不是随口感慨的空谈，老黄这次从技术、社会、人性各个维度分别探讨了这件事，明显最近对此有不少思考。 但这还不是全部，在 Lex Fridman的最新专访 中，黄仁勋还讲了更多… 暴论频出 ，火力全 开： Ilya错了，预训练远未触顶，合成数据将进一步推动数据规模扩张。 随后更进一步，驳斥了同行对 " 推理时Scaling " 这条路线的低估。 推理绝非轻量级计算。若说预训练是阅读，那推理就是思考，而思考远比阅读难。 而整场访谈最有趣的，无疑是在老黄所谓"AGI已经实现后"，他基于这个第一性原理，对未来所做出的展望。 Jay 发自 凹非寺 量子位 | 公众号 QbitAI 我们已经实现了AGI。 老黄最新的暴论，震惊了所有人。 OpenClaw就是Token时代的iPhone。来最常跟你聊天，或许是你的龙虾。 智能将成为一种可按需定价的商品，以Token的形式大规模流通。 程序员不会被AI取代。写代码其实并不是核心能力，这个群体将从3000万激增至10亿。 为了实现太空旅行，我可能会做个人形机器人，再通过模型的形式把我的意识上传进去。 智能这个词被过度神化了。人性、品格、同情心与慷慨，这些才是最宝贵的特质。 这是 ...

、 本周观点 风险提示 后续政策不及预期风险；中美关系变化风险；内容上线及表现不及预期风险；宏观经济运行不及预期风险；AI 技 术迭代和应用不及预期风险；政策监管风险。 敬请参阅最后一页特别声明 1 咖啡茶饮：1）咖啡：高景气维持；行业具备β性红利，各品牌依旧积极开店，茶饮品牌亦有所布局；阿拉比卡 咖啡期货价格下跌后续有望带来成本改善。2）茶饮：略有承压；4 月起受去年外卖补贴影响基数逐步提高，关注 数据波动风险。 电商：持续承压。1-2 月网上商品零售额 20812 亿元，增长 10.3%，受今年春节错期去年提前网购备货等因素影 响，1-2 月实物电商表现良好，但受国内消费环境影响，持续性仍需观察。 流媒体平台：本周腾讯音乐发布财报，2025年总收入329亿元，yoy+15.8%，经调整归母净利润95.9亿元，yoy+25%。 行业竞争加剧，预计今年在线音乐订阅收入增速将放缓，关注非订阅音乐收入开启第二增长曲线。 虚拟资产&资产交易平台：虚拟资产&资产交易平台：宏观波动，加密市场情绪仍然较低，币价承压。本周 SEC 与 CFTC 发布新监管指引：加密资产分为五大类，多数不属于证券；SEC 批准纳斯达克代币化 ...

Core Viewpoint - NVIDIA is positioned as an "accelerated computing company" rather than merely a GPU company, emphasizing the importance of the entire technology stack in AI development [2][10][24]. Group 1: AI Competition and Token Economy - The AI competition has shifted from merely computing power to producing high-quality results quickly and cost-effectively, with the entire process needing acceleration [4][5]. - Tokens are viewed as the core currency of the AI era, where smarter tokens can command higher prices, reflecting the efficiency of the models generating them [7][8]. - NVIDIA's acquisition of Groq and the introduction of Groq LPU aim to address the challenge of generating tokens with low latency, complementing existing GPU capabilities [9][10]. Group 2: Full-Stack Approach and Industry Integration - NVIDIA is transitioning from a focus solely on chips to a comprehensive understanding of applications, necessitating a full-stack approach to accelerate software and tools used by AI [12][20]. - The company aims to build AI factories and infrastructure globally, integrating various components like networking and storage to enhance overall system performance [22][26]. - The integration of AI with existing human tools, such as Excel and SQL, requires significant acceleration to keep pace with AI's rapid processing capabilities [14][15][30][31]. Group 3: Future of AI Models and Architectures - The limitations of current models like Transformers necessitate the development of new architectures that can handle long-term memory and continuous tasks more effectively [33][36]. - AI's ability to generate economic value is linked to its improved reasoning capabilities, allowing it to perform tasks beyond mere information generation [40][41]. - The emergence of coding agents signifies a shift where AI can assist in programming, enhancing efficiency and allowing engineers to focus on higher-level problem-solving [45][46]. Group 4: Role of CPUs and System Design - CPUs remain crucial in the AI ecosystem, with NVIDIA emphasizing the need for high-performance CPUs to prevent bottlenecks in GPU utilization [53][64]. - The design of CPUs like Vera focuses on high I/O bandwidth and single-thread performance to support the demands of AI applications [64][66]. - NVIDIA's strategy includes a collaborative approach with various architectures, ensuring that the best components are utilized for optimal system performance [66][87]. Group 5: Supply Chain and Market Dynamics - The current landscape shows that nearly all aspects of the supply chain are nearing capacity, making it challenging to scale any single component significantly [92][95]. - NVIDIA's proactive supply chain planning positions it favorably to meet future demands, despite potential constraints in power and chip availability [95][96]. - The company recognizes the importance of maintaining a competitive edge in the technology stack across all layers of AI development, from infrastructure to applications [98][99].

