Chinese artificial intelligence startup DeepSeek announced on May 23 that it is permanently reducing the price of its flagship V4-Pro model by 75%, a move that has sent ripples through China’s AI industry and drawn immediate attention from analysts tracking the country’s domestic chip supply chain. The price cut, confirmed by Reuters, makes V4-Pro one of the most cost-effective frontier AI models available to Chinese enterprise customers and signals that DeepSeek has resolved a critical bottleneck in its inference infrastructure.
The Economics of the Price Cut
The 75% reduction is not a promotional discount. DeepSeek has confirmed it is permanent, reflecting a structural change in the company’s cost base rather than a short-term marketing move. For enterprise customers, this dramatically alters the return-on-investment calculus for deploying V4-Pro at scale. Industries such as financial services, healthcare, and logistics, where AI inference costs have previously been a barrier to wide adoption, will now find the model accessible at a fraction of its earlier price.
The timing is significant. DeepSeek launched V4-Pro in late April 2026 after a period of supply-related delays, as EastFrontier reported when the model was first released. At launch, the model’s inference pricing was competitive but not dramatically cheaper than alternatives. The permanent 75% cut suggests that in the weeks since launch, DeepSeek has secured a substantially larger and more cost-effective supply of compute, and the most likely explanation, according to industry analysts, is a breakthrough in the availability of Huawei’s Ascend 950 AI accelerators.
The Huawei Ascend 950 Connection
The Ascend 950 is Huawei’s most advanced AI chip, designed to serve as a domestic substitute for Nvidia’s H100 and H200 GPUs. DeepSeek has publicly stated that V4 and V4-Pro were trained exclusively on Huawei Ascend hardware, making the company uniquely dependent on — and uniquely positioned to benefit from, improvements in Ascend chip supply and manufacturing yield.
If Huawei has achieved a meaningful ramp in Ascend 950 production, the cost per inference token would fall significantly, enabling DeepSeek to pass those savings on to customers while maintaining or improving its margins. This interpretation is supported by the scale of the price cut: a 75% reduction implies a very substantial improvement in underlying compute economics, not merely incremental efficiency gains from software optimization.
The implications extend well beyond DeepSeek. As EastFrontier has documented, Huawei’s Ascend ecosystem has been gaining momentum across China’s AI industry, with demand for the 910C already running approximately 20% above supply earlier this year. A supply breakthrough at the Ascend 950 level would represent a qualitative step forward for China’s entire domestic AI hardware stack, not just for DeepSeek.
Cascading Effects on China’s AI Market
The permanent price cut is expected to trigger a competitive response from other major Chinese AI model providers. Baidu’s ERNIE series, Alibaba’s Qwen family, Tencent’s Hunyuan models, and ByteDance’s Doubao are all priced in a range that V4-Pro’s new pricing undercuts significantly. These companies will face pressure to reduce their own inference prices or risk losing enterprise customers to DeepSeek.
This dynamic is part of a broader pattern in China’s AI market, where intense competition among well-funded players has driven prices down rapidly across the model stack. The price war benefits Chinese enterprises and developers, who gain access to frontier-level AI capabilities at costs that are increasingly difficult for Western providers to match. For global enterprise customers evaluating Chinese versus Western AI models, the cost differential is becoming a serious factor in procurement decisions.
Open Source and the Global Competitive Landscape
DeepSeek’s pricing strategy also intersects with its open-source commitments. The company has released several of its models under open-source licenses, contributing to a global developer ecosystem that now spans well beyond China. By making V4-Pro dramatically cheaper to access via API, DeepSeek is accelerating the adoption of its technology among developers who might otherwise default to OpenAI or Anthropic offerings.
The permanent nature of the price cut also sends a signal to the global AI market: DeepSeek is not competing on a temporary promotional basis but is building a sustainable cost advantage grounded in domestic hardware. As U.S. export controls continue to restrict Chinese access to Nvidia’s most advanced chips, the ability to achieve competitive inference costs using Huawei hardware is a strategic milestone for China’s AI industry, one that the V4-Pro price cut makes visible for the first time at this scale.
The broader context for this development is DeepSeek’s remarkable trajectory over the past twelve months. The company burst onto the global AI scene in early 2025 with the release of its R1 reasoning model, which demonstrated that frontier-level performance could be achieved at a fraction of the training cost of Western competitors. Since then, DeepSeek has maintained a relentless pace of model releases and infrastructure improvements, culminating in V4-Pro, a model that now combines frontier performance with what is arguably the most competitive inference pricing in the global market. The permanent 75% price cut is not the end of this story but the beginning of a new chapter: one in which DeepSeek competes not just on technical merit but on the structural economics of AI deployment at scale. For Chinese enterprises, developers, and the broader global AI ecosystem, the implications of that shift are profound and will take months to fully play out.