As Tencent first bid on Chinese AI chip startups, Enflamed seemly had the faith to compete with and even surpass Nvidia one day, but will the company be valued to almost 1 billion one year later as its counterpart?
A Chip. PHOTO: Credit to Unsplash
Chinese AI maker startup Enflame（Suiyuan Technology, 燧原科技）has raised CNY 340 million (USD 50 million)in Series Pre-A funding led by Tencent, Zhen Fund（真格基金）, Delta Capital, Yunhe Partners（云和资本）and Summitview Capital（武岳峰资本）in August 2018.
Founded in March 2018, Enflame is a Shanghai-based startup developing cloud-based deep learning chips for AI training platforms. The company focuses on developing chips based on a self-developed unique architecture design to achieve high computational power and efficiency with low cost for deployment in major machine learning frameworks. The company has research and development centers in Beijing and Shanghai.
Cerebras, a California-based startup building tailor-made chip for deep learning, raised its first round of investment–USD 27 million at a USD 64.5 million valuation—in the mid of 2016, the year Cerebras was founded. In three rounds of funding, Cerebras has raised USD 112 million, and its valuation has soared to USD 860 million as of August 2017, according to PitchBook.
The two companies have two things in common, one is they both focus on self-developed unique architecture design chips and the other is the correlation with AMD(Advanced Micro Devices).
Ever since the capabilities of Nvidia's graphics processors(GPUs) were recognized to be ideally suited to run deep learning algorithms, the company jumped at the heart of the AI revolution.
However, a number of startups like Cerebras and Cambricon believe they could build chips that outperform GPUs in deep learning applications. According to Feldman, GPU is not the best solutions for machine learning, it's just better than Intel's CPU(central processing unit), reported by Forbes.
As for cloud training platform, chipmakers battle to power artificial intelligence in cloud data centers as well. CFRA Research analyst Angelo Zino reckoned that Nvidia's Volta architecture is in a dominant position for most training applications in the cloud over the next couple of years, and also view fears of cloud providers designing their own processors for AI as overblown.
Amazon and China's Baidu keep broadening their use of Xilinx's FPGAs. Microsoft, Alphabet, Facebook, Alibaba and Tencent, these leading cloud service providers are pursuing artificial intelligence and exploring chip development to produce internal chip. While application-specific integrated circuits (ASICs) chips for artificial intelligence can cost less in production but remain high-expensive and time-consuming to design, according to Zino.
Looks like Enflame faces the problem, as they design chips special for AI training platform.
Andrew Feldman, co-founder and CEO of Cerebras once funded SeaMicro which was acquired by AMD for USD 334 million in 2012. ZHAO Lidong（赵立东）formerly worked for AMD China as a Sr. director and a former VP of Tsinghua Unigroup & President of RDA（锐迪科）, build his core chip scientist team with people possessing an average 15-years-experience in high-end chipset development and mass production.
October 2018, Enflame has purchased multiple licenses of Arteris FlexNoC interconnect IP(intellectual property) for use as the on-chip communications backbone of their artificial intelligence (AI) training chips for use in cloud data centers. Leveraging Arteris's NoC(network-on-a-chip) technology, Enflame is allowed to develop AI training systems-on-chip (SoC) that implement neural networks.
These chips use multiple hardware accelerators to increase the speed and accuracy of neural network training while reducing data center power consumption.
Baidu now plays in a variety of AI fields thanks to a new chip announced in July 2018 called Kunlun（昆仑）and an alliance with Intel. Alibaba invested a number of AI chip companies including DeePhi（深鉴科技） and Cambricon（寒武纪）. Though Tencent has an advantage in social big data, it remained the last one of BAT to deploy in AI fields. As Tencent first bid on Chinese AI chip startups, Enflamed seemly had the faith to compete with and even surpass Nvidia one day, but will the company be valued to almost 1 billion one year later as its counterpart?