DeepSeek announced on 22 May 2026 that the 75% discount on its V4-Pro model API, originally intended to expire on 31 May, will not be rolled back. The promotional rate has become the permanent list price. What was a marketing event is now a market floor — and the implications reach far beyond the pricing sheet.
The Numbers That Matter
V4-Pro now permanently costs $0.435 per million cache-miss input tokens and $0.87 per million output tokens. For context, comparable frontier-class reasoning from OpenAI or Anthropic runs at roughly four times that cost. Chinese models already represent 60 per cent of usage on OpenRouter, the popular AI API aggregator — a figure that tells you where cost-sensitive developers are already routing their workloads.
Why This Is Not Just a China Story
The permanent price cut puts direct pressure on every Western AI lab to either reduce pricing on mid-tier models or make a more compelling capability argument for the premium. Analysts estimate that Anthropic, OpenAI, and Google will make quiet pricing adjustments on their mid-range models within 30 days. For UK startups building AI-powered products, the practical effect is a significant reduction in inference costs on agentic and batch workloads — provided data residency requirements do not mandate UK or EU hosting.
The Caveat UK Businesses Must Understand
Data residency is the critical constraint. DeepSeek processes data through servers based in China, which creates real compliance risk for UK businesses handling personal data under UK GDPR, for financial services firms regulated by the FCA, and for any organisation with government contracts. British developers building consumer products or internal tools with no personal data may route freely; those in regulated industries should treat DeepSeek as out of scope for production workloads until a UK or EU-resident hosting option becomes available.
