Core Insights - The article discusses the efficiency and cost implications of open-source models like DeepSeek-R1 compared to closed-source models, particularly focusing on token consumption and its impact on overall reasoning costs [2][19]. Token Consumption and Efficiency - A study by NousResearch found that open-source models, specifically DeepSeek-R1-0528, consume four times more tokens than the baseline for simple knowledge questions, indicating significant inefficiency in straightforward tasks [2]. - For more complex tasks such as math problems and logic puzzles, the token consumption of DeepSeek-R1-0528 is reduced to about twice the baseline, suggesting that the type of question posed significantly affects token usage [3][6]. AI Productivity Index - An independent study by AI recruitment unicorn Mercor noted that models like Qwen-3-235B and DeepSeek-R1 have longer output lengths compared to other leading models, which can enhance average performance at the cost of increased token consumption [5]. Economic Value of Tokens - The economic value of tokens is determined by the model's ability to solve real-world problems and the monetary value of those problems, emphasizing the importance of creating economic value in practical scenarios [10]. - The unit cost of tokens is crucial for the economic viability of models, with companies like NVIDIA and OpenAI exploring custom AI chips to reduce inference costs [10]. Hardware and Software Optimization - Microsoft’s research highlighted that actual energy consumption during AI queries can be 8-20 times lower than estimated, due to hardware improvements and workload optimizations [11]. - Techniques such as KV cache management and intelligent routing to appropriate models are being explored to enhance token generation efficiency and reduce consumption [11]. Token Economics in Different Regions - There is a divergence in token economics between China and the U.S., with Chinese open-source models focusing on achieving high value with more tokens, while U.S. closed-source models aim to reduce token consumption and enhance token value [15][16]. Environmental Impact - A study indicated that DeepSeek-R1 has the highest carbon emissions among leading models, attributed to its reliance on deep thinking and less efficient hardware configurations [18]. Overall Cost Advantage - Despite the higher token consumption, open-source models like DeepSeek still maintain a cost advantage overall, but this advantage diminishes at higher API pricing levels, especially for simple queries [19]. Conclusion on AI Economics - The pursuit of performance is overshadowed by the need for economic efficiency, with the goal being to solve valuable problems using the least number of tokens possible [20].

以下文章来源于嬉笑创客 ，作者CB 谈创业事，谈经济，谈生活 本文来自微信公众号： 嬉笑创客 ，作者：CB，题图来自：AI生成 嬉笑创客 . 老罗对西贝预制菜开炮，但估计最后受伤的只是西贝，而预制菜就像冷兵器时代末的火器，非但 不会衰落，只会更进化普及。 很久以前和做餐饮互联网的朋友交流，已经明确得知，在商场和外卖平台上的餐厅，基本有9成 在9成的品类上采用了预制菜。 快速出餐、成本可控、后厨简单，如果不用预制菜和中央厨房，基本是不可能三角，尤其是中 餐。 过去十年餐饮进商场化的浪潮加上外卖化，基本将餐饮工业品化了，在扩大了餐饮供给的同时， 也让食物失去了锅气、烟火味和个性化。 现代人享受着工业化的产品，但依然怀念着田园时代的手工作业淳朴性，是人性使然，但很难实 现。 除非加钱。 先讲前者。 再讲后者。 但能收取溢价的必要条件是什么？是消费者愿意为新鲜、手工和现做服务付费，并且预制还是鲜 炒的信息能被清晰传递给消费者。 前者在中国已实属不易，后者更是难上加难。 还是说社会定价本身有扭曲，如果有，扭曲的原因是什么？当然，亏本定价已经是很中国的特色 了，在工业品和出口上已经有着充分研究，服务业也值得研究。一个可能 ...

Core Viewpoint - The article highlights NVIDIA's GTC 2025 event, emphasizing the shift in AI focus from training to inference, showcasing new hardware and software innovations aimed at enhancing AI capabilities and applications [1][3][30]. Group 1: Key Innovations and Products - NVIDIA introduced the Blackwell Ultra GPU series and the next-generation architecture Rubin, with plans for the Vera Rubin NLV144 platform to launch in the second half of 2026 and Rubin Ultra NV576 in the second half of 2027 [5][10]. - The Blackwell Ultra architecture significantly enhances AI performance, achieving a 1.5x improvement in AI performance compared to the previous generation, and offers a 50x increase in revenue opportunities for AI factories [8][10]. - The new CPO switch technology aims to reduce data center power consumption by 40MW and improve network transmission efficiency, laying the groundwork for future large-scale AI data centers [13][14]. Group 2: AI Inference and Software Upgrades - NVIDIA's new AI inference service software, Dynamo, is designed to maximize token revenue in AI models, achieving a 40x performance improvement over the previous Hopper generation [19][21]. - The introduction of AI agents and the Ll ama Nemo tr o n series models aims to facilitate complex inference tasks, enhancing capabilities in various applications such as automated customer service and scientific research [20][30]. Group 3: Robotics and Physical AI - NVIDIA launched the GROOT N1, the world's first open-source humanoid robot model, designed for various tasks such as material handling and packaging, indicating a significant step towards the commercialization of humanoid robots [25][30]. - The company also introduced new desktop AI supercomputers, DGX Spark and DGX Station, aimed at providing high-performance AI computing capabilities for researchers and developers [23][24]. Group 4: Market Sentiment and Future Outlook - Despite the significant technological advancements presented at GTC 2025, NVIDIA's stock price fell by 3.43% post-event, reflecting ongoing market concerns regarding AI spending and competition [28][29]. - Analysts suggest that while there are concerns about AI capital expenditure growth in 2026, the overall sentiment may improve due to the innovations showcased at the event [29][30].