# Another week, another DeepSeek, and another insane valuation for OpenAI Source: https://www.youtube.com/watch?v=mmdX3yjTVic Recap page: https://rapidrecap.app/video/mmdX3yjTVic Generated: 2025-12-01T14:35:48.955+00:00 --- ## Quick Overview Chinese AI advancements, exemplified by Moonshot AI's Kimi K2 model costing only $5.6 million to train and ranking second on an intelligence index, are rapidly closing the gap with US tech giants like OpenAI, whose models like GPT-4 and Gemini cost significantly more, leading to questions about the sustainability of massive US spending and the longevity of their AI lead. **Key Points:** - Moonshot AI's Kimi K2, an open-source reasoning model, cost less than $6 million to train, compared to $78 million for ChatGPT and $191 million for Gemini. - Kimi K2 ranks second on the Artificial Analysis' Intelligence Index (as of Nov 2025) with a score of 67, just behind GPT-4 (high) at 70. - The rapid progress of Chinese AI labs, which are developing comparable models at a fraction of the cost, is forcing investors to question whether massive US spending will yield better commercial returns. - Airbnb CEO Brian Chesky noted that while his company uses some OpenAI models, they rely heavily on Alibaba's Qwen model because it is fast and cheap, and that OpenAI's connective tools aren't quite ready. - The cost gap is striking: Kimi K2 cost $4.6 million to train, while OpenAI's next-generation model is expected before Sam Altman's trillion-dollar data center is built. - OpenAI recently achieved a $500 billion valuation, which Thomas Wolf of Hugging Face suggested was a rounding error compared to Moonshot's recent $3.3 billion valuation, fueling AI bubble fears. ![Screenshot at 00:04: The video opens with host Kevin Walmsley in Kunming, China, introducing the topic by displaying a screenshot of a previous video where DeepSeek's low training cost \($5.6 million\) was contrasted with ChatGPT \($78 million\) and Gemini \($191 million\).](https://ss.rapidrecap.app/screens/mmdX3yjTVic/00-00-04.png) **Context:** The video, titled 'Another week, another DeepSeek, and another insane valuation for OpenAI,' discusses the accelerating competition in the Artificial Intelligence (AI) sector, specifically highlighting the performance and cost efficiency of Chinese AI models relative to their US counterparts. Host Kevin Walmsley uses recent news regarding DeepSeek, Moonshot AI's Kimi K2, and OpenAI's valuation to frame a discussion about the shifting technological landscape and investment concerns. ## Detailed Analysis The video asserts that Chinese AI development is making rapid, cost-effective progress, challenging the dominance of US tech giants. A key example is Moonshot AI's Kimi K2 model, which cost under $6 million to train (specifically cited as $4.6 million in a later quote) and achieved the second-highest ranking (67) on the Artificial Analysis' Intelligence Index, trailing only GPT-4 (high) at 70. This rapid improvement, achieved on significantly smaller budgets and without the benefit of advanced US-controlled chips, contrasts sharply with the massive spending by US firms; for instance, OpenAI recently hit a $500 billion valuation, while Moonshot's was $3.3 billion, which co-founder Thomas Wolf called a 'rounding error' next to OpenAI's valuation, increasing fears of an AI bubble. Furthermore, Airbnb CEO Brian Chesky revealed that despite being friends with Sam Altman, Airbnb relies on Alibaba's Qwen model for its AI customer service agent because it is faster and cheaper than OpenAI's latest models, which he deemed not 'quite ready' for integration into their connective tools. This indicates that Chinese AI models are already competitive in commercial applications, raising the fundamental question for US investors: whether their massive spending translates into superior, commercially viable returns. ### AI Model Cost Comparison - DeepSeek cost $5.6 million to train - ChatGPT cost $78 million - Gemini cost $191 million ### Kimi K2 Ranking - Ranks No. 2 on Artificial Analysis' Intelligence Index with a score of 67 - GPT-4 (high) ranks No. 1 with 70 - Moonshot's Kimi K2 cost $4.6 million to train ### US vs. China AI Spending - US tech giants face questions on ROI for massive spending - Chinese labs match frontier capabilities at lower cost - Chinese models like Qwen and Kimi K2 are already being used in production ### Airbnb's AI Strategy - CEO Brian Chesky stated OpenAI's tools were not 'quite ready' - Airbnb relies heavily on Alibaba's Qwen model because it is fast and cheap - Airbnb's AI agent reduced resolution time from nearly three hours to six seconds ### Valuation Concerns - Moonshot's $3.3 billion valuation is a 'rounding error' next to OpenAI's $500 billion valuation - Increasing fears of an AI bubble are valid ![Screenshot at 00:04: Comparison of AI model training costs showing DeepSeek at $5.6M vs. ChatGPT at $78M and Gemini at $191M.](https://ss.rapidrecap.app/screens/mmdX3yjTVic/00-00-04.png) ![Screenshot at 00:39: A chart illustrating the performance gap closing, showing Kimi K2 Thinking ranked second \(67\) just behind GPT-4 \(high\) at 70.](https://ss.rapidrecap.app/screens/mmdX3yjTVic/00-00-39.png) ![Screenshot at 00:54: A Bloomberg article screenshot discussing that Moonshot AI's Kimi K2 is a turning point in AI, ranking highly despite low cost.](https://ss.rapidrecap.app/screens/mmdX3yjTVic/00-00-54.png) ![Screenshot at 01:34: A bar chart titled 'Closing the Gap' showing Kimi K2 Thinking ranked No. 2 behind GPT-4 \(high\).](https://ss.rapidrecap.app/screens/mmdX3yjTVic/00-01-34.png) ![Screenshot at 04:20: A Bloomberg article excerpt detailing Airbnb CEO Brian Chesky's comments that OpenAI's tools are not 'quite ready' for integration.](https://ss.rapidrecap.app/screens/mmdX3yjTVic/00-04-20.png)