黄仁勋GTC演讲全文：推理时代到来，2027营收至少万亿美元，龙虾就是新操作系统

Core Insights - NVIDIA's CEO Jensen Huang positions the company as a builder of "AI factories," predicting a demand of at least $1 trillion by 2027 [6][8][64] - The concept of "Token Factory Economics" is introduced, emphasizing that performance per watt is crucial for commercial monetization [15][71] - Huang asserts that Agents will replace traditional SaaS models, with a new workplace standard of "salary + token budget" emerging [26][29] Group 1: Performance and Demand - The global AI computing demand has exploded exponentially over the past two years, with significant increases in computational consumption as models evolve [7][64] - Huang previously forecasted a high-confidence demand of $500 billion, which has now been revised to at least $1 trillion for 2027 [8][64] - NVIDIA's systems are claimed to be the "lowest cost infrastructure" globally, with 60% of business coming from the top five cloud service providers [13][66] Group 2: Token Factory Economics - Future data centers will function as "factories" for producing tokens, shifting from mere storage facilities [71] - Huang categorizes future AI services into four commercial tiers based on token generation speed and pricing, with the highest tier priced at approximately $150 per million tokens [16][74] - The architecture of NVIDIA allows for high throughput at the free tier while achieving a 35-fold performance increase at the highest value inference tier [18][68] Group 3: Technological Innovations - The introduction of the Vera Rubin system, which has achieved a 350-fold increase in token generation speed over two years, is highlighted [21][82] - NVIDIA's acquisition of Groq aims to enhance inference performance through a unique combination of hardware and software [22][78] - The company is also developing a space-based data center, Vera Rubin Space-1, to extend AI computing capabilities beyond Earth [30][88] Group 4: Software and Ecosystem Transformation - The emergence of Agents is set to transform SaaS companies into AaaS (Agent-as-a-Service) providers, with a focus on specialized AI agents [28][94] - OpenClaw is introduced as a pivotal open-source project that enables the management of resources and execution of tasks by intelligent agents [27][90] - The future workplace will require engineers to have an annual token budget, reflecting a significant shift in employment structures within the tech industry [29][94]