Goldman Sachs: AI Enters Execution Era, Competition Shifts to Workflows
TREE NEWS reports: After a two-day tour of Silicon Valley’s AI ecosystem, Goldman Sachs has concluded that the industry is moving from models that ‘answer’ to agents that ‘execute.’ The shift, detailed in a new report, signals that AI commercialization is evolving from per-seat subscriptions to consumption-based and outcome-based pricing, while value migrates from model weights to proprietary data, business context, and domain expertise.
What Happened
Goldman’s 8th-9th August visits to AI startups, leading venture funds, and researchers at Stanford, UC Berkeley, and UC San Francisco revealed a consensus: enterprises are less constrained by model capability than by controllability. In legal, risk, insurance, and audit workflows, questions of responsibility, traceability, and error correction are as critical as raw intelligence. The most automatable workflows—like invoice processing—share three traits: clear decision boundaries, verifiable outcomes, and rollback capability.
Market Implications
Frontier vs. open-source models: The report suggests a division of labor. Frontier models will handle high-value, high-reliability tasks where errors are costly, while open-source models will dominate the bulk of inference tokens. One VC firm predicts 90% of inference tokens will flow to open-source models within 12-18 months.
World models as the next growth curve: Researchers are shifting focus from LLMs to world models that understand physics, causality, and dynamic interaction. These require far more compute and proprietary data, potentially extending the AI infrastructure buildout. Goldman estimates compute demand could grow 24x over five years, benefiting cloud providers like Microsoft, Oracle, and CoreWeave.
Value chain redistribution: As agents move into production, value shifts from model providers to those owning trusted content, validated domain models, and regulatory relationships. Information service providers with these assets will likely penetrate enterprise environments first.
Key Takeaways for Investors
- Monitor AI pricing models—consumption and outcome-based pricing will disrupt software economics.
- Watch for winners among enterprise software firms with deep domain data and workflow integration.
- Compute demand remains structurally bullish for cloud and GPU infrastructure players.
- Open-source model adoption will pressure pricing for commoditized AI services but expand the market.
