DeepSeek confirmed today that its V4 models will run entirely on Huawei chips, with no Nvidia hardware required. To understand why that is a big deal, a little context helps. Nvidia makes the chips that power almost every major AI model in the world. The US government introduced export controls specifically to stop China from accessing them, betting that cutting off the hardware would slow Chinese AI development down long enough for American labs to pull ahead.

What appears to have happened instead is that it gave China a deadline.

There is one more detail that landed particularly hard today. DeepSeek did not just build around Nvidia. They shut them out entirely. Typically when a major AI lab releases a new model, chip companies like Nvidia get early access so they can prepare their software in advance. DeepSeek denied Nvidia that window for V4. Only Chinese chip companies, including Huawei, had early access. Nvidia found out about the model the same time everyone else did.

As The Information put it, Nvidia is now getting squeezed from both sides. US export controls limit what it can sell abroad, and Chinese labs are actively steering away from its chips at home. The dominant position Nvidia has held in AI infrastructure is looking less permanent than it did a year ago.

To understand what DeepSeek building on Huawei actually means, it helps to zoom out a little.

In January 2025, a Chinese AI company almost nobody outside the industry had heard of dropped a model that matched GPT-4. It cost a fraction of what American labs were spending to build. It was open source. It was free. Nvidia lost $600 billion in market cap in a single day. That company was DeepSeek. And today's news suggests it was not a fluke, it was a sign of something much larger taking shape.

The US and China are running fundamentally different strategies when it comes to AI. The American approach has been to build the most powerful models possible, fund frontier labs like OpenAI, Anthropic, and Google, and sell access through subscriptions while chasing AGI. The Chinese approach has been to embed AI across existing industries, make consumer AI free, monetize through ecosystem, and deploy at scale across ports, power grids, hospitals, and manufacturing. Less about who has the smartest model, more about who has AI running everywhere first.

Export controls fit neatly into the American strategy. If you believe raw capability is the thing that matters most, keeping the best chips out of Chinese hands makes sense. But if China's strategy was never about having the most powerful model, and was always about building useful AI at scale across their entire economy, then cutting off Nvidia access did not hurt the plan. It just forced Chinese labs to accelerate their own hardware ecosystem faster than they otherwise would have.

DeepSeek V4 running on Huawei chips is the clearest signal yet that acceleration worked. The export controls did not slow China down. They made China less dependent on the US than ever.

Two countries. Two playbooks. One of them just stopped needing the other's hardware entirely.

Anthropic quietly shipped something useful today. Microsoft 365 connectors are now available on every Claude plan, meaning you can connect Outlook, OneDrive, and SharePoint directly to Claude and bring your emails, documents, and files into the conversation.

It is a small update on paper and a meaningful one in practice. Claude already had Google Workspace connectors, but Microsoft 365 is where most of the corporate world actually lives. The gap between asking Claude a question and asking Claude a question with full context of your actual work just got a lot smaller.

Connect yours at claude.ai/customize/connectors.