PPIO Co-Founder and CEO Yao Xin
AsianFin -- While the daily cost-to-profit ratio of 545% might be a theoretical estimate made by DeepSeek, it doesn't reflect the practical realities of the industry, Yao Xin, the co-Founder and CEO of PPIO, a leading distributed cloud service provider, told AsianFin on Tuesday.
A recent article by DeepSeek said the company has hit a theoretical daily cost-to-profit ratio of 545%, based on pricing across web, app, and API usage. This figure has sent shockwave across the market, but many industry experts remain skeptical.
"If this figure applies to the whole industry, no one would be in this business," Yao noted, adding that such a model is not sustainable in the long run. He pointed out that despite DeepSeek's innovative technology, data shows that the service operates at full capacity for almost 16 hours a day, failing to meet the increasing demand, which he considers "unsatisfactory."
PPIO, founded in 2018, has adopted DeepSeek's model. The company offers distributed cloud computing services at the network edge using a "pay-as-you-go" business model. After DeepSeek's V3/R1 model was released, PPIO was the first to integrate it. According to a third-party evaluation by SuperCLUE, PPIO's DeepSeek R1 integration achieved 100% accuracy in web stability and provided reliable support for clients and developers.
Yao emphasized that PPIO's unique value proposition lies in its "elastic" resource allocation, which allows the system to adjust dynamically based on user demand, thus alleviating the pressure on DeepSeek's official servers. This is particularly relevant considering DeepSeek's recurring service outages.
On March 6, PPIO launched the DeepSeek R1/V3 Turbo version, which triples throughput speed compared to previous models, and is priced as low as 1.6 yuan per million tokens, with a 20% discount for early adopters.
Looking toward the future of AI infrastructure, Yao is confident that the profitability of MaaS (Machine-as-a-Service) companies, like PPIO, doesn't lie in short-term margins but in achieving open-source and standardized solutions. Yao predicts that, within three years, the cost of AI model inference could drop by over 1,000 times.
Distributed computing will play a pivotal role in this reduction as demand for computing power soars. PPIO has already demonstrated this trend by reducing the price of its 8-billion-parameter model by over 50% since May of last year, and the cost of API calls has decreased tenfold in just one year.
In an exclusive interview with AsianFin, Yao expanded on the importance of flexibility in AI infrastructure. He discussed how PPIO's distributed computing approach ensures elasticity and stable service.
During the Spring Festival, when DeepSeek's services were strained, PPIO integrated its system to help ensure DeepSeek's R1 service maintained 99.9% availability. For Yao, the future of the AI infrastructure industry hinges on companies' ability to efficiently allocate idle resources and scale globally.
Yao also shared insights into PPIO's approach to the AI ecosystem. He believes that vertical, fine-tuned models will be the key to a prosperous AI ecosystem, rather than relying on a large number of foundational models. Moreover, he sees significant opportunities for domestic computing power in inference scenarios, especially as the demand for AI-related applications continues to grow.
As the competition in China's AI market intensifies, Yao is upbeat, saying that PPIO's focus on open-source technology, elastic resource allocation, and cost reduction will help it succeed. He remains optimistic about the role of domestic AI companies in reducing inference costs and believes that China could soon enter an era of "free AI applications."
