ZH reported, citing a May 26 report from China Daily.
A Chinese artificial intelligence startup, DeepSeek, has announced that it will permanently maintain steep discounts on its flagship DeepSeek-V4-Pro model, effectively pushing inference costs to some of the lowest levels in the global AI industry.
The move is intensifying competition in the rapidly evolving AI sector, where pricing, efficiency and infrastructure optimization are becoming just as important as model capability.
Permanent Discount Reshapes Pricing Expectations
DeepSeek said the promotional pricing previously offered on its V4-Pro API — originally scheduled to end this week — will now be made permanent.
The company has effectively reduced the model’s official pricing to around one quarter of its original planned rate.
Under the revised structure, input costs for cached requests fall to 0.025 yuan per million tokens, while standard input is priced at 3 yuan per million tokens and output at 6 yuan per million tokens.
Industry observers note that these levels place DeepSeek among the lowest-cost providers of large language model APIs globally.
Pressure Builds Across the AI Industry
The announcement comes at a time when global AI infrastructure costs are rising, driven by heavy investment in data centers, advanced chips and energy-intensive model training.
Against that backdrop, DeepSeek’s aggressive pricing strategy is widely seen as a direct challenge to existing pricing norms in the AI industry.
Rather than competing solely on model size or performance benchmarks, AI companies are increasingly competing on inference efficiency — the cost of actually running models at scale.
This shift is redefining how AI businesses are structured, especially as commercial deployment becomes more important than model development itself.
Efficiency Over Scale: A Structural Shift
Analysts say DeepSeek’s strategy reflects a broader transformation in China’s AI sector.
Instead of relying purely on large-scale computing infrastructure, companies are increasingly focusing on optimizing model architecture, reducing dependency on high-end foreign chips, and improving efficiency per unit of computation.
According to industry commentary, DeepSeek has adapted its models to domestic computing platforms such as Huawei’s Ascend chip ecosystem, helping reduce hardware costs and improve deployment flexibility.
This approach allows companies to lower prices while maintaining competitiveness — a dynamic that is now reshaping the economics of AI services.
A New Cycle of Price Competition
Experts describe the current phase as a transition from capability competition to cost competition.
Wang Peng, a researcher at the Beijing Academy of Social Sciences, said DeepSeek’s pricing move should not be viewed as simple subsidy-driven price cutting.
Instead, he argued it reflects underlying architectural improvements and cost advantages built into the system itself.
According to this view, lower pricing can help attract users, expand ecosystem adoption, and further reduce marginal costs — creating a reinforcing cycle of scale and efficiency.
China’s AI Market Is Moving Toward “Efficiency Wars”
The DeepSeek case highlights a broader trend in China’s artificial intelligence industry: competition is increasingly shifting toward deployment efficiency rather than raw model performance.
As AI adoption expands across industries such as manufacturing, finance, education and consumer services, the cost of inference — running AI systems at scale — is becoming a decisive factor in commercial viability.
Companies that can reduce these costs gain a significant advantage in scaling real-world applications.
This is particularly important in China’s AI ecosystem, where integration with the real economy and large-scale deployment is a key strategic priority.
Implications for the Global AI Industry
The global AI sector is now entering a new competitive phase where pricing pressure could reshape business models across the industry.
If ultra-low-cost inference becomes sustainable, it could accelerate AI adoption across emerging markets and lower the barrier to entry for smaller companies and developers.
At the same time, it raises questions about long-term profitability and the sustainability of aggressive pricing strategies in an industry still heavily reliant on infrastructure investment.
Beyond Price: A Structural Competition
Ultimately, DeepSeek’s move reflects a deeper shift in how AI competition is unfolding.
The battle is no longer just about who builds the most powerful model.
It is increasingly about who can deliver intelligence at the lowest cost, at the largest scale, and with the most efficient infrastructure.
In that sense, the global AI race is entering a new phase — one defined not only by capability, but by economics.
And in this emerging landscape, pricing itself has become a strategic weapon.