We need to talk about what happened this past Sunday (May 24th), because it completely shifts the unit economics of AI development.
DeepSeek just announced a permanent 75% price cut on their flagship 1.6-Trillion parameter V4-Pro model API. It now costs roughly $0.83 per million output tokens. Let that sink in. For a model that is natively hitting SWE-bench scores rivaling the top closed-source models, it’s practically free.
Why did this happen? It’s not just “cheap marketing.” It’s a massive hardware shift. Because of US export controls on advanced Nvidia GPUs, DeepSeek optimized their entire infrastructure to run on Huawei’s Ascend 950 supernodes. By bypassing the restricted, overpriced Nvidia hardware monopoly, they dramatically lowered their compute costs and passed the savings directly to developers.
Meanwhile, we are paying a massive premium in the Google and Anthropic ecosystems just to subsidize Big Tech’s reliance on Nvidia.
As a engineering guy why should I pay a “brand tax” for Gemini 3.1 Pro when DeepSeek V4-Pro integrates flawlessly with my custom Opencode setup, gives me a 1M context window, and costs pennies on the dollar?
Are the days of US infrastructure dominance over? Is anyone else migrating their heavy reasoning workloads to V4-Pro this week?