The New Rule for Picking AI Winners | The a16z Show

Revenue Growth & Market Scale

• Anthropic and OpenAI are currently generating more monthly revenue growth than Meta, Google, and Microsoft combined.
• The combined revenue run rate of Anthropic and OpenAI is projected to reach $200 billion by the end of this year.
• The revenue growth of these two entities alone is expected to exceed the total revenue added by the entire public software universe.
• The collective profit of Fortune 500 companies is approximately $2 trillion annually, suggesting a massive total addressable market for AI adoption.

Valuation Trends & Exit Thresholds

• The threshold for the top 1% of venture exits has increased tenfold over the last 24 months, moving from $10 billion in 2020 to $32 billion as of yesterday's update.
• Recent exit data shows a doubling of top-tier exit values approximately every two years, with February 2025 updates noting a $20 billion threshold.
• There is a possibility that combined entity valuations for leading AI firms could exceed $100 billion by September.
• 80% of companies in the current AI universe are likely overvalued, while a small subset of market leaders are potentially undervalued.

Technology Adoption & Diffusion

• Actual technology diffusion into the broader economy remains under 5%, despite high adoption rates in coding and tech-forward sectors.
• Current enterprise usage is largely "skeuomorphic," focusing on efficiency gains for existing workflows rather than native AI-driven business model changes.
• Usage takeoff in new verticals like legal is beginning to mirror the rapid adoption previously seen in software development.
• Most resource allocation in leading companies is currently directed toward product innovation rather than internal process automation.

Value Capture & Business Model Evolution

• The "half-life" of AI startups is shrinking rapidly, with 40% of companies on last year's Forbes AI 50 list dropping off by the current year.
• Venture capital strategies are shifting toward backing market leaders at the earliest stage, accepting higher loss ratios to capture the "power law" upside.
• Defensibility is increasingly tied to securing a position in the "token path" and managing relationships with frontier model providers.
• Open source and local model adoption is gaining importance due to immediate cost pressures, potentially accelerating faster than anticipated.
• Competition among frontier labs will determine token pricing; fewer competitors may lead to higher prices, while increased competition could drive prices down 10x year-over-year.

Supply Chain Constraints & Bubble Analysis

• The market is currently supply-constrained rather than demand-constrained, with data center capacity expected to remain tight until late 2028 or early 2029.
• The speaker asserts there is no current bubble, citing a lack of excess supply that would typically destroy unit economics in bubble scenarios.
• A potential bubble could only materialize if an algorithmic breakthrough drastically reduces model size and token consumption, creating an oversupply of intelligence.
• Global data center build-outs are estimated to require $5 trillion in capital expenditure, potentially yielding $1–2 trillion in revenue.

Public Markets & Future Industry Structure

• The inclusion of hyper-growth AI companies in public market indexes is expected to revitalize the public market ecosystem, which has seen a 50% reduction in the number of listed companies over 20 years.
• Palantir is identified as the only major software company currently growing at 70%+ rates, contrasting with the sub-30% growth of other "Mag 7" peers.
• Venture firms are scaling early-stage platforms to support companies that face "big company problems" (e.g., international expansion, complex cloud deals) much earlier than in previous cycles.
• Future VC industry value creation over the next five years will depend heavily on the market structure of model labs, the role of open source, and the volume of token consumption.

Global Competitive Landscape

• Leading Chinese LLMs are approximately six months behind US counterparts in capability but are priced 10x cheaper, presenting a classic innovator's dilemma scenario.
• There is significant uncertainty regarding the market share capture of lower-tier models versus the insatiable demand for frontier intelligence.
• The speaker notes that while 80% of companies may fail, the historical track record of diversified portfolios suggests long-term value appreciation even when the majority of bets fail.

Forward-Looking Statements

• The speaker predicts that the next generation of companies will be significantly larger than predecessors, with the largest firms potentially becoming the dominant economic forces of the next decade.
• A shift from reactive to proactive AI engagement is expected to occur in both consumer and enterprise sectors within the next 12 months.
• Consumer attention markets are expected to undergo a significant transformation as AI tools reduce the time spent on routine tasks, creating new opportunities for VC-backed consumer applications.
• The industry is currently in a phase of rapid learning, with the speaker admitting to changing their priors on value capture and scale faster than ever before in their career.