These Chinese AI Companies Could Be The Next DeepSeek

Forbes

10-04-2025

Illustration by Fernando Capeto for Forbes; Icons by Ruby/Noun Project
InJanuary, a little-known Chinese AI lab named DeepSeek rocked the world when it released an advanced open source model that rivaled those of US tech giants, seemingly using a fraction of their resources. A who's who of AI's juggernauts, from OpenAI to Anthropic, lauded the company's achievements while defending their own progress and methods. President Donald Trump called it a 'wake-up call.'
But beyond that viral moment, spun up by AI's own frothy hype cycle and a frenzy of geopolitical concerns, DeepSeek's emergence had a deeper impact: It put the spotlight on Chinese AI, and gave it a face on an international stage. DeepSeek and other Chinese companies were not included on the AI 50 list, which honors the most promising privately-held companies in artificial intelligence, because their financials and business practices are opaque. But they are worth highlighting as many of them make significant impacts beyond China, thanks to an emphasis on open source models made freely available for anyone to use.
Forbes
Many of the Chinese AI models gaining traction are made by the country's tech giants. There's video generation app Hunyuan owned by tech conglomerate Tencent, the $92 billion (2024 revenue) maker of WeChat. The company claims its recent 'reasoning' AI models, which can answer complex questions by breaking them into smaller sub-questions, outperforms DeepSeek's flagship models. Then there's Doubao, a consumer-focused app from TikTok parent ByteDance, which has built spatial models that analyze physical environments and generate 3D landscapes. There's also Qwen, a family of large language models from the ecommerce behemoth Alibaba, which has amassed more than 90,000 enterprise users on the company's cloud platform.
'Alibaba is kind of the leading big tech AI champion in China, comparable to what Google or Meta represent in the U.S.,' Rob Toews, a partner at Radical Ventures, told Forbes. As of publication, models from Alibaba and DeepSeek were among the top 5 trending models on Hugging Face, a widely used hub for open source AI models and datasets.
'In the realm of open source releases, there is no such barrier. There is no Great Firewall.'
DeepSeek's meteoric rise opened the floodgates for other Chinese startups. In March, Wuhan-based startup Butterfly Effect launched an AI system called Manus, which it claims can autonomously browse the web and do things like search for apartments, analyze stocks and design websites. The tool has some pitfalls, ranging from making incorrect assumptions about the task at hand to crashing while processing large amounts of text. But its release was praised as an emerging rival to OpenAI's service, Operator. Fueled by the likes of Twitter cofounder Jack Dorsey, buzz around the company has drawn investor interest: Butterfly Effect is reportedly in talks to raise funding from US-based investors at a $500 million valuation, according to The Information.
While DeepSeek built its models by finetuning other models from Meta and Alibaba, Manus managed to use Anthopic's Claude models off the shelf. 'It's basically a much better version of what OpenAI is trying to do with Operator," Toews said.
China has been making significant leaps in humanoid robotics too. Agibot, founded in 2023 by former Huawei 'genius youth' recruit Peng Zhihui, claims it has already manufactured over 1,000 AI-powered bipedal robots and reportedly plans to grow that number to 5,000 robots by the end of the year, in a bid to match Elon Musk's plans for Tesla's general purpose robot Optimus. Earlier this month, the company hired Luo JianLan, who previously worked at Google X, Alphabet's moonshot factory, to lead its research efforts: 'I hear about them when I'm talking with my robotics companies and they're talking about competition they're seeing in China,' Aaron Jacobson, partner at venture capital firm NEA said. 'This one's come up a few times now.'
Prominent investor and AI expert Kai-Fu Lee, who helped Google and Microsoft set up their outposts in China, founded his AI startup called 01.AI in 2022. It recently pivoted from training its own open source models to using DeepSeek's AI to build enterprise apps across areas like gaming, law and finance. The company has raised some $200 million at a $1 billion valuation, according to PitchBook. It's one of China's 'Six Tigers,' an elite group of top AI companies in the country that includes the multimodal AI developer MiniMax AI and model maker Moonshot AI, both of which have attracted investment from Alibaba.
China's advances in AI come as political relations with the U.S. grow ever more fraught. In 2022, then-President Biden issued export controls on semiconductor companies, including crucial chipmakers like Nvidia and AMD, restricting sales of their most powerful hardware in the country, aimed at throttling China's AI growth. Now, a trade war has broken out between the two nations created by President Trump's 'Liberation Day' tariffs, with the U.S. imposing 125% duties on exports.
'This is the power of a centralized state that can say, 'We're going to try to go in that direction.''
China's gains in AI are driven in part by an emphasis on academic research and open source publishing at universities, Russell Wald, executive director of Stanford University's Institute for Human-Centered Artificial Intelligence (HAI), told Forbes. In 2018, China said it wanted to be the leader in AI by 2030, and marshaled significant academic resources toward that goal. Now the country is producing most of the world's AI research. As of 2023 it accounted for about 70% of all granted patents and produced 23% of the world's AI publications and citations. 'This is the power of a centralized state that can say, 'We're going to try to go in that direction.'' Wald said. (He added that the downside, however, is censorship in AI models by the Chinese government might put off western users.)
So far the strategy is working. Earlier this week, HAI released its annual AI Index, which shows the AI race between the U.S. and China narrowing. The U.S. is still producing most of the world's cutting edge AI, with U.S. companies releasing 40 'notable models,' defined as 'particularly influential models within the AI/machine learning ecosystem.' China, in second place, released 15 such models. And the country is closing the gap in model performance: Two years ago, the U.S. led by double digit points on various benchmark tests. Last year, China reached 'near parity,' according to HAI.
The country's open source approach, which lets anyone download the model and build applications with it, has made it easy for Chinese companies to have global impact. For decades, the U.S. and China have had starkly different tech ecosystems, with products and services only available in one region or the other. But DeepSeek's success showed the AI community that Chinese labs could circumvent those restrictions with open publishing. 'In the realm of open source releases, there is no such barrier,' Jeff Boudier, head of product and growth at Hugging Face, told Forbes. 'There is no Great Firewall.'
Meanwhile, given the country's strong cultivation of AI talent at its universities and support for R&D, Wald sees a lot of promise not just in DeepSeek and its current cohort, but what follows them. 'What I think is more interesting is the company you don't know about right now, and you might hear about a year or two from now.'