Sequoia AI Ascent 2025 Keynote: Unpacking AI's Trillion-Dollar Market Opportunity, the Race for the Application Layer, and the Future Agent Economy

Sequoia Capital

Summary:

This keynote from Sequoia AI Ascent 2025 outlines AI's market opportunity, which is estimated to be at least 10 times larger than cloud computing, impacting both services and software. The speakers explain why AI's current boom is happening now, citing the established technological foundation and accelerated distribution mechanisms, such as widespread internet connectivity and social media platforms. They advise startups to focus on the application layer by building moats through customer-centric solutions, leveraging unique data flywheels, and specializing in specific industry verticals. Key metrics for success include durable revenue (not just "vibe revenue"), a path to healthy gross margins, and data flywheels that directly impact business metrics. The presentation also predicts the rise of an "agent economy" where AI agents will transact and collaborate, emphasizing the need for persistent identity, seamless communication protocols, and robust security. This future requires adopting a "stochastic mindset" and operating with maximum velocity.

AI Ascent 2025 Opening Remarks and Framework [00:00:00]

Sequoia Capital partners Pat Grady, Sonya Huang, and Constantine Ostrovsky introduce the AI Ascent 2025 keynote, setting the stage to share perspectives learned over the past year in AI. They utilize a simple framework for market analysis: "What?", "So What?", "Why Now?", and "What Now?". The focus for the session will be on updating their thinking regarding the "So What?", "Why Now?", and "What Now?" aspects, assuming the audience already understands "What AI is."

The Enormous Market Opportunity of AI [00:01:27]

The "So What?" section highlights AI's massive market potential.

• Market Size Comparison: AI represents a market opportunity at least an order of magnitude (10x) larger than the cloud computing transition.
  • The cloud market reached $400 billion in revenue, exceeding the global software market at its inception.
  • AI services are starting from a much larger base, with a potential endpoint 10-20 years from now being "absolutely massive."

    

• Attack on Profit Pools: AI is impacting both services and software profit pools.
  • Companies are evolving from selling tools (software budget) to selling outcomes (labor budget), progressing from co-pilots to autopilots.
  • Both the total addressable markets (TAMs) for software and services are open for disruption.

Why AI is Happening Now [00:03:04]

The "Why Now?" section explains the imminence of AI's widespread adoption.

• Technological Foundation: AI is imminent, not just inevitable, because all the precedent conditions are in place.
  • This includes existing infrastructure for compute, networks, data, distribution, and talent.

    

• Accelerated Pace of Innovation: Things are happening faster than ever before.
  • This acceleration is driven by the "New Physics of Distribution," which requires three elements for widespread adoption:
    • Awareness: The launch of ChatGPT on November 30, 2022, instantly garnered global attention for AI. In contrast, the cloud transition required extensive "gorilla marketing."
    • Desire: Combined monthly active users on platforms like Reddit and the artist formerly known as Twitter (X) are between 1.2 and 1.8 billion, providing channels for discovering new technologies. These platforms barely existed at the start of prior transitions.
    • Action/Accessibility: There are now 5.6 billion internet users, effectively covering every household and business globally, compared to only 200 million during the cloud transition.

      

• Market Vacuum: The "rails are in place," meaning there are no significant barriers to AI adoption. This creates a "tremendous sucking sound" in the market.
  • Macroeconomic factors like tariffs and interest rates are considered noise compared to the rising tide of technology adoption.
  • Startups must move at maximum velocity to fill this vacuum before others do.

    

Strategies for Winning in the AI Landscape [00:05:32]

The "What Now?" section focuses on how startups can play to win in the AI era.

• The Race for the Application Layer: The primary value accrual is at the application layer.
  • While there was significant "white space" last year, the market is filling up, but opportunities remain.

    

  • Companies that achieved billion-plus revenues in prior transitions often did so at the application layer.
  • Foundation models are increasingly competing by expanding into the application layer through reasoning, tool use, and inter-agent communication.

    

• Building Moats Across the Merchandising Cycle (Customer-Back Approach): Startups not building vertically integrated businesses should adopt a customer-back approach to build defensible positions ("moats").
  • Vision (Opinionated): Provide clear, opinionated solutions, as customers may not know what they want from AI.
  • Product (Solution): Offer end-to-end solutions that solve problems comprehensively, rather than just throwing a tool over the fence.
  • Engineering (Data Flywheels): Build data flywheels based on the usage data of your own product, creating unique, proprietary advantages that others lack.
  • Marketing (Domain-Specific Trust): Be "of the industry, for the industry" (e.g., Open Evidence for medical, Harvey for legal) to build deep domain-specific trust.
  • Sales (Lingua Franca / Account Control): Speak the customer's language and establish strong account control, which foundation models are less likely to achieve.
  • Support (FDE's / Talk to the Bot): Offer comprehensive support, potentially through sophisticated bots, to embrace customers fully.

    

Key Metrics for AI Companies [00:09:01]

Key metrics that matter for AI companies are distinct from traditional software.

• Revenue:
  • Beware "Vibe Revenue": Initial excitement or "tire kicking" might not translate to durable behavior change.
  • Inspect Adoption, Engagement, Retention: True revenue reflects genuine, sustained product usage and customer trust. Trust is paramount; customers will wait for product improvements if they trust the company.
• Margins:
  • Slope, Not Intercept: Current gross margins may be low, but the critical factor is the slope of the margin curve. Costs of goods sold (COGS), like cost per token, are expected to continue decreasing.
  • Value Capture: As companies move from selling tools to outcomes, their price points should increase, leading to healthier gross margins over time.
• Data Flywheel:
  • Must Move a Business Metric: A data flywheel is only valuable if it directly impacts and improves a core business metric (e.g., better retention, higher conversion).
  • If a data flywheel doesn't move a business metric, "either you don't have a data flywheel or it just doesn't matter."
  • This is considered one of the best moats to build.

