Alibaba Says It Has an Answer to DeepSeek: Superior AI

It’s the Lunar New Year today, so many markets are closed in Asia. But the ramifications of DeepSeek’s advances in Artificial Intelligence continue to dominate the discourse.

Alibaba Group Holding BABA (HK:9988) shares are up in premarket U.S. trading as developers claim strong performance for a new version of the group’s Artificial Intelligence model, Qwen.

The shares gained 6.7% by Tuesday’s close and look set to add an additional 2.7% at the open today.

Alibaba says in a blog post that its Qwen2.5-Max system outperforms Deepseek-V3, as well as OpenAI’s GPT-4o and the Llama-3.1 system from Meta Platforms META.

Hong Kong’s market is shuttered today and for the rest of the week, having closed after a half day’s trade on Tuesday. We’re ushering in the Year of the Snake, which feng shui experts tell us is a time to rely on intuition, charm and transformation. That’s compared with the auspicious Year of the Dragon we are exiting, a time dominated by power, honor and success.

There’s been no trading on Wednesday in mainland China, Hong Kong and Taiwan, as well as Indonesia, Malaysia, Mongolia, the Philippines, Singapore, South Korea and Vietnam, all closed in honor of the Lunar New Year. Indonesia and the Philippines resume trade tomorrow, but the markets in greater China remain closed the rest of this week.

So we will have the unusual situation that Alibaba shares will be trading on Wall Street while walled off in Hong Kong, where they gained 1.2% on the half day for Tuesday, before the company’s announcement. Hong Kong and China stocks have not reflected the selloff on Wall Street, with the region surprising tech analysts with the advances in Artificial Intelligence achieved apparently with the use of inferior materials.

Will we see any reversal of the trend that has seen U.S. stocks shoot higher, a time of “U.S. exceptionalism,” when the S&P 500 and U.S. economy could do no apparent wrong? I said at the start of this calendar year that the one-way predictions of bull U.S. markets concern me.

We have received a sudden jolt of surprise in the form of DeepSeek’s claims of achieving results to rival or outdo the likes of OpenAI at a fraction of the cost. But here in China, there’s also skepticism about its claims.

I’d note that DeepSeek still says it is relying on chips from Nvidia NVDA, just not the top-flight ones that sell for $30,000 to $40,000.

It is also possible that DeepSeek has got its hands on higher-end Nvidia chips than U.S. export controls allow. The U.S. Department of Commerce last month reportedly asked the chip designer and its distributors to investigate how its chips have been making their way into China. The Information reports that smugglers have used tactics such as duplicating server serial numbers to arrange “grey market” shipments into China of chips that are under U.S. export controls.

Still, there’s no doubt that China churns out engineers and computer scientists at an unprecedented clip. Many have struggled to find jobs, resulting in cheap labor costs for the likes of Hangzhou-based DeepSeek. The company published the results of its latest tests of DeepSeek 3.0 on its home page, claiming performance to rival Qwen2.5, Llama 3.1 and GPT-4o.

Now we have the Alibaba Cloud team pointing to performance that outdoes DeepSeek-V3, as well as the rivals from OpenAI and Meta.

It goes to show that DeepSeek is at least not alone in developing AI models that match those in the West. 

OpenAI is claiming it has evidence that DeepSeek has been using the American company’s AI model to train its system, according to Bloomberg. This is really rich, considering that OpenAI and Microsoft MSFT are being sued first by The New York Times and then by eight other newspapers as well as numerous authors claiming that the OpenAI system has been trained on their news articles.

U.S. investors are able to play the Alibaba claims today. Be forewarned that we won’t have full price disclosure until next week when Asian trade resumes. This “war” over Artificial Intelligence looks set only to intensify. 

At the time of publication, Alex Frew McMillan had no position in any security mentioned.