Until recently, the major advancements in AI were almost exclusively in the US. Every major LLM (think OpenAI, Gemini, Anthropic, LLaMa) was based there, most access to capital for AI is in the West Coast of the US. It was common knowledge that to compete in this market, companies need huge access to capital, compute, expensive data and expensive talent.
Until last week, when the news that a relatively unknown Chinese company called Deepseek, released its R1 reasoning model which is equivalent to the performance of ChatGPT’s o1 model, but is fully open-source and available at a fraction of the cost (3-5% of OpenAI’s model). It was also said that the model cost only $6M to train, compared to the billion+ dollars it took to train the US model. The news sent the US stock market into a spiral, causing the market to shed $1 trillion in value, with Nvidia alone making the biggest single day drop of $600 billion in market cap. New details about the real costs of training the model (much higher) and the source of the data (distilled from OpenAI against the T&Cs) are slowly surfacing, but it’s clear that Chinese AI startups are here to stay.
As AI reshapes industries worldwide, Chinese companies are at the forefront of developing cutting-edge models and applications. It remains to be seen how the US/China trade war and curbs on AI chip exports will impact the future of this competition, but there have already been significant strides made in the past few weeks. Let’s dig in to the main players.
DeepSeek: A Rising Star in AI
DeepSeek, a relatively unknown AI firm based in Hangzhou, has quickly emerged as a formidable player in China’s AI ecosystem. The company claims its AI models rival those of top U.S. firms at a fraction of the cost. DeepSeek reportedly trained its large language model (LLM) for under $6 million—significantly less than OpenAI’s estimated costs for ChatGPT (the figure has been widely contested and is likely to be 400x higher in reality),
The news about Deepseek spread like wildfire and the Deepseek app became the most downloaded app in the US, surpassing ChatGPT. With its overnight success, the conspiracy theories about data being shared with China, coupled with an alleged hack of Deepseek, raised security concerns with the Pentagon, who immediately moved to block government employees from using it.
Unlike many other AI startups, DeepSeek benefits from the backing of a major Chinese quantitative hedge fund, allowing it to offer its AI models and applications for free or nearly free. This approach has garnered significant traction, raising questions about the sustainability of its pricing model but also positioning it as a potential game-changer in AI-driven industries like drug development.
DeepSeek’s bold claims briefly shook confidence in the AI investment landscape. Investors and analysts questioned how the company could achieve such efficiency, with some doubting that it truly trained its models for under $10 million. Additionally, concerns around the accuracy and energy efficiency of DeepSeek’s models surfaced, prompting a more cautious evaluation from industry players. However, despite the initial panic, venture capital funding for AI remains robust.
Days after the R1 debut, Deepseek released Janus Pro 7B, a vision AI model capable of generating photorealistic images, answering visual questions and performing captioning tasks. The open-source model outperforms competitors like OpenAI’s DALL-E 3 and Stable Diffusion 3 on key benchmarks, all while running on fewer computational resources.
The Dragons, Tigers and Their Challenges
While DeepSeek thrives, many other Chinese AI startups, collectively known as the “six little dragons”—MiniMax, Moonshot AI, Zhipu AI, Stepfun, Baichuan, and 01.AI—face increasing challenges. These companies, which initially aimed to build general-purpose LLMs, are now contending with a price war and a general reluctance among Chinese businesses to pay for AI subscriptions.
To stay competitive, several of these startups are shifting strategies:
- MiniMax is focusing on international expansion with its AI conversation app, Talkie, targeting U.S. consumers.
- Baichuan AI is pivoting toward specialized AI models for finance and healthcare.
- 01.AI has partnered with Alibaba to focus on AI applications in e-commerce rather than competing directly with OpenAI.
- Zhipu AI has developed “Ying,” a text-to-video AI model, opening new avenues in content generation.
This review of the Chinese AI ecosystem by Spacecadet Ventures, goes deeper into each one.
ByteDance: A Dominant Force in AI
ByteDance, the parent company of TikTok, has cemented its dominance in China’s AI landscape. The company’s AI offerings have made it increasingly difficult for smaller startups to compete, as ByteDance provides powerful AI tools at an extremely low cost. For example, its multimodal AI model is priced at just 0.003 yuan per 1,000 tokens.
One of ByteDance’s most innovative AI developments is OmniHuman, an advanced AI system capable of transforming individual photos into highly realistic videos. OmniHuman generates full-body videos with natural movements, speech, and expressions, setting a new standard for AI-driven digital avatars. The system was trained on over 18,700 hours of video data, allowing it to produce lifelike motion effects that have broad applications in entertainment, gaming, and social media.
