

Semiconductor startups had an exceptionally robust 2021 as investors increased expectations for chipmakers to fuel higher efficiency and greater utility for AI processing.

While VC-backed startups are unlikely to overtake the likes of Nvidia and Intel, this push could help companies step up the development of competitive chip architectures needed for specific use cases in 2022.

VC funding for global semiconductor startups more than tripled year-over-year in 2021, with $9.9 billion invested across 170 deals, according to PitchBook data. The segment includes companies building AI chips, intelligent sensors and devices, and compression algorithms that optimize AI and machine learning models for deployment.

Today’s enthusiasm for semiconductors is accompanied by critical supply bottlenecks and chip shortages that continue to drag on in 2022 despite a rise in supplies. Furthermore, AI chips are undergoing a shift in demand from data centers to emerging use cases such as smartphones and wearable devices. This has led to a faster growth rate for integrated processors compared to discrete processors, according to a recent PitchBook report on the sector.

The AI ​​and ML semiconductor market reached an estimated $35.9 billion in 2021, achieving 60% growth over 2020 likely as a result of the pandemic. But startups in the sector are far from operating at a scale that can threaten industry giants.

“Nvidia, AMD and Intel are innovating at too fast a rate for these challengers to overtake them in the short-term,” said Brendan Burke, an emerging tech analyst at PitchBook. It will take a misstep by any of these leaders to open market share for challengers this year, given the capital investments these players are making, he said.

Last year, SambaNova Systems’ $676 million Series D set a VC deal value record for a pure-play AI chip company, making it the most valuable AI chip startup outside of China. While the company claims that its data center training chips are superior to Nvidia’s for training large language models, it has only recently begun to commercialize and has not been independently benchmarked by MLPerf—a consortium that aims to provide unbiased evaluations of training and inference performance for hardware, software and services.

Another notable startup in the space is Cerebras Systems, which has designed the largest neural network chip in the market. In November, the company raised a $250 million Series F led by Alpha Wave Ventures.

This could be the year for more startups to participate in leading benchmarks for AI inference, including MLPerf Edge Inference and Tiny, to demonstrate the advantages of their architecture compared to incumbents. The number of submitters in both categories have the chance to double this year, Burke said.

“VC-backed startups have developed great technology for some specific use cases but don’t outcompete incumbents like Intel, AMD and Nvidia on cost, so they don’t actually ship many chips,” Burke said.

Featured image by Weiquan Lin/Getty Images

Leave a Comment