DeepSeek, a Chinese AI startup, said it has made a 75% price cut on its flagship V4‑Pro model permanent, keeping access at roughly a quarter of its original cost and underscoring how hardware and geopolitics are reshaping AI economics; the move ties into Huawei’s Ascend 950 chips and follows coverage involving Abu Sultan in Bengaluru, Eduardo Baptista and Tomasz Janowski. The change affects API pricing and could reshape who can afford high-end models in China and beyond, while raising questions about supply, demand and how export controls influence the market.
DeepSeek’s announcement shifts the pricing landscape for its most powerful model. The V4‑Pro API is now priced between 0.025 and 6 yuan per million tokens, down from a previous range of 0.1 to 24 yuan per million tokens. In dollar terms that translates to about $0.0035 to $0.83 per million tokens, a big cut that makes heavy use scenarios far cheaper and widens access for developers and companies that had balked at earlier rates.
That 75% reduction brings V4‑Pro down to roughly a quarter of its original price, and DeepSeek says the change is permanent. The company did not spell out whether the permanent cut is a direct result of higher availability of Huawei’s Ascend 950 chips, which it used to tune V4’s performance. Still, the timing and technical link to Ascend hardware are hard to ignore when you think about the compute behind large models.
Hardware availability matters because high-end AI depends on dense, powerful compute. Huawei’s Ascend 950 has been a key element for companies like DeepSeek that want to push model performance without relying on foreign GPUs that are harder to ship into China. U.S. export controls have blocked Nvidia’s most advanced chips from the Chinese market, and that has created space for domestic alternatives — even as limits on chipmaking equipment constrain how quickly those alternatives can scale.
DeepSeek originally launched V4 last month and flagged scarcity for the Pro tier. It said the Pro version would cost up to 12 times more than the lighter Flash edition because of “constraints in high-end compute capacity,” limiting broad availability at launch. The company anticipated that Pro pricing would slide once Huawei’s Ascend 950 supernodes are produced in larger quantities during the second half of the year, a timeline that ties pricing directly to supply growth.
The move to lower prices has clear market implications. For startups, researchers and smaller firms, a cheaper Pro tier removes a major budget barrier and could spur fresh experiments and product ideas built on more capable models. For competitors and cloud providers, it signals that pricing pressure may be looming if other firms follow suit or if hardware supply increases faster than demand for premium compute.
There are trade-offs to watch. While cheaper access can democratize advanced AI, it also depends on sustainable compute supply and the ability to maintain performance at scale. Production limits on chipmaking gear mean Huawei and others might still face bottlenecks, and that could create episodic price swings even after an overall downward reset. Investors and customers will be watching whether this price cut sticks through fluctuating hardware availability.
Beyond immediate economics, the DeepSeek decision highlights how technology, trade policy and chip manufacturing intertwine. Export rules shaped which chips are available where, local chip development supplied alternatives, and startup pricing strategies adapted accordingly. Abu Sultan in Bengaluru and Eduardo Baptista reported on the development, with Tomasz Janowski noted in the editorial chain, capturing how these technical and policy forces are playing out in real time.
