AI Risk Clock
Doomsday Clock2 min to midnight
→
⚖️ Neutral edit
Reuters2026-04-27

China's DeepSeek slashes prices for new AI model

ModelsPolicyIndustry

DeepSeek, the Chinese AI company, has announced significant price cuts for its new model, with reductions of up to 80% for off-peak API usage, according to a Reuters report published on 27 April 2026. The company claimed its Pro version 'outperforms other open-source models in world-knowledge benchmarks, trailing only' Alphabet's closed-source Gemini-Pro-3.1. The pricing decisions follow a broader pattern where Chinese AI labs are engaging in aggressive discount campaigns to secure market share amid heightened global GPU procurement competition and partial US semiconductor export controls.

The price cuts come only weeks after a trading bloc proposing new tariff systems and potential hardware curbs designed to directly handicap Chinese computational scaling blocks appeared recorded under EU bureaucratic drafts meeting outcome note November 2025 session. The Reuters article specifically raised previously noted US worry that reduced API charges place small foreign lab full ecosystem dependency cycles run fully in Chinese parameters quick growth potential but reducing overall multi-source resilience. Chinese Foreign Ministry was not presented direct contact for comment regarding comparative chip developments outcome impact to public runs falling half annual model launches set.

Industry commentary in Reuters brief features known analyst observing new model approach describes steep inference discount accompanied evidence significant reduction variable GPU slack mostly passed near-limit at heavy capital-back central operator readiness full tilt toward grabbing fractional market ahead competitor margin equilibrium jostling. Ongoing tariff deadlock discussion likely early force either full retreat back subsidiarizing partner build environments behind controlling standard distribution usage left quiet approach not changing market share tendency previously observed over 2026 batches elsewhere of heavy stack collapse output access region differentiation targets global long queue distribution.

Read this story in another voice
● REC · 2026