Written by 
Another large model startup has temporarily set aside its ambition for a consumer Super App and shifted its focus to the seemingly more profitable B-side track.
Following Baichuan Intelligence’s repeated emphasis on focusing on the medical scene, on November 6th, 01.AI made a high-profile disclosure of its ToB (To Business) solution based on the Yi model. In mid-October, this company had already launched its first industry application product AI 2.0 Digital Human targeting retail and e-commerce scenarios while releasing its latest flagship model Yi-Lightning.
This press conference can be seen as a systematic review of 01.AI’s ToB strategy.
This system includes three business lines, namely the newly launched “RuYi” digital human solution for e-commerce live streaming, office meetings and other scenarios, the AI Infra solution refined based on its own AI Infra capabilities, and the existing Yi API and open model training platform. 01.AI summarizes it as a “Infra + Large Model + Application” integrated strategy.
In an industry that increasingly values companies’ commercialization levels, 01.AI directly invited several heavyweight enterprise clients to the launch event. For the “Six Small Tigers” (Zhipu AI, Baichuan Intelligence, Moonshot AI, Minimax, 01.AI, StepFun) who are at the forefront of trends and facing commercialization and valuation doubts, this is a necessary endorsement.
The main customers of the “RuYi” digital human solution include Yum China, KIDSWANT, Turing Intelligent Computing, Letao Entertainment, Zhiketong etc. According to the company’s introduction,this business currently covers local life live broadcasting,AI companionship ,IP image ,office meetings ,media marketing,and other scenarios.It can be generalized to finance,customer service,traning,and many ToB even ToC (consumer-level) scenes in future.
The company’s AI Infra (AI infrastructure) capabilities are also being used for commercial purposes. 01.AI is one of the enterprises in the industry that pays special attention to model-based co-construction. In today’s environment where model performance improvement is gradually slowing down, relying on AI Infra to enhance training and inference efficiency, reduce computing costs, its importance is increasingly prominent.
The company believes that the construction and operation of large-scale AI clusters currently face a series of challenges. In terms of power supply, liquid cooling technology, computing capacity, network connectivity, storage solutions, scheduling systems as well as fault monitoring and positioning all require a significant amount of effort and resources investment. Therefore, 01.AI refines its own AI Infra capabilities into complete solutions and collaborates with governments and enterprises to build large-scale model computing power and service platforms.
In fact, the reason why the business models of leading companies in the AI 1.0 era are controversial mainly lies in the “non-standardized nature” heavy customization emphasis on delivery” characteristics of ToB businesses.
Regarding this issue, Kai-Fu Lee, founder & CEO of 01.AI stated that AI 2.0 represented by large-scale model technology is reshaping productivity patterns across various industries but only by entering core business systems deploying ToB applications quickly lightweightly at scale can maximize cost reduction increase efficiency for enterprises.
Ken Qi, co-founder of 01.AI added that from Day 1 onwards the company has been exploring the boundaries capabilities big models have in various scenarios avoiding business problems faced during AI 1.0 era focusing on leveraging strong generalization ability big models finding deep enough application scenarios while considering inference costs scaling under premise enterprise customer ROI positive.
However, stepping out of the ToB perspective, large-scale entrepreneurial companies have a more macro commercial issue to address.
SEE ALSO: 01.AI Releases New Flagship Model Yi-Lightning
In China, it may correspond to a much smaller application ecosystem. How many AI companies can it support to survive in the end?
Kai-Fu Lee’s conclusion is that China can have as many companies as the United States, but with different strategies. Although the financing is not up to 10% of OpenAI’s, the training cost of 01.AI accounts for 3%, and the inference cost accounts for 40%. In addition, “none of the top 10 large models in America will land in China,” while Chinese companies not only have opportunities to land in domestic markets but also go global. In this competitive environment, Kai-Fu Lee believes that Chinese companies even have greater opportunities in ToC.
Looking at the next step for large model startups from an intuitive perspective, how to balance between their own business income and valuation may be an important breakthrough point for sustainable operation under healthy conditions.
An investor in the field of AI large models told reporters that American AI company valuations are generally around 20-30 times ARR (Annual Recurring Revenue). However, whether it is American or Chinese leading AI companies, this ratio has actually exceeded 40 times.
Taking OpenAI as an example, according to The Information report, its latest ARR may be around $3.4 billion (different institutions and media have different views on the specific business composition of this ARR), but its valuation has reached $157 billion, which is about 46 times higher than its former value.
At this point, can 01.AI’s ToB strategy help accelerate its growth in ARR and become a ‘unicorn’ capable of withstanding revenue doubts?
Ken Qi told reporters that in fact, the judgment of ARR also needs to be divided by business type, which can generally be divided into ToB, ToC, and Professional Consumer modes. If we only look at API revenue, domestic companies are far behind OpenAI, but in the overseas scenario of the Professional Consumer mode, if it can achieve a Product-led state (product-driven growth), it is possible to reach a level 20-30 times higher.
“For ToB, I think it’s unlikely to reach 40 times; even reaching 20-30 times is not very likely,” Ken Qi pointed out. This involves many factors such as the level of domestic customer unit price and still relatively high overall customization demand.
From the perspective of 01.AI, Ken Qi believes that if we single out digital human businesses, achieving a level 20-30 times higher is highly possible. However, from the current multi-line parallel systematic business composition of the company’s perspective, a clear calculation logic has not yet been formed.