DeepSeek V4 Preview Drops, and This Time It’s Running on Chinese Chips

DeepSeek is back with a preview of V4, its next-generation open-source model, and the timing is hard to ignore. It’s been roughly a year since the company shook up the AI world with V3 and R1, and now they’re taking another swing at the US incumbents.

The headline claim is that V4 can go toe-to-toe with closed-source systems from Anthropic, Google, and OpenAI. That’s a bold statement, but given how much ground DeepSeek covered in the last year, it’s not entirely dismissible. The company says the biggest improvements are in coding — which makes sense, since that’s where AI agents are making their real money right now. Think ChatGPT Codex, <a href="https://design.allwinchina.org/ai-tools/claude-code/" title="Claude Code review”>Claude Code, all that jazz.

But the part that actually caught my attention isn’t the benchmark numbers. It’s the hardware. DeepSeek explicitly highlights compatibility with domestic Huawei technology. That’s a big deal. For anyone who’s been watching the chip sanctions saga, this is the kind of milestone that actually matters. China’s AI industry has been scrambling to reduce dependence on Nvidia hardware, and DeepSeek just showed that you can build a competitive model without relying on the latest Blackwell chips.

Of course, this is just a preview. We don’t have full benchmark results yet, and the real test will be when the final model drops and people start running it on actual workloads. But the direction is clear: open-source models are catching up, and they’re doing it on Chinese silicon.

I’m not going to pretend this means the US is losing the AI race overnight. But if DeepSeek V4 delivers on its coding promises, and runs efficiently on Huawei hardware, then the conversation shifts from “can China compete?” to “how fast can they scale?”