Xiaomi has dramatically re-entered the AI price wars, slashing its MiMo-V2.5 series API prices by up to 99% on May 27, 2026.
This move is a direct response to a rival's gambit. Just days earlier, another major Chinese AI firm, DeepSeek, made its own 75% price cut permanent, effectively setting a new, rock-bottom price floor for the entire industry. Xiaomi’s decision to match this price signals a clear strategic choice: prioritize user acquisition and market share over immediate profitability. The timing was also critical, as Xiaomi's recent Q1 financial results showed declining profits, adding pressure to stimulate growth in new areas like its AI services.
This aggressive pricing strategy reveals a major split within China's AI landscape. First, there's the 'scale-first' camp, now led by DeepSeek and Xiaomi, who believe that attracting a massive developer base with ultra-low costs is the key to long-term success. Second, there's the 'monetization-first' camp, including firms like Zhipu AI and ByteDance, which are moving in the opposite direction. They have been raising API prices and introducing paid subscription tiers, focusing on generating revenue and proving the business viability of their models now.
But how can Xiaomi and DeepSeek afford such low prices? The answer lies in the hardware. The critical enabler is a strategic shift towards domestic computing infrastructure, specifically Huawei's Ascend AI chips. By building their systems on this local hardware instead of relying on more expensive and supply-constrained Nvidia GPUs, these companies have significantly lowered their inference costs—the cost of running the AI model for users. This hardware advantage gives them the economic foundation to sustain a prolonged price war.
This domestic price war also has global implications. While Chinese firms are racing to the bottom, major Western players like OpenAI have been increasing prices for their newest models, such as GPT-5.5. This creates a stark price contrast. Xiaomi's new pricing makes its models over 100 times cheaper than GPT-5.5 on certain tasks. This huge gap could incentivize developers worldwide, especially those working on price-sensitive applications, to route their AI traffic to these Chinese providers, fundamentally reshaping the global AI platform market.
- API (Application Programming Interface): A set of rules and tools that allows different software applications to communicate with each other. In this context, it's how developers access and use AI models like Xiaomi's.
- Inference Cost: The computational cost incurred when an AI model is used to make a prediction or generate a response after it has been trained. Lower inference costs are key to offering cheaper services.
- Token: The basic units of text or code that an AI model processes. Pricing is often based on the number of tokens in the input and output.