The stealthy Chinese startup closes a second round of funding, tapes out AI inference accelerator for the data center.
Chinese startup Vastai Technologies has closed a second major round of funding, a Series A+ raising RMB 500 million (about $77 million). The fabless semiconductor startup is still in stealth mode, but has already taped out its first chip, an AI inference accelerator for the data center.
Vastai Technologies raised $50 million in a previous Series A round in November 2020. The company’s total funding to date is around $133 million, including an initial seed round of $8 million.
In an interview with EE Times, Vastai Technologies CEO John Qian said that he plans to invest the funding in the company’s products, IP and people.
“To make a good product, it takes that much [funding] to get it done,” Qian said. “The $50 million we raised in the last round is being used to develop our current product. Hardware alone won’t sell itself — it has to have its software, and we have to build an ecosystem to make it easier for customers to switch to our hardware. So we are focusing on a few applications at the moment, and we’re going to put together teams to help our customers migrate from their current hardware to our hardware.”
The $77 million from the latest funding round will be used to develop future products, he added.
The company has already taped out silicon for its first chip, an AI inference accelerator for both public cloud and enterprise data center applications. There are more and more enterprises in China wanting to have their own Internet Data Centers, Qian said, because they want to control their own hardware in order to optimize for their specific application.
“We view this as a big opportunity and we want to focus on that,” Qian said. “[Enterprise customers] are [keen on] reducing op-ex, so this is where we’re targeting. Of course, our product can go into the public cloud as well.”
Qian believes there is still an opportunity for startups like Vastai Technologies in the public cloud — even though hyperscalers from AWS and Google to Alibaba and Baidu are making their own AI accelerators to meet their specific needs.
“They still need some help, in my opinion, because [their solutions are] more specific, tailored to their own application,” Qian said. ASICs “tailored too much to your application may not scale because the algorithms you use might change, and you have to put in millions of dollars to invest in this silicon. If you have to recycle your servers pretty much every three years, it’s an investment, a commitment you have to make. I’m not sure they will continue to do that.”
Vastai’s first product is a general-purpose inference accelerator for the data center. It is suitable for computer vision and video processing applications, but can also handle natural language processing (NLP) workloads including Bert. The company says these two workloads constitute around 90% of data center AI workloads today.
Asked about future product lines and applications, Qian said that the same chip already taped out could also handle some edge applications. It can be used as a PCI accelerator alongside a host CPU at the edge, as it can be scaled down to run in the 15-W range (the chip can be run at between 15 and 150 W). Further into the future may come a training chip, but for now the focus is on inference.
“In terms of inference and training, we do see that the inference market is growing much, much faster than the training market,” Vastai Technologies CTO, Louis Zhang, told EE Times. “So that’s where we are targeting initially with our with our chip. Also, it’s also easier for the customer to switch from GPU to AI accelerator chips for inferencing versus training, because for training the software ecosystem is very hard to break; it’s much easier for the customer to adopt customized silicon for inferencing. [Inference] is where we see the market, and the money, is.”
Vastai Technologies has its major operations in China, with design centers in Shanghai and Beijing, plus an R&D center in Toronto, Canada.
The latest round of funding was led by Matrix Partners China and China Internet Investment Fund and joined by existing shareholders Redpoint China Ventures, 5Y Capital, SAIF Partners, Glory Ventures, Sirius Capital and Yuan Capital.
This article was originally published on EE Times.
Sally Ward-Foxton covers AI technology and related issues for EETimes.com and all aspects of the European industry for EETimes Europe magazine. Sally has spent more than 15 years writing about the electronics industry from London, UK. She has written for Electronic Design, ECN, Electronic Specifier: Design, Components in Electronics, and many more. She holds a Masters’ degree in Electrical and Electronic Engineering from the University of Cambridge.