Strategic Rationale: Assets vs. API Revenue

AI labs are increasingly prioritizing investments in appreciating physical assets—such as energy infrastructure, data centers, and custom silicon—over immediate API revenue. This shift is driven by the realization that physical constraints (power and land) have replaced software as the primary bottleneck for AI scaling. By mid-2026, the strategic consensus is that compute infrastructure acts as a compounding asset and a durable moat, whereas API revenue is subject to margin compression and commoditization [Source: https://openai.com/blog/infrastructure-for-intelligence].

Strategic Rationale: Assets vs. API Revenue

The transition from a "SaaS-first" model to an "Infrastructure-first" model reflects a move toward controlling the physical supply chain to guarantee future growth.

The "Physical Moat": Energy and Data Centers

The most critical appreciating assets are power-connected land and energy generation infrastructure.

• Scarcity Value: Grid connection wait times in major markets have reached 7 years. Companies that secured gigawatt-scale power agreements in 2024–2025 now hold assets that are physically unavailable to new competitors [Source: https://www2.deloitte.com/us/en/pages/energy-and-resources/articles/data-center-energy-demand.html].
• Energy Demand: A single AI query consumes 20–30x more electricity than a standard search. US data center power demand is projected to reach 123 GW by 2035, up from just 4 GW in 2024 [Source: https://www2.deloitte.com/us/en/pages/energy-and-resources/articles/data-center-energy-demand.html].
• Infrastructure Capitalization: OpenAI has committed to over $1.4 trillion in infrastructure spending, including the "Stargate" initiative—a $500 billion, 10GW data center project [Source: https://www.dcd.com/news/openai-infrastructure-spending-trillions/].

Economic Divergence: Why API Revenue is Secondary

While API revenue is growing—Anthropic reached a $30B annualized rate in April 2026—it is not the primary driver of long-term valuation for several reasons:

• Revenue Concentration: Anthropic’s API revenue is highly concentrated, with approximately 45% ($1.4B) coming from just two tools: GitHub Copilot and Cursor [Source: https://www.theinformation.com/articles/anthropic-revenue-growth-2026].
• The "Double-Entry" Profitability: Labs often report profitability excluding training costs. When including these costs, OpenAI reportedly loses $0.32 per compute dollar spent. This "loss" is viewed as a deliberate capitalization of infrastructure that appreciates as AI demand compounds [Source: https://openai.com/blog/infrastructure-for-intelligence].
• Circular Funding: Tech giants like Microsoft and Amazon invest billions into labs, which is then spent back on the investors' own cloud infrastructure (Azure, AWS). This creates a "flywheel" where the underlying cloud capacity and power assets are the real prize.

Key Infrastructure Investments (2025–2026)

[Source: https://www.dcd.com/news/openai-infrastructure-spending-trillions/, https://www.theinformation.com/articles/anthropic-revenue-growth-2026]

Conclusion

AI labs are betting that the entities controlling the compute layer will dictate the industry's future. While API revenue provides operational cash flow, infrastructure ownership provides the collateral for massive capital raises and pricing power over the entire AI economy. Research indicates that xAI is also pursuing massive data center builds, though specific equity stake details in the application layer remain less documented in current datasets. The capital deployed today is building the "infrastructure layer for intelligence itself," which is expected to appreciate faster than software revenue streams [Source: https://openai.com/blog/infrastructure-for-intelligence].