Today's signal

DeepSeek launched V4 on April 24, a frontier-class open-source AI model optimized to run on Huawei's Ascend chips, with Chinese chipmakers receiving exclusive early access that Nvidia and AMD were denied. It is the first model of this caliber built around a domestic Chinese hardware stack, not an American one.

Why it matters

For the last three years, US export controls on advanced chips were the primary lever Washington used to slow China's AI progress. DeepSeek V4 is the clearest signal yet that the lever is losing its grip. By training and deploying on Huawei's Ascend 950 hardware and pricing V4-Pro at $3.48 per million output tokens against GPT-5.5's $30, DeepSeek has demonstrated that frontier AI is now achievable without Nvidia. Jensen Huang said at GTC in March 2026 that "there's no question we need to have American tech stack in China." His argument: push China out of the American ecosystem and they build their own. DeepSeek V4 is that argument becoming real. The irony is sharp. Huang is being criticized for wanting to sell chips to China, while the consequence of not selling them just showed up at the frontier.

The take

Export controls were always a bet, not a guarantee. The bet was that chip scarcity would cap Chinese AI capability. DeepSeek keeps proving the opposite: scarcity forces efficiency, and efficiency compounds. V4 is not just a model launch. It is the first nail in the assumption that American hardware controls the global AI stack. Washington should be paying close attention, and so should every enterprise currently making long-term AI infrastructure decisions.

The number

That is how much cheaper DeepSeek V4-Pro is per output token compared to OpenAI's GPT-5.5, at $3.48 versus $30 per million tokens. This is a frontier-class model running on Chinese domestic chips, open-sourced for anyone to download and run.

Read the full breakdown at https://analyticsdrift.com/