Apple Chose Tongyi and ERNIE in China, Not DeepSeek — What Does This Tell Us?

Honestly, when I saw this news, I was in the middle of tuning a RAG prompt with DeepSeek.

My phone buzzed with a notification: Apple confirmed partnerships with Alibaba and Baidu in China, integrating Tongyi Qianwen and ERNIE into Apple Intelligence. I paused — wait, what about DeepSeek?

In recent months, DeepSeek has practically become synonymous with domestic large models. Voices in the tech community saying “DeepSeek crushes GPT” have been rising, and it indeed often tops various benchmark leaderboards. Yet when Apple chose partners, they didn’t pick it?

This is quite interesting.

Let me put the conclusion first: Apple not choosing DeepSeek doesn’t mean DeepSeek’s technology is lacking. On the contrary, this might be a landmark event signaling the large model industry’s shift from “technology competition” to “engineering deployment.”

Strong Technology ≠ Deployable

We technologists easily fall into a trap: thinking that as long as the model is strong enough, everything else is secondary. But for a company of Apple’s magnitude, partnership considerations are far more complex than “whose benchmark score is higher.”

First is stability. Apple’s products serve hundreds of millions of users. They can’t have the API working today and down tomorrow. Tongyi and ERNIE are backed by Alibaba and Baidu’s cloud infrastructure, battle-tested during events like Double 11. DeepSeek’s technology is indeed strong, but honestly, it’s still lacking in stability validation at massive concurrency scales.

Second is compliance costs. Deploying large models in China involves filing, auditing, content safety — all hard requirements. Alibaba and Baidu have already navigated the entire process; Apple can essentially “move in with luggage.” If DeepSeek had to start from scratch on these, the time cost is something Apple can’t afford to wait for.

Another point many overlook: ecosystem synergy. Tongyi Qianwen has deep integration with Alibaba’s e-commerce ecosystem and Alipay scenarios; ERNIE is linked with Baidu Search and Maps. Apple wants AI that “can do work,” not just “can chat.” From this perspective, choosing these two makes sense.

So Did DeepSeek Lose?

I don’t think so.

Apple’s choice reflects more of the “current stage” market logic, not the “technical endgame.” Just like how the iPhone didn’t choose the best camera back then, but rather the solution that could best integrate into the system — this didn’t prevent the later explosion of mobile photography.

DeepSeek’s value lies in proving that “Chinese teams can build world-class large models.” This signal itself is more important than winning the Apple deal. It sets a benchmark for the entire industry: the technical ceiling can be broken through.

Moreover, Apple not choosing doesn’t mean other manufacturers won’t. In fact, I understand several phone manufacturers are already in deep partnership talks with DeepSeek — just not as high-profile as Apple.

A Deeper Question

This makes me think of an old question: In the AI era, how should we weigh “technical leadership” versus “commercial success”?

OpenAI’s GPT series is indeed technologically strong, but what really makes money is the ChatGPT product and the operational system behind it. Anthropic’s Claude is better than GPT in some scenarios, but commercialization progress clearly lags.

Technology comes first, but technology alone isn’t enough. Engineering capability, ecosystem integration, compliance experience, business negotiation — these “soft skills” are often more important than model scores during deployment.

My personal feeling is: as a former big tech algorithm engineer, I both hope to see technical players like DeepSeek win, and understand why Apple made this choice. This isn’t a binary choice of “technology vs business,” but rather different priority orderings at different stages.

Final Question for You

If you were the AI lead at a phone manufacturer and had to choose a deep partnership between DeepSeek, Tongyi Qianwen, and ERNIE, which would you pick? The most technically advanced, or the most “plug-and-play”?

Welcome to share your thoughts in the comments.