Recently, the AI startup "Infini-Quest" affiliated with Tsinghua University has garnered widespread attention with its nearly $500 million Series A funding round. This company, established only one year and four months ago, has already surpassed the $1 billion mark in total funding, becoming the king of financing in the domestic AI infrastructure field. Its shareholder lineup is nothing short of luxurious, attracting state-owned capital and local funds from across the country, as well as renowned venture capital firms like Shunwei Capital and DCVC, even brokerage direct investments and industrial investments have flocked to join.
How did "Infini-Quest" rise to prominence? This can be traced back to the strong background of its founding team. Most of the team members come from Tsinghua University, particularly the NICS-EFC lab, which was established in 2008 and has deep accumulations in the field of electronic engineering. The head of the lab and the director of the Department of Electronic Engineering at Tsinghua University, Wang Yu, is one of the founders of "Infini-Quest." CEO Xia Lixue and other co-founders are all Wang Yu's students, with rich research experience in the field of artificial intelligence, making them an elite team from "Tsinghua Electronics."
As early as 2016, the NICS-EFC lab at Tsinghua University incubated the AI chip company Deephi Technology, which was acquired by the global FPGA giant Xilinx for $300 million in 2018. Unlike Deephi Technology, which improved chip efficiency at the algorithm level, "Infini-Quest" chose a unique path: through the collaborative optimization of software and hardware, to achieve more efficient computing power support.
At the end of 2022, the emergence of ChatGPT set off a wave of the large-model era, bringing an excellent opportunity for the development of "Infini-Quest." Xia Lixue pointed out that the emergence of large models means that the same model can support multiple scenarios, allowing them to maximize the use of their previously accumulated technology and experience.
The reason why "Infini-Quest" can attract investors, in addition to the strong team and the opportunities of the large-model era, is also due to the urgency of market demand. Although the domestic computing power market seems to be oversupplied on the surface, in reality, it is imbalanced in supply and demand, with many enterprises urgently needing to integrate various heterogeneous chip resources to achieve efficient operation.
"Infini-Quest" has introduced a set of AI model computing power formulas, clearly identifying the three influencing factors of computing power: chip computing power, optimization coefficient, and cluster size. By optimizing these factors, they hope to significantly enhance the performance of AI models. The team's goal is to break down the barriers between models and computing power, achieving full-stack technical capabilities in software and hardware.
The company's flagship product, the "Infini-AI Heterogeneous Cloud Platform," is a manifestation of this idea. The platform supports mixed training of multiple chips, providing users with convenient services, making the process of using and developing models simpler and more efficient.
Since its inception, "Infini-Quest" has consistently adhered to the concept of "breaking the boundaries between chips and models," committed to achieving more efficient utilization of computing power through the joint optimization of software and hardware. More exciting is that their goal is to reduce the cost of large-model deployment by 10,000 times, and they have already achieved a 1,000-fold reduction at this stage, with unlimited potential for the future.