一万亿 Blackwell和Rubin的订单 到2027年 至少1万亿美元 性能提升50倍 我们的单个token成本是全球最低的 你不可能打败我们 推理之王 Hello 大家好 欢迎来到英伟达2026年GTC大会 又是一年过去了 我们来看看黄仁勋要讲什么新故事 ChatGPT爆发的三年半之后 黄仁勋开始停止了讲芯片的故事 他开始瞄准更大的叙事 token经济 而这 将是更有野心且更持久的市场蛋糕 这个视频 我们将拆解英伟达的 “五层蛋糕”生态体系 一万亿美元的收入是怎么算出来的 基于Groq的LPU 将给英伟达带来什么新的机会 还有黄仁勋在光通信上的提前布局 以及未来Scale-across（跨域扩展）的 AI工厂机会 以及OpenClaw的里程碑之后 英伟达押注的token经济学将会如何发展 给算力市场将带来了如何的挑战 token的推理使用量事实上是一直在暴增 每一个组件开始亮红灯 我讲亮红灯是指就开始大缺货 然后开始大涨价 现在的产能还是没有办法跟当时老黄 所预计的产能达到一致 这个是我们从来没有看到过的 （内存）supercycle（超级周期） 大概率也要到2028年才能有 实质性的（供应）上涨 在整 ...

Core Insights - The article discusses the evolution of tokens in the context of AI and cryptocurrency, highlighting the shift from crypto tokens to AI tokens, which are now seen as essential for productivity rather than speculative assets [6][42]. - It emphasizes that while crypto tokens are driven by speculation, AI tokens are driven by their utility in enhancing productivity and decision-making processes in various industries [42][43]. Group 1: Token Evolution - In 2009, an anonymous individual created the concept of "token" through computational power, leading to the birth of the cryptocurrency economy [6]. - By March 2025, a new type of token was defined, where computational power is used to generate tokens that are immediately consumed in AI inference processes, marking the acceleration of the AI economy [6][12]. - The article draws parallels between the original token concept and the new AI token economy, stating that both involve inputting computational power to produce valuable outputs [7][21]. Group 2: Economic Models - Huang Renxun's presentation at NVIDIA GTC 2026 focused on the economic model of token production, pricing, and consumption, rather than specific hardware specifications [11][12]. - He proposed a structured approach to data center power allocation, dividing resources into tiers based on performance and pricing, which reflects a comprehensive understanding of the token economy [12][24]. - The relationship between inference efficiency and token consumption was illustrated, showing different pricing tiers for various AI models, ranging from free to $150 per million tokens [16][24]. Group 3: Scarcity and Competition - The article discusses the concept of artificial scarcity created by Nakamoto through Bitcoin's capped supply of 21 million coins, which serves as a value anchor for the cryptocurrency economy [26]. - In contrast, Huang Renxun's approach to scarcity is based on physical laws, emphasizing the substantial investment required to build data centers and the inherent limitations of resources like land and electricity [27][28]. - The competition in the AI hardware market is likened to the historical evolution of mining hardware, with NVIDIA positioning itself as a leader in the AI token economy by defining market standards and usage scenarios [33][39]. Group 4: Market Dynamics - The article highlights that the demand for crypto tokens is largely speculative, while AI tokens are essential for operational efficiency, as companies like Nestlé utilize them to enhance supply chain decision-making [42][43]. - The distinction between the two types of tokens is critical: crypto tokens are held for value appreciation, whereas AI tokens are consumed for immediate utility, leading to a more stable economic model for AI tokens [43][44]. - The article concludes that the AI token economy is less prone to bubbles compared to the crypto token economy, as its value is derived from actual usage rather than speculation [44][45].

