Sequoia Capital held a 6-hour AI closed-door meeting, revealing 10 future trends of AI (full speech attached at the end of the article)

Written by

Caleb Hayes

Updated on:June-19th-2025

The 2025 Sequoia Capital AI Summit brought together the world's top founders and thought leaders in the field of artificial intelligence to discuss the latest developments in AI technology, industry trends, challenges and unlimited opportunities in the future. This document aims to extract the ten most valuable core viewpoints from the summit, and combined with the in-depth explanations of the guests, present you with a clear picture of the future development of AI. These insights not only reveal the direction of technological evolution, but also provide valuable strategic references for entrepreneurs, developers, investors and followers in the AI wave.

1. Trillion-level AI opportunities have arrived, and the value of the application layer is highlighted

Sequoia Capital pointed out at the beginning of the summit that artificial intelligence is sweeping the two major markets of services and software at the same time with an unprecedented trend, and its potential market size is expected to far exceed the period of cloud computing transformation. Although the capabilities of basic models are becoming increasingly powerful and penetrating into the application layer, the summit generally believes that truly lasting business value and rich profit returns will eventually be realized at the application level that can solve the pain points of specific industries and complex user needs. This view points the way for participants in the AI field. Compared with investing huge sums of money to pursue the construction of the next general basic model, entrepreneurs and developers should focus on how to cleverly use existing and future AI technologies to deepen specific vertical industries or functional scenarios. The core is to provide solutions that can directly solve customers' actual problems and create measurable results. In his speech, Sequoia Capital partner Packardium emphasized that AI companies should " focus on customers and think in reverse " and focus on solving complex problems that may require human intervention. This is the core and value of competition. This means that the business model will gradually shift from traditional sales of software tools and budgets to sales results and even direct provision of intelligent labor.

2. The “Intelligent Economy” is taking shape, and human-machine collaboration opens up a new paradigm

The summit predicts that the commercial application of AI agents, which is still in its infancy, will gradually mature in the next few years and eventually evolve into a complete " agent economy ". In this new economic system, AI agents will no longer be just tools for information transmission. They will be able to transfer resources, execute transactions, record behaviors, understand trust and reliability more autonomously, and collaborate efficiently with human users and other agents. Sequoia Capital partner Constantine outlined this blueprint in his speech, emphasizing that the agent economy is not intended to replace humans, but is "completely built around people". This will profoundly change the future work model and business processes, and AI agents will become efficient digital assistants and partners for individuals and organizations. However, to achieve this grand vision, the technical level still faces three core challenges: persistent identity (the consistency and memory of the agent itself and its understanding of users), seamless communication protocols (standardized protocols that support information transmission, value transfer and trust building between agents, such as Anthropic's MCP) and strong security guarantees (establishing trust and security mechanisms in agent interactions where physical contact is not possible).

3. AI Infrastructure: “Steel, Servers, and Electricity” of the New Industrial Revolution

The rapid development of artificial intelligence, especially the training and reasoning of large models, has put forward unprecedentedly stringent requirements for computing infrastructure. The summit compared the construction of AI infrastructure to a new "industrial revolution", with "steel, servers and electricity" as its core elements. In the future, the energy consumption and power density of AI data centers will far exceed those of traditional data centers, liquid cooling technology will become standard, and the entire data center will be regarded as a unified, highly optimized computing unit, namely the "AI factory" . In an interview, Chase Locke Miller, co-founder of Crusoe, deeply revealed the unprecedented scale and complexity of AI infrastructure construction. He pointed out that energy supply has become a key bottleneck restricting the development of AI, which has not only promoted the trend of building data centers in areas rich in renewable energy, but also spawned the need to build a new generation of power generation facilities including small modular reactors (SMRs) to support AI factories. In this competition, startups like Crusoe, with their agility, vertical integration capabilities and solutions based on "first principles", have shown the potential to surpass traditional technology giants in the rapid deployment of ultra-large-scale AI infrastructure. At the same time, the trend of "sovereign AI" and data localization is also driving investment and construction of local AI factories around the world.

4. Business model disruption: outcome-oriented pricing, reshaping AI value exchange

Brett Taylor, former Facebook CTO and current co-founder of Sierra, said at the summit that artificial intelligence is giving rise to a new business model that is outcome-oriented. In the future, when companies purchase AI services, they will increasingly tend to purchase an AI agent that can complete specific tasks and bring measurable business results, and pay for the actual value it creates, rather than simply paying for software seats or usage time. This "outcome-based pricing" model heralds a major disruption to the traditional SaaS subscription system . Brett Taylor believes that start-ups can gain an advantage in the competition with industry giants with this more flexible and more directly linked pricing strategy to the core value proposition of customers. He reviewed the history of the software industry and pointed out that the innovation of business models is often one of the key factors for Salesforce to beat Siebel Systems and ServiceNow to surpass many traditional ITSM companies. For AI start-ups, business model innovation should not be ignored as a powerful competitive weapon.

5. The survival rules of the founder of OpenAI: When starting an AI business, don’t compete with giants for big models

OpenAI founder Sam Altman sent a clear signal to AI entrepreneurs at the summit: OpenAI will continue to focus on building smarter core AI models and providing core AI subscription services. His core advice to entrepreneurs is: " Don't try to do core AI subscription services, but there is still a lot of room for development in other areas ." This view is widely interpreted as an emphasis on the core business barriers of AI giants, as well as a field division and strategic guidance for ecosystem participants. Sam Altman believes that startups should avoid direct competition with giants such as OpenAI in the fields of basic large model research and development and core subscription services with huge capital investment and extremely high technical barriers. On the contrary, broader opportunities exist in using the APIs and SDKs provided by giants to build products and services with differentiated value in massive application layers, specific market segments or surrounding ecosystems.

