A friend who works in tech in Hangzhou messaged me last night. "Guess what," he said. "Alibaba just sent out an internal memo. No more Claude Code. Not for any project. Not even for personal learning on company machines."
I asked if this was a security thing. He laughed. "They said it's a data compliance issue. But come on. You know what this is really about."
The Memo
The internal notice, which went out to Alibaba's cloud and AI divisions earlier this week, was pretty blunt. Employees were told to stop using Anthropic's Claude Code — the AI coding assistant that's been gaining serious traction among developers — effective immediately. The stated reason: concerns about source code being processed on US-based servers, potential violations of China's data security laws, and the ever-present specter of "regulatory compliance risk."
A few weeks earlier, a similar directive had quietly circulated at Tencent. ByteDance, interestingly, hasn't issued a blanket ban — but they've been pushing their internal Doubao and a new coding assistant built on their own models pretty aggressively. The message is clear without being explicit: use our stuff. Not theirs.
This isn't coming out of nowhere. The US has been tightening AI-related export controls for three years now, restricting access to advanced chips, limiting what models can be exported, and generally treating AI technology as a strategic asset rather than a commercial product. The Claude Code ban is, in many ways, the mirror image of that policy — China telling its tech giants: American AI tools are not welcome here.
What Developers Are Actually Saying
The official line is about data security. The unofficial reaction from developers I've spoken to is... mixed.
One engineer at Alibaba Cloud told me he'd been using Claude Code for about four months. "Honestly, it was better than anything we had internally. Faster. Smarter about context. Our own tools feel like they're two generations behind." He's now been told to switch to Tongyi Lingma, Alibaba's in-house AI coding assistant. "It's fine for boilerplate. For anything complex, I'm spending twice as long as I was with Claude."
Another developer was more philosophical. "Look, I get it. The company can't have our entire codebase flowing through servers in Virginia. That's not paranoia — that's just how geopolitics works now. But it still sucks."
Some teams are reportedly exploring workarounds — running open-weight models locally, using DeepSeek's API (which is hosted in China and doesn't raise the same compliance flags), or, in a few cases, just using Claude Code on personal devices and manually porting over the useful output. None of these are great solutions, but developers are resourceful.
The practical impact is going to be uneven. Junior developers who rely heavily on AI coding assistants will feel the loss most acutely. Senior engineers who use these tools for the boring 30% of their work — boilerplate generation, test writing, documentation — will be annoyed but mostly fine. The productivity hit is real, though, and it'll show up in project timelines over the next few quarters.
The Bigger Picture
This isn't really about Claude Code. It's about what happens when AI becomes a strategic industry and the two countries that dominate it stop pretending the market is open.
Think about the trajectory. In 2023, the US restricted exports of advanced AI chips to China. In 2024, the restrictions expanded to cover more hardware and some model weights. In 2025, there were murmurs about restricting access to US-based AI APIs. Now, in mid-2026, the restrictions are flowing in both directions — China telling its companies not to use American AI tools, the US telling its companies not to use Chinese ones.
What we're watching is the slow decoupling of two AI ecosystems. On one side: OpenAI, Anthropic, Google, Microsoft, NVIDIA. On the other: DeepSeek, Alibaba's Tongyi, ByteDance's Doubao, Baidu's ERNIE, Huawei's Ascend chips. The tools are broadly similar. The infrastructure underneath them is diverging fast.
For developers caught in the middle — people who just want to write good code and go home — none of this feels great. A tool that made you 40% faster yesterday is suddenly off-limits, not because it stopped working, but because of a geopolitical chess game you didn't sign up for.
The Alternatives Aren't Terrible, Though
If you're an Alibaba developer who just lost Claude Code access, what do you actually switch to?
The most obvious answer is DeepSeek's API. It's Chinese-hosted, which means no data compliance issues, and the latest DeepSeek-V3 model handles code generation about as well as Claude 3.5 Sonnet on most benchmarks. The API pricing is a fraction of what Anthropic charges — we're talking maybe $0.14 per million input tokens versus Anthropic's $3. Something like a 20x difference. For a company the size of Alibaba running thousands of developer seats, that math adds up fast.
Then there's Tongyi Lingma, Alibaba's own coding assistant. I've tested it briefly through a colleague's account, and it's... fine. Solid on Chinese-language documentation and comments, which actually matters more than you'd think if your entire team communicates in Mandarin. It handles standard CRUD operations and API wiring without problems. Where it stumbles — and where Claude Code really shined — is on complex architectural reasoning and debugging subtle issues across multiple files. But Alibaba has every incentive to close that gap, and they will.
Huawei's PanGu models are also in the mix, along with Baidu's Comate. The competitive landscape for Chinese AI coding tools in mid-2026 actually looks a lot like the US landscape did in early 2024 — multiple players, fast improvement, no clear winner yet. The Alibaba ban might be the thing that forces one of them to get really good, really fast.
What Actually Happens Next
I think we'll see three things play out over the next year.
First, Chinese AI coding tools are going to get better, fast. DeepSeek has already shown it can compete with Western models on technical benchmarks. The Claude Code ban creates a captive audience of millions of developers who now have to use domestic alternatives — and their feedback, frustration, and feature requests will accelerate those tools' improvement. Necessity is a hell of a motivator.
Second, this will accelerate the open-weight model movement in China. If companies can't use US-hosted APIs, they'll gravitate toward models they can run on their own infrastructure. DeepSeek, Qwen, and Yi all offer open-weight versions. This might actually be the healthier long-term outcome for Chinese AI development — less dependence on foreign APIs, more investment in local infrastructure.
Third, and most frustratingly, we're going to see more of these bans, not fewer. The AI decoupling hasn't peaked yet. If anything, Claude Code getting banned at Alibaba is an early indicator of where things are heading. Every major tech company in both countries is going to have to pick sides eventually, whether they want to or not.
I asked my friend in Hangzhou what he's going to do. He shrugged. "I'll use whatever they tell me to use. But I'm keeping my personal Claude subscription. For my weekend projects, obviously."
Obviously.
Written by James Whitfield. Based on reporting from internal sources at Alibaba and broader industry analysis of US-China AI policy developments.
