The AI opportunity: Sequoia Capital's AI Ascent 2024 opening remarks

Macro Opportunity and Historical Context

• Sequoia Capital positions Generative AI as the start of the "plateau of productivity" following the hype cycle's peak and trough of disillusionment.
• AI is defined by three distinct capabilities: creation (images, text, video), reasoning (multi-step agentic logic), and human-like interaction.
• The team draws a parallel between the current AI shift and the cloud transition (2010–2024), noting cloud software revenue grew from $6B to $400B at a 40% CAGR.
• Unlike the cloud transition which replaced software with software, AI is projected to replace services with software, potentially unlocking a Total Addressable Market (TAM) in the tens of trillions.
• Sequoia hypothesizes this represents the "single greatest value creation opportunity mankind has ever known."
• The current AI wave is additive to previous tech eras (transistors, PCs, Internet, Cloud, Mobile), with necessary ingredients finally becoming available: cheap/computing, efficient networks, mobile connectivity for 7B people, and abundant data.
• Historical analysis suggests the next decade will be defined by AI, following a pattern of technology waves that began in the 1960s and evolved through the 2000s.

Current Market Traction and Financials

• Generative AI has reached $3 billion in aggregate revenue in its first year of public availability, a milestone that took the SaaS market nearly a decade to achieve.
• Klarna has implemented OpenAI to handle two-thirds of customer service inquiries, effectively automating the work of 700 full-time agents.
• AI has found product-market fit in legal services (e.g., Harvey automating legal analysis) and software engineering (moving from code assistance to self-contained AI software engineers).
• Despite high user engagement, the funding environment has been inverted: over $50 billion was spent on NVIDIA GPUs in 2024, while only ~$3 billion in revenue has materialized.
• Current AI usage metrics show low retention compared to mobile peers; usage is currently sporadic ("once a week/month") rather than daily.
• User churn is primarily driven by a gap between expectations of "magical" performance and the reality of model reliability.

Predictions for 2024 and Near-Term Evolution

• Prediction 1: 2024 will shift AI from "co-pilots" (assistive tools) to "agents" that can execute tasks entirely without human intervention, particularly in software engineering and customer service.
• Prediction 2: New research in "inference time compute" and value iteration will enable models to reason and plan, moving beyond pattern parrotting to handle complex cognitive tasks.
• Prediction 3: A migration from consumer/prosumer apps to high-stakes enterprise applications (healthcare, defense) driven by techniques like RLHF, prompt chaining, and vector databases to achieve five-nines reliability.
• Prediction 4: As prototypes move to production, priority will shift to latency, cost, and model/data ownership, causing a shift in compute balance from pre-training to inference.
• Prediction 5: The "one-person company" will emerge as a viable business model, leveraging AI networks to manage complex workflows previously requiring large teams.
• Prediction 6: Companies like Klarna will increasingly use AI to drive specific KPIs, eventually causing entire organizational structures to function as interconnected neural networks.

Long-Term Societal and Economic Impact

• AI is categorized as a productivity revolution following a historical pattern: human + tool → human + machine assistant → human + machine network.
• Historical data shows AI-driven productivity leads to deflation in costs for critical sectors (software, electronics) and could similarly reduce costs in education, healthcare, and housing.
• The cost per revenue unit for S&P 500 companies is expected to decline rapidly as AI integration accelerates.
• Computing architecture is shifting from rote pixel storage to "conceptual" rendering, where data is understood as multidimensional points (Platonic forms) rather than binary matrices.
• Future interfaces will be generated and contextualized rather than static, allowing computers to understand the broader meaning of data (e.g., "why" a concept is presented, not just "what" it is).
• The ultimate trajectory envisions entire companies functioning as interoperable, self-optimizing neural networks, breaking down silos between customer support, growth engines, and personalization.
• The goal of this technological shift is to enable "more problems can be tackled by more people" through massive efficiency gains.