6. The Future Engine of AI: The Rise of Reinforcement Learning and “Test-Time Computing”

Dan Roberts of OpenAI emphasized in his speech that the strategic focus of artificial intelligence is shifting from traditional pre-training computing that relies on large-scale data to reinforcement learning ( RL ) and "test-time computing" . He believes that future AI models will significantly improve their performance and ability to solve complex problems by investing more time and computing resources in "thinking" and reasoning when receiving tasks or problems (i.e., the testing phase). This represents a new dimension of AI's ability to expand. The common perception of "pre-training is the cake and reinforcement learning is the cherry" in the past may be completely overturned, and reinforcement learning may even dominate the allocation of computing resources in the future. This means that the development of AI models will not only rely on "rote memorization" of massive data, but also emphasize improving the level of intelligence and output quality through human-like "deliberate thinking", multi-step reasoning, and dynamic interaction with the environment during the execution of specific tasks.

7. AI Programming Revolution: Subverting Development Processes and Forcing Organizations to Improve Efficiency

Mike, Anthropic's Chief Product Officer, pointed out at the summit that efficient AI programming tools (such as Anthropic's Claude model for programming) are like a "magnifying glass" that will make the inefficiency of other links within the enterprise organization more prominent and unbearable. When the speed of code generation and iteration is greatly improved due to AI assistance, traditional, lengthy coordination meetings, communication barriers, and slow decision-making processes will become a greater bottleneck restricting the release of overall productivity. AI is not only profoundly changing the way individual engineers program - for example, Mike mentioned that more than 70% of the code in pull requests within Anthropic is generated by Claude - but also forcing companies to fundamentally rethink their product development processes, team collaboration models, and even the overall organizational structure. In order to fully utilize the efficiency dividends brought by AI programming, organizations must make corresponding adjustments and changes to adapt to this new, high-speed development rhythm.

8. “Environmental Intelligence”: Beyond Chat Interaction, Achieving Seamless Integration

LangChain CEO Harrison Chase proposed the emerging concept of "Ambient Agents" at the summit and distinguished it from traditional chat agents. Unlike chatbots that require users to actively initiate interactions, "Ambient Agents" can passively listen to event streams (such as new emails, calendar reminders, system notifications, etc.), run autonomously in the background, and may handle multiple concurrent events at the same time, and the requirements for interaction delay are relatively lower. This indicates that the application form of AI will gradually evolve from the current "chat partner" with explicit dialogue as the main interaction method to an "invisible assistant" that silently assists users in handling daily affairs and managing workflows in the background. For example, an email ambient agent can automatically listen to received emails and try to reply, arrange meetings or remind relevant personnel accordingly. Although this type of agent runs in the background, human-computer collaboration is still crucial, and the interaction mode will change to users approving or rejecting the agent's action suggestions, editing the agent's operations, answering questions for the agent when it encounters obstacles, and reviewing and correcting the decisions in the agent's historical steps through the "time travel" function.

9. Deep Research: AI-driven deep information analysis and insight engine

OpenAI's Esa detailed its Deep Research capabilities at the summit. This is not a simple information retrieval tool, but an AI system that can browse massive online resources, perform deep reasoning, and generate detailed reports with comprehensive citations comparable to professional research analysts in 5 to 30 minutes. Deep Research represents a major advance in AI in the field of information acquisition, processing, and analysis. It achieves its powerful functions through a version of the O3 model that is fine-tuned specifically for optimizing web browsing and data analysis. The original intention of its development was to hope that the model could directly learn and perform actual tasks in users' daily lives and work. At present, Deep Research can efficiently complete complex tasks that traditional manual research would take hours or even longer to complete. Future development directions include continuously improving the reliability of the model, integrating its capabilities into the core reasoning model, and introducing private data sources such as the company's internal knowledge base. The more groundbreaking goal is to allow Deep Research to evolve from the current information integration and report generation to the stage where it can "take action" autonomously.

10. The game and future of open source AI: huge potential, “decentralization” may be the key to sustainable development

Regarding the future of open source AI, the panel discussion at the summit pointed out that although open source models (such as Llama, DeepSeek, etc.) currently account for only about 20% to 30% of the overall usage of inference tokens and have not yet become the absolute mainstream, their inherent transparency and the ability to bring together the talents of developers around the world give them great development potential. The sudden rise of the Chinese model DeepSeek also confirms the vitality of the open source AI field and the dynamic nature of the possible changes in leaders. The development of open source AI is not smooth sailing. It faces many challenges in terms of computing power acquisition, business model construction, and coping with potential regulation. Some guests pointedly pointed out that if the open source community fails to successfully establish a sustainable, decentralized open source model service and inference provider system that is not monopolized by a few large technology companies, then the future may still be dominated by closed source models. Therefore, "decentralization" is considered to be one of the key factors to ensure the healthy and sustainable development of the open source AI ecosystem. It can motivate more participants to contribute and thus change the current rules of the game.

The ten insights presented at the 2025 Sequoia Capital AI Summit clearly outline the grand blueprint for the accelerated evolution of artificial intelligence technology, profound changes in the industrial landscape, and the intelligent transformation of future society. Whether it is a giant leading technological breakthroughs, an innovator working hard in the application field, or everyone who pays attention to this era of change, they all need to deeply understand these trends and actively embrace changes in order to gain insight into opportunities and grasp the future in this magnificent wave of artificial intelligence.

(Click below to view the original text and view all the content of the Sequoia AI Summit )

Creation is not easy, every attention , like , repost and read is the greatest support for me. Thank you for your support! #AI #人工智能#红杉财富#红杉 #智能体 #openai