ondo generated $15.2m revenue in january-february with $944m in tokenized stocks. 60% market share. token is down 87% from december peak, trading below coinbase listing price. 1.94b tokens just unlocked. infrastructure thesis intact, token absorbing maximum pain. separation between product success and token performance rarely this extreme ...

Core Viewpoint - Nvidia's recent financial report serves as a "stress test" for the AI hardware sector, reflecting strong performance amid concerns about sustainability in capital expenditures for AI [2][3]. Financial Performance - Nvidia reported a 73% year-over-year revenue increase, rising from $39.3 billion to $68.1 billion, with over 91% of sales coming from the data center segment [3][4]. - The data center business generated $62.3 billion in revenue, exceeding expectations of $60.69 billion, with net profit nearly doubling to $43 billion, or $1.76 per share [4][5]. - The company forecasts next quarter's revenue at $78 billion, significantly above analyst expectations of $72.6 billion, excluding potential revenue from China [3][4]. Product Developments - The upcoming Vera Rubin system is expected to enhance performance by tenfold per watt, crucial for data centers facing power constraints [6]. - Nvidia is expanding its supply chain from Asia to the U.S. and Latin America to enhance resilience and meet growing AI infrastructure demands [8]. Market Trends - The demand for Nvidia's networking technology, particularly NVLink and Spectrum-X, has surged, with network device sales increasing by 263% year-over-year to $10.98 billion [13]. - Nvidia's gaming segment saw a 47% year-over-year revenue increase to $3.7 billion, but a 13% decline quarter-over-quarter, attributed to memory resource constraints [15]. Strategic Investments - Nvidia has invested $17.5 billion in private enterprises and infrastructure funds, primarily to support early-stage startups, while also pursuing a significant partnership with OpenAI [17][19]. Future Outlook - Nvidia's CEO Jensen Huang emphasizes that computational power equates to revenue, as the demand for AI-driven workloads is expected to grow exponentially [20][21]. - The company is committed to delivering new AI infrastructure annually, with a focus on performance improvements that exceed Moore's Law [21]. Industry Insights - Huang argues that AI will enhance, rather than threaten, existing software companies, as businesses will leverage agentic AI to improve efficiency [23][24]. - The shift towards AI-driven software will necessitate substantial investments in computational resources, with the potential for revenue generation through token production [25][26].

I spent a lot of time thinking about Flying Tulip FDV, this is a problem we had during the original pitches as well.Normal raises are simple, there will be X amount of tokens, so X * price.FT raise doesn't technically raise against the token, it raises against an American (exercise-anytime) put option with no expiry. So this then led me down the rabbit hole of how would you price that?So then instead, I tried to price perpetual American put options against similar apps; Aave, Ethena, Uniswap, Hyperliquid, e ...

Ethereum crypto game @AxieInfinity is introducing a new token that will be airdropped to stakers and those who play its upcoming land game. https://t.co/XVJNl4lBxL ...

Vitalik Buterin said the top priorities are Decentralized Social (DeSoc) and smarter DAOs, but many at the application layer have strayed from these goals. The recent surge in memecoins reflects this shift—by early 2025, even Trump launched a memecoin, and when he later issued a second token, MELANIA, the original TRUMP token had effectively lost its relevance. ...

What's your favorite winter-themed token? ⛄️ ...

TOKEN REVIEW: $JASMY - Does @Jasmy_Global have a native token to match its impressive industry partnerships...? Read now https://t.co/caI4AhwLWa ...