    

The Evolution of AI and the Agent Economy [00:12:37]

AI applications are rapidly advancing and paving the way for a new economic paradigm.

• AI App Engagement:
  • In 2023, AI apps had poor engagement ratios (Daily Active Users/Monthly Active Users), suggesting hype exceeded reality.
  • By 2024, this has dramatically improved, with apps like ChatGPT showing DAU/MAU ratios climbing towards levels seen in established platforms like Reddit.
  • This indicates increasing value derived from AI by users.

    

• Deeper Applications and Breakthroughs:
  • Initial usage was fun and viral (e.g., "jiblify everything").
  • Now, deeper applications are emerging across various sectors:
    • Advertising: Creating highly accurate and aesthetically pleasing ad copy.
    • Education: Visualizing complex concepts instantly.
    • Healthcare: Aiding in patient diagnosis (e.g., Open Evidence).

      

  • "Her" Moment for Voice: Voice generation has crossed the "uncanny valley," with AI voices becoming indistinguishable from human voices, rapidly closing the gap between science fiction and reality.

    

  • Coding as a Breakout Category: AI coding has achieved "screaming product market fit," exemplified by Anthropic's Claude 3.5 Sonnet.
    • It's fundamentally changing the accessibility, speed, and economics of software creation.
    • Users are "vibe coding" their own alternatives to existing tools.

      

• New Vectors for AI Progress (Technology Out):
  • While pre-training scaling is slowing down, new research breakthroughs are emerging to scale intelligence.
  • Key vectors include: Reasoning Models, Synthetic Data, Tool Use, and Agentic Scaffolding.

    

  • These combine to make AI capable of increasingly sophisticated tasks, as measured by benchmarks like "Meter."

    

  • Innovation is particularly strong at the "product-model boundary," with examples like OpenAI's Deep Research and Google's NotebookLM.

    

• The Battleground: Application Layer (Reaffirmed):
  • Despite initial debates, the application layer remains where value will ultimately accrue, with foundation models increasingly competing here.
  • The first "cohort of winning AI applications" has emerged (e.g., ChatGPT, Harvey, Glean), with a "next cohort on the rise" across diverse markets.

    

• Agents: From Prototypes to Hardened Systems:
  • Many new companies will be "agent-first," evolving agents from pieced-together prototypes to robust systems.
  • Two paths for building robust agents:
    1. Orchestration with rigorous testing and evaluations.
    2. Agents trained and fine-tuned end-to-end on specific tasks.

       

• Vertical Agents: Deeper Tech, Deeper Customer Value:
  • Vertical agents represent a significant opportunity for founders with deep domain understanding.
  • These agents are trained end-to-end for specific workflows using reinforcement learning on synthetic and user data.
  • Evidence suggests vertical agents can outperform humans in specific tasks:
    • Cybersecurity: XBow's agent can outperform human penetration testers.
    • DevOps: Traversal's AI troubleshooter is better than the best human troubleshooters.
    • Networking: Meter's agents outperform networking engineers.

      

• Entering an Abundance Era: Code is the first market category to "tip" into an abundance era.
  • This offers a preview of what happens when labor becomes cheap and plentiful, making "taste" the scarce asset.
  • The continued progress of coding agents will redefine the technology landscape and serve as a harbinger for other industries transformed by AI.

What's Next: The Agent Economy and Mindset Shifts [00:20:28]

Constantine Ostrovsky discusses the mid and long-term predictions for AI, focusing on the agent economy and its implications.

• The Agent Economy:
  • A year ago, AI Ascent focused on agents as machine assistants evolving into "machine networks" (now called "agent swarms").

    

  • This will mature into an "agent economy" where agents don't just communicate information but also transfer resources, make transactions, and understand trust and reliability.
  • This economy is human-centric, with agents and humans working collaboratively.

    

• Technical Challenges to Achieve the Agent Economy:
  • The speaker highlights three key technical challenges.

    

  • 1. Persistent Identity: Agents need to maintain a consistent personality and understand individual users over time, addressing challenges in true memory and self-learning.

    

  • 2. Seamless Communication Protocols: The development of foundational protocols (like TCP/IP for the internet) is crucial for agents to effectively transfer information, value, and trust.

    

  • 3. Secure Trust: With face-to-face interaction absent, security and trust mechanisms for agents become even more critical, fostering a new "cottage industry" around this.

    

• Living in the Agent Economy: Mindset Shifts:
  • 1. Stochastic Mindset: A departure from deterministic computing, requiring managing a distribution of outcomes rather than a single predictable result. Human and AI behavior will be less precise, like remembering "around 73" instead of exactly "73."

    

  • 2. Management as a Computer in an Agent Swarm: The role of human managers will shift from direct supervision to understanding what agents can and cannot do, providing higher-level managerial decisions like blocking processes and giving feedback.

    

  • 3. Leverage Over Uncertainty: This new era offers significantly more leverage on tasks (100% leverage), but with far less certainty on the exact manifestation of the outcome. Success requires managing this uncertainty and associated risks.

    

  • They have seen companies scale faster than ever before with fewer people.

    

• Reinventing the Economy: AI agents will continue to merge and form complex "neural networks of neural networks," leading to unprecedented levels of leverage.
  • This will reinvent individual work, rewire companies, and fundamentally recreate the entire economy.