From budget-friendly breakthroughs to mind-blowing models, China’s AI underdogs are redefining the game finds Satyen K. Bordoloi
In Liu Cixin’s Three-Body Problem series, aliens sabotage Earth’s technological progress to keep humanity in check. The US, in a somewhat less extraterrestrial but equally dramatic move, tried to do the same to China: slapping export bans on advanced chips and blacklisting companies. And let’s be honest, if there’s one thing both Biden and Trump agree on, it’s their shared love for hating China.
But here’s the twist: instead of stifling China’s AI growth, these restrictions have only fuelled its rise. As I wrote in a previous Sify piece, China’s AI industry is not just catching up to the US; it’s sprinting ahead and doing so with a flair that’s both impressive and, dare I say, cheeky.
Taking the discussion ahead from that article, here are some Chinese AI models who rolled up their sleeves and swore: “We’ll show you how it’s done.”
The Rise of DeepSeek-V3: Doing More with Less
DeepSeek is a large language model (LLM), much like ChatGPT. Except it is open source, much like Meta’s Llama. Its previous models were already making waves. But then came DeepSeek-V3, strutting onto the scene like it owned the place. This LLM didn’t just perform well on global benchmarks—it did so on what Andrej Karpathy, a founding member of OpenAI, called a “joke of a budget.”
DeepSeek is the ultimate budget-savvy overachiever. With just over 2,000 Nvidia H800 GPUs—costing a cool $40 to $50 million—and a mere $5 million spent on training, this scrappy underdog is going toe-to-toe with ChatGPT’s latest iteration, which reportedly burned through billions of dollars. And here’s the kicker: DeepSeek isn’t just keeping up; it’s often every US LLM on most benchmarks. Let that sink in. If $55 million can do what billions used to, the AI industry is about to flip faster than an omelette at a roadside breakfast joint.
No wonder Nvidia’s Senior Research Manager, Jim Fan, commenting about DeepSeek in a post on X on 27 December, summed up perfectly: “Resource constraints are a beautiful thing.” And boy, is he right. DeepSeek-V3 is proof that necessity isn’t just the mother of invention; it’s the mother of efficient invention. By focusing on cost-effective training methods and leveraging China’s vast data resources, DeepSeek has created a model that rivals GPT-4 without breaking the bank.
DeepSeek’s models are designed to work in multiple languages, making them a global contender.

Kling: Where Text Meets Video Magic
If DeepSeek is the brainiac of the group, Kling from Kuaishou Technology is the artist. This Chinese company has been making waves in the text-to-video (TTV) space for half a year, turning written descriptions into high-quality video content. Imagine typing “a dragon kissing a rainbow unicorn at sunset” and getting a cinematic masterpiece in seconds. That’s Kling for you.
While OpenAI announced its TTV model Sora in February 2024, it launched it only six months after Kling in December. Despite the delay, users quickly pointed out that Kling is far ahead of Sora in terms of video quality and distortions. And best of all, Kling’s subscription is cheaper than most TTVs.
Kling is a blessing for content creators, advertisers, and educators. Filmmakers and YouTubers can now visualise scripts and concepts at the speed of typing. Marketers can generate dynamic, narrative-driven video ads tailored to specific campaigns in a jiffy. Educators can also easily use TTV to turn lessons into engaging visual content.
Kling’s secret sauce lies in integrating natural language processing (NLP), computer vision, and generative adversarial networks (GANs). It’s like having a Hollywood production team on your laptop, ready to bring your wildest ideas to life.
Vidu 2.0: The Future of Video Generation
If DeepSeek is the budget genius of Chinese AI, then Shengshu Technology’s Vidu 2.0 is the speed demon of video generation. Launched as the latest iteration of its video creation platform, Vidu 2.0 makes high-quality video production faster, cheaper, and easier than ever. How? By combining a universal vision transformer (U-ViT) model with a proprietary interference accelerator, Shengshu has slashed costs by more than half compared to industry averages.
But it’s not just about speed and savings. Vidu 2.0 is packed with features that make video creation a breeze. Forget writing tedious prompts; their templates let users add detailed actions, props, or settings with just a few clicks. The “reference-to-video” feature ensures consistency in subjects and styles, tackling the fundamental problem with AI video generation – inconsistency. And with plans to expand its template library, Vidu 2.0 is only getting better.
Shengshu, surprisingly, did it in under two years. Founded in 2023, it has quickly gone from research to revolutionising the creative industry. Alongside Vidu, they’ve also developed VoxCraft, a 3D asset generation platform. But Vidu 2.0 is their crown jewel, bringing us closer to a future where AI-driven storytelling adapts to every creator’s vision. It’s not just a tool—it’s a creative revolution in a box.

