The Economics of Generative AI: Two Years Later-Value Chain Still Inverted

The generative artificial intelligence (AI) industry has experienced significant growth over the past two years, with annualized revenue expanding from approximately $90 billion to about $435 billion. Despite this substantial expansion, the economic structure of the generative AI value chain remains largely unchanged, emphasizing the dominance of specific layers and highlighting strategic opportunities for investors and businesses.

The Value Chain is Still Inverted

Two years ago, I noted that the compute layer captured approximately 83% of all revenue and 87% of gross profit within the generative AI ecosystem. This trend has persisted, with semiconductor companies (Semi) and infrastructure providers (Infra) continuing to dominate the value chain, while application developers (Apps) remain marginalized.

Semiconductor Layer: A One-Player Game

The semiconductor layer is overwhelmingly dominated by NVIDIA, which reported a data center business revenue of $62 billion in the last quarter, annualizing to approximately $250 billion. Broadcom’s AI semiconductor business, which includes custom accelerators for major tech companies like Google, Meta, and ByteDance, contributes an additional $34 billion. High bandwidth memory purchased directly by hyperscalers for custom chip programs adds another estimated $25 billion. The concentration is extreme, with NVIDIA accounting for about 80% of the semiconductor layer's revenue.

Infrastructure Layer: A Competitive Landscape

The infrastructure layer, which includes AI-specific cloud services, has grown to approximately $75 billion in annualized revenue. Major players such as Azure, AWS, GCP, and Oracle each contribute between $10 billion and $20 billion in AI-attributable revenue. CoreWeave adds about $6 billion, while Baseten, Together, Modal, and other inference providers make up the remainder. Unlike the semiconductor layer, this segment is relatively evenly distributed across the major cloud providers, making it a more competitive space.

Application Layer: A Two-Player Game

The application layer, which directly serves end customers, has grown to about $60 billion in annualized revenue. However, despite its proximity to users, this layer remains highly concentrated and less profitable compared to the semiconductor and infrastructure layers. The dominance of a few major players in the application space suggests that it is a challenging market for new entrants.

Key Risks

1. Market Concentration: The extreme concentration in the semiconductor and infrastructure layers poses significant risks for smaller players and new entrants. Companies outside these dominant positions may struggle to gain traction or profitability.
2. Technological Advancements: Rapid advancements in AI technology could disrupt current market leaders, particularly if breakthroughs occur that reduce the reliance on high-end semiconductors or cloud infrastructure.
3. Regulatory Scrutiny: Increased regulatory scrutiny of AI and data privacy could impact the business models of companies across all layers of the value chain, potentially leading to higher compliance costs and operational challenges.

The Opportunity

1. Diversification in Infrastructure: While the semiconductor layer is dominated by NVIDIA, the infrastructure layer offers more opportunities for diversification. Companies that can innovate in cloud services, particularly those focused on specialized AI applications, may find significant growth potential.
2. Application Layer Innovation: Despite its current challenges, the application layer remains a critical component of the generative AI ecosystem. Companies that can develop innovative and user-friendly applications with strong value propositions have the potential to capture market share and drive revenue.
3. Strategic Partnerships: Forming strategic partnerships with key players in the semiconductor and infrastructure layers can provide smaller companies with the resources and technology needed to compete effectively.

Conclusion

The generative AI industry's economic structure remains inverted, with semiconductors and infrastructure continuing to capture the majority of the value. While this presents challenges for new entrants, it also highlights opportunities for strategic investments and innovation in less concentrated segments. As the market continues to evolve, companies that can navigate these dynamics effectively will be well-positioned for long-term success.