$ cat deepseek-v4-huawei.mdx

DeepSeek V4 - first frontier AI without a single Nvidia chip

Apr 24, 2026 · #deepseek #huawei #nvidia #china #usa #ascend #chip-wars #sanctions

DeepSeek V4 - first frontier AI without a single Nvidia chip

Chinese company DeepSeek is launching V4 - the first frontier-level AI model in the league of ChatGPT and Claude. Everything runs exclusively on Chinese Huawei Ascend chips. Without a single Nvidia chip.

DeepSeek V4 Huawei chips

Reuters confirmed on April 4, 2026: Chinese company DeepSeek is launching a new AI model called V4. League comparable to GPT-5 and Claude Opus.

But pay attention now.

V4 doesn’t use a SINGLE Nvidia chip

This is the first time in history a frontier-level AI model runs without American hardware. Everything runs on Chinese Huawei Ascend 950PR.

Specs:

  • 1 trillion parameters - on par with GPT-5 and Claude Opus
  • 1 million context - that’s how much text the model holds at once
  • 2x delayed vs. original plan (Chinese New Year, February 2026) - reason: optimization for domestic hardware

Three years of US sanctions - and the Chinese built it themselves

The US has spent three years blocking exports of Nvidia’s latest chips to China. The strategy: don’t let China build competitive AI without our hardware.

Plot twist - they built their own.

What’s more - they gave Nvidia a slap. DeepSeek didn’t give Nvidia early access to V4. Priority went exclusively to Chinese chip makers. It’s a deliberate inversion of standard industry practice.

Reaction in the Chinese chip market

Alibaba, ByteDance, Tencent ordered hundreds of thousands of Huawei Ascend chips. Chip prices jumped 20% in three weeks.

This isn’t a one-off incident. It’s a strategic pivot of the entire Chinese AI ecosystem - from Nvidia to Huawei.

Jensen Huang’s reaction

The Nvidia CEO told Reuters: “a big threat to American dominance.”

Translation - US sanctions were designed for a world from five years ago. In that world, China couldn’t build its own frontier chips. Today they can. Those sanctions don’t work anymore.

The US figured China wouldn’t build AI without their chips. China is saying - thanks, we have our own.

What’s next

Three scenarios:

Scenario 1 - sanctions escalation: The US adds new sanctions (on production technology, on Huawei partners, on talent). China adds new chips. Spiral.

Scenario 2 - two parallel ecosystems: The world splits into “American AI stack” (Nvidia, OpenAI, Anthropic, Apple with Gemini) and “Chinese AI stack” (Huawei, DeepSeek, Alibaba, ByteDance). They live side by side, don’t cooperate.

Scenario 3 - US policy reset: Washington admits sanctions lost their meaning and lets up. Politically unlikely.

Most probable - scenario 2. The AI world splits into two systems.

My take

The worst thing that could happen to America’s chip wars strategy - is the realization that it didn’t work. And now we have public proof.

DeepSeek V4 isn’t an experiment. It’s a product on the market. With real customers (Alibaba, ByteDance, Tencent). With real hardware (Huawei Ascend). With real performance (ChatGPT and Claude tier).

The question isn’t whether the US will react - it’s how. New sanctions are obviously coming. But each new sanction only accelerates the moment China becomes fully self-sufficient. Classic blowback - the harder you block, the harder you force self-sufficiency.

For us watching from Europe - the conclusion is simple. The AI world in 2030 will look different than anyone expected. Not one winner. Two. And in between - the rest of the world picking sides.

Sources

$ cd ../