The Supporting Cast: Inspur, Zhipu AI, ByteDance, and MiniMax
While DeepSeek, Kling, and Vidu 2.0 are hogging the limelight, let’s not forget the stellar supporting cast of companies like Inspur, Zhipu AI, ByteDance, and MiniMax, who are quietly (or not so quietly) making waves with their own AI innovations.
Take Zhipu AI, for instance. This start-up recently found itself on the US Department of Commerce’s Entity List—a dubious honour that restricts it from buying US tech without special approval. The Biden administration cited “national security concerns,” but let’s be honest: it’s also a backhanded compliment to Zhipu’s growing prowess. Being banned by the US is practically a rite of passage for Chinese tech companies these days.
Then there’s MiniMax, whose Hailuo AI (of the viral AI kissing videos fame), and ByteDance’s Dreamina, are giving Kling and Vidu 2.0 a run for their money in the text-to-video (TTV) space. These platforms are innovating, and by offering multilingual capabilities beyond Chinese, including English and other languages, they indicate their global ambitions. Why dominate a single market when you can conquer the world?

Why China’s AI Boom is No Fluke
Together, these, and other companies prove that China’s AI ecosystem isn’t a one-hit wonder but a new tech symphony where each player brings unique notes. As I wrote in the previous article for Sify, they result from a careful orchestration. With nearly a billion and a half people, a digital footprint the size of a small planet and lax regulation for data scraping, China has no shortage of data to train its models. Add significant government support, from funding to favourable policies, and you have a recipe for success.
Top universities and tech companies are joining forces to accelerate innovation while cutting-edge computational facilities and AI supercomputers provide the backbone of this ecosystem. It’s a perfect storm of resources, talent, and ambition—and it’s paying off in spades. India, are you listening?
What This Means for the World
China’s advancements in artificial intelligence are significant not only for the nation itself but also for the world. Similar to the space race during the Cold War, the rivalry between the US and China in AI is accelerating innovation at an incredible pace, motivating both countries to surpass each other in this competition. These technologies are transforming entire industries and creating new business models and opportunities for those who are willing to embrace them.
However, with great power comes great responsibility. As AI evolves, so do ethical and regulatory challenges. Take, for example, what Bloomberg columnist Catherine Thorbecke found when she attempted to create a video clip of Chinese President Xi Jinping. The request was denied, highlighting the platform’s adherence to regulations that prevent manipulation of Chinese political figures even though similar content about Donald Trump was allowed. How would this be addressed; we do not know yet.
Yet, there’s no denying that China’s AI industry is proving that restrictions and embargoes are little more than speed bumps on the road to innovation. With companies like DeepSeek, Kling, and Vidu 2.0 leading the charge, China isn’t just catching up to the US—it’s poised to leave it in the dust.
Get ready, folks. In the race for AI dominance, Chinese models are outpacing those from the US. Whether you’re a tech enthusiast, a business leader, or just an everyday user, one thing is clear: the competition for the future of AI will definitely be exciting.
In case you missed:
- Deep Impact: How Cheap AI like DeepSeek Could Upend Capitalism
- Unbelievable: How China’s Outsmarting US Chip Ban to Dominate AI
- What are Text-to-Video Models in AI and How They are Changing the World
- How Lionsgate-Runway Deal Will Transform Both Cinema & AI
- AI Taken for Granted: Has the World Reached the Point of AI Fatigue?
- A Manhattan Project for AI? Here’s Why That’s Missing the Point
- Rise of the Robolympics: When R2-D2 Meets Rocky Balboa
- Rise of Generative AI in India: Trends & Opportunities
- Susan Wojcicki: The Screaming Legacy of The Quiet Architect of the Digital Age
- 9 new ways to power data centers: the unthinkable to the absurd