Just finished watching the latest All In Podcast, Nvidia's financial report exploded, Salesforce rebounded, and the 30-year US Treasury bond broke 5.2% NVIDIA's financial report thoroughly exposes AI capital expenditure at its peak Nvidia's Q2 revenue was $96.2 billion, a sharp increase of 106% year-on-year, with a record breaking core net profit of $60 billion and a gross profit margin of 75%. What's even more explosive is the management's guidance for next year's growth, which is as high as 70%, far exceeding Wall Street's expectation of 45%. The management also made it clear that this number was given under limited production capacity, which means that the demand itself has not been fully met. Sacks pointed out that NVIDIA's performance basically shattered the narrative that AI's capital expenditure was just a short-term foam, which proved that the investment curve of Jianli has a strong long-term continuity. Although the performance is so impressive, Nvidia's dynamic valuation has been suppressed due to the rapid outbreak of profits, and the forward P/E ratio is now in an attractive range. The program also mentioned something Nvidia has been doing recently: investing billions of dollars in intensive mergers and acquisitions, including the acquisition of Hugging Face and the Poolside team, with the aim of gaining deep control over the distribution portal for open-source models and code agents. The boundaries of the entire industry are also melting, and ultra large scale cloud manufacturers are starting to develop their own chips. Nvidia, in turn, is developing its own models, providing cloud services, and offering end-to-end APIs. The giants are moving towards a comprehensive competition of full stack vertical integration. Salesforce skyrockets by 20% in a single day, debunking the SaaS doomsday theory Salesforce's surge of over 20% in a single day directly challenges the simple inference that traditional SaaS will be phased out due to AI intelligence being able to write code in the market. The analysis of several guests is that the moat of enterprise level software lies in the long-term accumulation of compliance, security architecture, over 20 years of bug fixing experience, and its position as the source of core system facts. These things cannot be easily replaced by building an AI agent or writing a few prompts. Chamath proposed a clear three-stage framework to understand the evolution of AI: the first stage is the "brain," which refers to the question answering ability of the basic large language model. The second stage is "hands and feet", which involves equipping the model with eyes, hands, and a user interface to enable it to truly work. The third stage is' domain experts', who combine internal private domain data and workflows to train the agent into a qualified salesperson, legal or customer service representative. Salesforce's approach this time is also quite clever, directly and deeply integrating with Anthropic's Claude, and even willing to let Claude become the front-line interaction entrance for users. Sacks and Jason's viewpoint is that the ultimate winner of SaaS software in the future is to build the best Agent interface (API or command-line tool), turning external AI into the "super user" within the system, and unleashing the value of complex functions that have been buried under traditional UI. The program also distinguishes the fate of vertical and horizontal SaaS. Workflow software that focuses on unique business proprietary processes (such as breeding, physical research and development in specific industries) will be more favored due to its customized value, while general horizontal tools (such as CRM, email, and spreadsheets) rely on ecological networks and data bases to solidify their position. The 30-year US Treasury bond has broken through 5.2%, and a refinancing tsunami is approaching This is the most tense part of this episode. The yield on 30-year US Treasury bonds has climbed to a high of 5.2% to 5.3%. The current average debt cost of the federal government is 3.4%, and for every 1 percentage point increase in interest rates, the United States has to pay an additional 1.25% of GDP in interest each year. This snowball is quite alarming. What is more difficult is that as many as $10 trillion of outstanding treasury bond will expire in the next 12 months, which needs to be refinanced at a higher interest rate, the deficit pressure will only further deteriorate. Finance Minister Besant tried to suppress the long-term yield by increasing the scale of long-term treasury bond bond repurchases from $2 billion to $4 billion each time, but legendary investor Druckenmiller wrote a criticism, saying that the Ministry of Finance's intervention in prices can cure the symptoms but not the root cause, and the real root cause is the runaway fiscal deficits and spending problems of Congress and the government. Several guests also discussed the institutional dilemma behind this, the structural "tragedy of the commons" coupled with the incentive mechanisms of bipartisan politics, making it extremely difficult to cut spending. Friedberg also issued a warning that long-term deficits and inflation will continue to erode residents' purchasing power, and may trigger serious institutional and fiscal breakpoints in 2026-2028, or even 2030-3022 (a time point when social security funds and some state finances are under pressure). However, everyone unanimously agrees that the productivity index explosion brought by AI is the only antidote for the US economy to truly "outperform debt" without triggering a vicious debt spiral. Several interesting derivative topics One reason is that Druckenmiller used AI to polish his column in The Wall Street Journal, which sparked a small controversy. Jason felt that not declaring the use of AI was somewhat damaging to his sincerity and the author's own voice. Chamath, Sacks, and Friedberg argue that AI, like Excel, calculators, or electronic music software, is essentially a mind amplifier. What matters is whether the viewpoint itself is original and whether the logic holds up. There is no need to engage in moral blackmail on whether digital tools are used or not. The second breakthrough is Moderna's personalized cancer vaccine (mRNA neoantigen therapy). The principle is to extract tumor samples from patients for DNA sequencing, identify the unique mutation antigen fingerprint of the tumor, and then use mRNA to guide the expression of this antigen in cells in the body, thereby accurately activating the immune system to hunt down cancer cells. Friedberg questioned the commercialization aspect, pointing out that this underlying immunotherapy has actually been developed with public research funding for decades, and the technology cost itself is not high. However, pharmaceutical companies are charging up to $500000 in treatment fees through regulatory and patent barriers. He called for the future to develop towards low-cost treatments overseas and more accessible civilian healthcare.