Group 1 - The article discusses the evolution of tokens, comparing the original crypto tokens introduced by an anonymous creator in 2009 to AI tokens defined by NVIDIA's CEO Jensen Huang in 2026, highlighting the shift from speculative value to practical utility in the AI economy [4][30]. - Huang's presentation at the NVIDIA GTC 2026 emphasized a comprehensive economic model for token production, pricing, and consumption, indicating a structured approach to AI token economics [7][14]. - The relationship between power consumption and token output is illustrated through a graph presented by Huang, which categorizes different pricing tiers for AI tokens based on their performance and usage [9][10]. Group 2 - The article contrasts the scarcity mechanisms of crypto tokens, which can be altered through forks, with the physical limitations imposed by data center infrastructure that Huang describes, emphasizing the natural scarcity of AI tokens [17][20]. - Both the crypto mining and AI inference processes are framed as converting electricity into value, with the article noting the historical evolution of hardware in both sectors, from CPUs to specialized ASICs for mining and AI [15][21]. - NVIDIA's strategic positioning in the AI token economy is highlighted, showing how it has moved from being a hardware supplier to defining market standards and usage scenarios for AI tokens, unlike competitors in the crypto mining space [26][27]. Group 3 - The article identifies a fundamental difference in the motivations behind crypto tokens and AI tokens, with crypto tokens driven by speculation and AI tokens driven by productivity and immediate utility [30][31]. - The demand for AI tokens is linked to their practical applications in business, contrasting with the speculative nature of crypto tokens, which rely on belief in future value [31][32]. - Huang's assertion that "tokens are the new commodity" reflects a consensus in the industry regarding the established value of AI tokens, as evidenced by widespread usage in various applications [14][33].

Core Insights - The token economy in both China and the U.S. is heavily reliant on electricity, with each country facing unique challenges in this regard [1][3] - The U.S. is experiencing a power shortage due to outdated generation and grid infrastructure, limiting token production [1][2] - In contrast, China faces high token production costs due to relatively low-efficiency hardware, impacting the overall cost of token generation [1][3] Group 1: U.S. Challenges - Microsoft CEO Satya Nadella emphasized that the real issue is not a shortage of GPUs but a lack of electricity, which restricts token production and monetization [1] - Major U.S. tech companies are in a race for AI infrastructure investment, which has turned into a competition for electricity supply [1][2] - The construction of large-scale data centers in the U.S. is progressing from 1GW to 10GW, with companies like Crusoe targeting significant capacity increases [1][2] Group 2: Infrastructure and Policy - Silicon Valley giants are urging the White House for support in developing infrastructure, particularly the power grid, to match the pace of AI innovation [3] - OpenAI has suggested that the U.S. needs to add 100GW of electricity capacity annually to compete effectively in AI against China [3] - The U.S. added 51GW of power capacity last year, while China added 429GW, highlighting a significant "power gap" [3] Group 3: China's Challenges - China's AI infrastructure is built on domestic chips, which currently have lower efficiency, leading to increased demand for computational power [3][4] - ByteDance's daily token calls have surged from 16.4 trillion in May to 30 trillion in September, indicating a rapid increase in computational needs [3] - The cost of electricity for a major cloud provider in China is estimated at 8-9 billion yuan for 1GW annually, reflecting the high operational costs associated with domestic chip usage [5] Group 4: Efficiency and Cost - The competition in the token economy involves not just hardware but also the software, tools, and the electricity and cooling systems required to operate them [4] - Huawei's CloudMatrix 384 has shown a significant increase in total computational power but at a much higher energy cost compared to NVIDIA's latest offerings [5][6] - The average industrial electricity cost in the U.S. is approximately 9.1 cents per kWh, while certain regions in China have reduced costs to below 4 cents per kWh, indicating a competitive advantage for Chinese data centers [6]

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].

Core Viewpoint - The article discusses the evolution and future of the prepared food industry, emphasizing that while there are criticisms, the trend towards industrialization and prepared meals is likely to continue and evolve rather than decline [5][10]. Group 1: Prepared Food Industry Dynamics - A significant majority (90%) of restaurants in shopping malls and on delivery platforms utilize prepared foods, driven by the need for quick service, cost control, and simplified kitchen operations [5]. - The industrialization of food has expanded supply but has also led to a loss of traditional cooking flavors and personalization in meals [5][6]. - Consumers express a desire for fresh, handmade meals but are often unwilling to pay a premium for them, complicating the market for differentiated food offerings [9][10]. Group 2: Consumer Behavior and Market Challenges - The willingness of consumers to pay for quality and differentiated products is limited, particularly in lower-tier markets where traditional cooking is often a low-cost alternative [9]. - The prevalence of low-cost, industrialized products suppresses the market for handmade items, as consumers are conditioned to expect lower prices [9][10]. - The article highlights the challenge of effectively communicating the quality and freshness of prepared foods to consumers, which is essential for justifying higher prices [9]. Group 3: Economic Implications and Industry Profitability - Despite the widespread adoption of prepared foods, the restaurant industry's profit margins remain low, raising questions about market pricing and hidden costs [10]. - The high turnover of service staff in the restaurant industry aligns with the need for standardized, industrialized products, further entrenching this trend [10]. - The article suggests that the future may see a clearer distinction in pricing between handmade and industrialized food products, similar to trends observed in developed markets [10].

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].