DeepSeek makes 75% V4-Pro price cut permanent on Huawei Ascend supply

TL;DR:

• Chinese AI startup DeepSeek will permanently lock in a 75% price cut on its flagship V4-Pro model, with API pricing now between 0.025 and 6 yuan per million tokens (about $0.0035 to $0.83) depending on usage type, down from 0.1–24 yuan previously.
• The company has not directly confirmed the reason but said in May that V4-Pro pricing was expected to fall sharply once Huawei’s Ascend 950 supernodes began shipping in volume in the second half of the year.
• V4-Pro pricing originally ran up to 12 times more expensive than the lighter V4 Flash model due to what DeepSeek described as “constraints in high-end compute capacity” – a constraint the price cut implies is now easing.

The Reuters piece is short but carries an outsized read. DeepSeek’s pricing is a leading indicator of how quickly Chinese compute supply – specifically Huawei’s Ascend 950 line – is closing the gap on the Nvidia-shaped market that US export controls have walled off. A permanent 75% drop is the kind of move only made when the underlying supply position is judged solid for the next price-elastic window.

Why the price cut matters beyond China

V4-Pro at a quarter of original API pricing puts a credible frontier-class model into the same cost range as US frontier models’ lighter tiers. For UK enterprise AI buyers building cost models, this is the second permanent floor-shift in 12 months (after the original V4 family launch). Token costs of $0.0035 at the low end and $0.83 at the high end give a wide envelope: high-volume use cases like document processing, customer-support summarisation and code generation that were borderline on cost at $0.10–1.00 per million tokens become structurally viable.

UK security and data-residency questions remain: DeepSeek’s hosting and training data flows mean direct enterprise adoption is constrained in regulated sectors, and AISI evaluation work has not (publicly) covered V4-Pro to the same depth as Claude or Gemini. But the API economics now anchor a competing reference price.

What the cut signals about chip supply

Reuters notes Huawei AI chip sales have benefited from US export controls that prevent Nvidia from selling top-tier semiconductors in China, while separate US controls on chipmaking equipment have constrained Huawei’s ability to scale Ascend production. DeepSeek’s confidence in making the cut permanent implies the Ascend 950 supernode pipeline is now meaningful enough to backstop frontier-tier inference at scale. If accurate, that materially shifts the strategic picture for Chinese AI cost structure going into 2027.

Looking forward

Expect Western frontier labs to respond either with corresponding pricing moves at the mid-tier (Anthropic, Google, OpenAI) or with capability tiering – charging frontier capability at frontier prices while pushing low-tier tasks to cheaper models. For UK SMEs running multi-model architectures, the practical move is to refresh cost models on multi-model routing now: V4-Pro at the new pricing rewrites several decision boundaries on which model handles which class of workload.