This video of Einstein speaking, was generated from a single image:
Additionally, ByteDance has dedicated a team within its video editing platform, CapCut, to further AI development, signalling its commitment to creating next-generation AI-powered content tools.
Alibaba: A Key AI Investor and Innovator
Alibaba remains a central player in China’s AI expansion, acting both as an innovator and a backer of emerging AI startups. The company has invested in:
- MiniMax and Moonshot AI, supporting their advancements in generative AI.
- 01.AI, collaborating on a joint research lab focused on smaller, industry-specific AI models.
- Its own AI model, Qwen 2.5-Max, which Alibaba claims outperforms DeepSeek, OpenAI, and Meta’s offerings in key benchmarks. Alibaba’s Qwen2.5-Max uses a “mixture-of-experts” architecture which allows it to perform complex tasks like reasoning and coding efficiently, while also significantly reducing infrastructure costs (by up to 60%). Qwen can code, search the web and generate images and videos in high resolution.
Alibaba’s cloud division continues to push AI adoption across multiple industries, reinforcing the company’s role as a driving force in China’s AI ecosystem.
Kling and Krea: Pioneering AI-Generated Video
As generative AI evolves, China is making significant progress in AI-powered video creation. Two key players in this space are:
- Kling, a text-to-video model developed by Kuaishou, which has been likened to OpenAI’s Sora in terms of capabilities. Kling allows users to generate high-quality videos from text prompts, with features like smooth motion, extended video sequences, and high resolution.
- Krea, another AI-driven video platform, is pushing the boundaries of realistic and interactive AI-generated content.
These advancements highlight China’s growing expertise in multimodal AI and its potential to redefine digital storytelling and content creation.
The Global AI Investment Landscape
Despite the excitement around DeepSeek, venture investments in Chinese startups have nearly flatlined in 2024, at a time where global investments (particularly in AI) are booming.
Approximately 30% of all venture capital dollars invested worldwide went to AI companies, and AI funding crossed the $100 billion mark in 2024.
SoftBank is reportedly in talks to invest between $15 billion and $25 billion into OpenAI (as part of a $40 billion round), just days after DeepSeek’s announcement. If completed, this round would value OpenAI at $340 billion, nearly doubling its valuation from October. This move signals that investors are still placing enormous bets on established AI leaders.
Similarly, private equity giant Blackstone has reaffirmed its commitment to AI infrastructure, investing heavily in data centers and storage companies essential for AI growth. With over $8 billion committed to AI-related infrastructure in Spain and the U.S., Blackstone remains bullish on the sector despite emerging competition from China.
However, many investors remain skeptical of DeepSeek’s ability to gain traction outside China. Regulatory hurdles and data privacy concerns make it difficult for Chinese AI firms to penetrate the U.S. and European enterprise markets. While DeepSeek’s technology has impressed, its long-term global impact is still uncertain.
Implications for the Global AI Landscape
China’s AI sector is at a crossroads, with some companies achieving rapid growth while others face significant obstacles. DeepSeek’s low-cost AI models, ByteDance’s aggressive expansion, and Alibaba’s strategic investments showcase the dynamic and competitive nature of China’s AI industry. Meanwhile, challenges like regulatory constraints, pricing pressures, and the difficulty of monetizing AI continue to shape the future of the sector.
As Chinese AI companies refine their strategies—whether through international expansion, sector-specific AI solutions, or new business models—their influence on the global AI ecosystem will only grow. The next few years will determine which of these rising stars solidify their positions as global AI leaders and how their innovations shape the future of artificial intelligence. Meanwhile, despite the initial shock of DeepSeek’s arrival, the broader AI investment landscape remains as strong as ever, proving that the AI boom is far from over.
I'm a former general partner at google ventures, head of Google for Entrepreneurs in Europe and founding head of Campus London, Google's first physical hub for startups.
I'm also the founder of Techbikers, a non-profit bringing together the startup ecosystem on cycling challenges in support of Room to Read. Since inception in 2012 we've built 11 schools and 50 libraries in the developing world.
- Don’t Sleep on Consumer Tech: AI is Changing Everything - March 16, 2025
- Weekly Firgun Newsletter – March 14 2025 - March 14, 2025
- Israeli Pre-Seed Report 2025 - March 12, 2025
