Something significant is happening in China's artificial intelligence industry and it started with a price tag.
When DeepSeek unveiled its V4 models at aggressively low costs, it didn't just turn heads. It sent shockwaves through an entire ecosystem of developers, cloud providers, and infrastructure companies who suddenly found themselves scrambling to keep up. Weeks later, the aftershocks are still being felt.
The speed at which Chinese AI companies have responded tells you everything about how competitive this market has become.
Xiaomi, better known for its smartphones and electric vehicles, has jumped into the AI race with surprising aggression. The company slashed the API costs for its MiMo-V2.5 model by 99% a move that immediately resonated with developers. Usage exploded almost overnight, with the model processing 1.7 trillion tokens in a single week, representing growth of over 999 per cent from the week before.
That kind of response doesn't happen unless the market was hungry for it.
Meanwhile, AI unicorn MiniMax is taking a different approach with its newly launched flagship model, MiniMax M3. Rather than simply cutting prices across the board, the company is experimenting with a hybrid model pairing token-based billing with monthly subscription plans. It is a strategy that acknowledges a growing reality: not every user wants to pay the same way.
The transition, however, has not been entirely smooth. Some users found that their token consumption ran out far faster than anticipated, exhausting monthly quotas before they expected. MiniMax apologised and moved to protect existing users with unlimited access, but the episode highlighted just how delicate the balance between pricing innovation and user trust can be.
What is emerging in China's AI sector goes beyond a straightforward race to the bottom.
Companies are getting smarter about how they charge for AI, layering in new subscription tiers, restructuring billing systems, and targeting different user groups with more tailored offerings. For enterprise customers running complex, high-volume workloads, the total cost of completing a task matters far more than what a single token costs. Developers working on that insight are starting to rethink their entire pricing architecture.
The shift reflects a deeper tension at the heart of the industry how do you grow your user base rapidly while building a business that can actually sustain itself? There are no easy answers, and the experimentation happening right now is essentially the industry thinking out loud.
The pricing pressure isn't staying neatly contained within AI companies. It is bleeding into the broader technology infrastructure that powers them.
Cloud providers, many of whom built their recent growth strategies around hosting AI workloads, are now being forced to make uncomfortable adjustments. Tencent Cloud, for instance, cut its API prices for DeepSeek's V4 models by as much as 97.5% a dramatic reduction that would have been unthinkable just a year ago.
The dynamic driving this is structural. Because DeepSeek's models are open-weight meaning they can be hosted by multiple vendors rather than controlled by a single provider cloud companies are competing fiercely for the same developers. When anyone can offer the same model, price becomes the primary battleground. Research has found that open models cost, on average, just a fraction of what proprietary closed-source alternatives typically charge, and that gap is only widening as competition intensifies.
Chinese companies have leaned into this advantage harder than almost anyone else. Benchmarking data currently shows Chinese firms occupying the top positions globally when it comes to the most aggressive pricing discounts in AI inference.
Most analysts believe offering AI services cheaply is not just a competitive tactic. It is also a way to attract users, and users generate the one resource that remains genuinely scarce: high-quality training data. Industry experts have pointed out that data availability continues to be one of the most significant constraints on improving AI model performance. In that context, aggressive pricing starts to look less like a sacrifice and more like a long-term investment.
Not everyone is celebrating the falling prices. For companies that invested heavily in cloud computing infrastructure and server capacity to meet AI demand, the economics are shifting in uncomfortable ways.
Shares in Chinese cloud and server-rental companies have softened since the wave of price cuts began, reflecting investor anxiety about what a world of ultra-cheap AI inference means for businesses built around selling computing power. If the cost of running AI drops dramatically and permanently, the demand for outsourced infrastructure may not keep pace with earlier projections.
China's AI pricing war is still very much in motion. Companies are adapting, experimenting, and in some cases stumbling but the direction of travel is clear. AI services are getting cheaper, faster than most anticipated, and the companies that survive this period will be the ones that find a way to build genuine value beyond the price tag.
The race has changed. It is no longer just about who can build the most powerful model. It is about who can build a business around it.
