May 6, 2025, 8:00 PM ETConference Presentation, Keynote, Panel

AI's Trillion-Dollar Opportunity: Sequoia AI Ascent 2025 Keynote

The AI opportunity encompasses both the services market (projected to be an order of magnitude larger than the 400 billion global software market at the cloud transition's start) and the software market.
Sequoia updated its view that AI will disrupt both profit pools, transforming business models from selling tools/software budgets to selling outcomes/labor budgets.
The "Don Valentine" framework for evaluating AI is structured around: What (it's imminent, not just inevitable), So What (market magnitude), Why Now (infrastructure readiness), and What Now (strategy).
Infrastructure readiness (compute, networks, data, distribution, talent) and additive technological waves position AI to scale faster than previous transitions like cloud or mobile.
Distribution physics have fundamentally shifted: unlike the cloud era where marketing was required to generate awareness, AI achieved immediate global attention (ChatGPT, Nov 2022) across 1.2–1.8 billion social media users and 5.6 billion internet households.
There is significant "white space" in the application layer, though competition is intensifying as foundation models push deeper into verticals via test-time compute and reasoning.
Sequoia's investment thesis prioritizes revenue and free cash flow over "unicorn" valuations, focusing on companies that reached $1B+ in revenue during prior tech transitions.

Investors distinguish between "vibe revenue" (tire-kicking) and durable behavior change, requiring proof of adoption, engagement, and retention.
Trust is prioritized as the primary asset; customer trust in the company's ability to evolve the product supersedes current product perfection.
Gross margins are expected to improve over time as COGS (cost per token) decline by ~99% in 12–18 months, while pricing power increases as companies move up the value chain to selling outcomes.
A functional "data flywheel" must tie directly to a specific business metric; otherwise, it is considered non-existent or irrelevant for building a moat.
The "95/5 rule" for building AI companies states that 95% involves standard startup fundamentals, while the remaining 5% focuses on AI-specific execution like end-to-end solutions and industry-specific language.
Competitive moats are best built by addressing complex problems with human-in-the-loop processes and developing industry-specific solutions (e.g., Harvey for law, OpenEvidence for medicine).

AI-native applications have seen a dramatic shift in engagement metrics, with ChatGPT's daily-to-monthly active user ratio approaching Reddit levels, signaling a move from hype to utility.
AI applications are expanding into deeper verticals including advertising (copy creation), education (concept visualization), and healthcare (diagnostics).
Voice generation technology has crossed the "uncanny valley," with demos from Sesame and others challenging the Turing test.
Software coding has emerged as the first category to achieve "screaming product market fit," with models like Claude 3.5 Sonnet drastically improving software creation accessibility and economics.
Pre-training scaling has slowed, but breakthroughs in reasoning (OpenAI o3), synthetic data, tool use, and agentic scaffolding are driving new intelligence scaling.
Innovation is blurring the lines between research and product, highlighted by breakthroughs like Deep Research and Notebook LM.
The application layer is identified as the primary locus of value accrual, with the first cohort of "killer apps" (e.g., ChatGPT, Harvey, Glean, Cursor) already established.

The next wave of AI development is predicted to evolve from isolated agents to "agent swarms" and eventually a full "agent economy" where agents transfer resources, execute transactions, and manage trust.
A critical technical challenge for the agent economy is "persistent identity," requiring agents to maintain consistent personality/understanding over time and retain memory of user interactions.
Seamless communication protocols (e.g., Model Context Protocol/MCP) are identified as the necessary infrastructure for transferring information, value, and trust between agents.
Security and trust verification will become a dominant cottage industry as human-to-agent interactions require new standards of reliability.
The economy is shifting toward a "stochastic mindset," replacing deterministic programming with probabilistic outcomes that require new management strategies to handle uncertainty.
Management roles will evolve to focus on blocking processes and providing feedback to agents rather than traditional oversight, potentially creating a "one-person unicorn" era.
The ultimate trajectory involves a "neural network within neural networks," where individual functions merge into complex agent clusters that reinvent work, corporate structure, and the broader economy.