For editorial information only. Not investment advice.
Key takeaway: The AI race is shifting from model ownership to bottleneck ownership: power, grid access, cooling, land, chips and industrial execution. That is where sovereign wealth funds, public pensions and other long-duration allocators have a structural edge.
For most of the AI boom, markets treated artificial intelligence as software: scalable, asset-light and close to zero marginal cost. That framing is breaking. The next phase of AI looks less like SaaS and more like heavy industry — dependent on power, cooling, land, transmission, chips and sovereign-scale balance sheets. For long-term asset owners, the strategic question is no longer simply who owns the best model. It is who owns the bottleneck.
Three Structural Infrastructure Observations
The capex wall. AI infrastructure is turning hyperscale economics into a heavier capital-expenditure model. Alphabet spent $52.5 billion on capital expenditures in 2024 and $91.4 billion in 2025, an increase of roughly 74%. The company frames this spending as technical infrastructure — servers, network equipment and data centers — to support growth, including AI products and services. The key point for allocators is not that one company spent aggressively in one year; it is that the AI stack requires repeated reinvestment across chips, servers, networking, real estate and energy systems.
The watt wall. Compute scaling is becoming a grid-allocation problem. The IEA projects global data center electricity consumption will roughly double to about 945 TWh by 2030, with data center electricity demand growing at about 15% per year from 2024 to 2030. In the United States, the IEA expects data centers to drive around half of total electricity-demand growth through 2030. Ireland is the live stress test: data centers consumed 6,969 GWh in 2024, equal to 22% of total metered electricity consumption and more than urban households at 18%.
The industrial AI pivot. Asia is not only chasing consumer chatbots. Japan and China are directing AI into the physical economy. METI’s FY2026 AI-and-semiconductor-related budget rose to JPY 1.239 trillion, up from JPY 332.8 billion in FY2025, including JPY 387.3 billion for AI foundation models, data infrastructure, AI robots and physical AI. That push is reinforced by demographics: Japan’s working-age population was 73.73 million in 2024, only 59.6% of the total population. China’s AI Plus initiative similarly aims to integrate AI deeply across industry, the economy and society.
Implications for Long-Horizon Asset Owners
Own bottlenecks, not just narratives. For sovereign capital, pensions and insurers, the investable question is shifting from model ownership to bottleneck ownership. The most durable scarcity may sit in baseload and dispatchable power, transmission, grid interconnection, cooling, land, permitting expertise and industrial operators with privileged energy access. These assets can hold pricing power, but they must be underwritten with regulatory, permitting, community and commodity-price risk in mind.
Re-underwrite technology cash flows. AI winners may still compound, but the assumption that software economics remain structurally asset-light deserves a fresh stress test. As infrastructure intensity rises, free cash flow quality matters more. Public-market multiples already reflect a harsher separation between scarce compounders and the rest of software: Meritech noted that only eight public software companies traded above 10x next-twelve-month revenue, while 73% traded below 5x.
Look for operator alpha in real assets. The next phase of AI diligence should not stop at whether a company has an AI strategy. Allocators should ask whether management can convert compute into measurable operating gains: lower energy intensity, higher equipment uptime, faster logistics, better safety, improved utilization and verifiable cost reduction. The alpha is not simply in owning compute; it is in owning operators that can absorb compute profitably.
The Allocator Lens
The sovereign AI race is not just a race for national models. It is a race for electricity rights, grid access, physical infrastructure, industrial data and operational absorption. That makes it unusually well-suited to long-duration capital. The natural owners of the next AI bottleneck may not be the most speculative technology investors. They may be the asset owners already built to hold infrastructure, utilities, energy-transition assets, logistics networks and industrial platforms through multi-decade cycles.
The next decade of AI will be priced not only in tokens, chips and model performance, but in megawatts, interconnection queues and the ability to turn computation into real economic throughput.
Antoine Tigneres is a macroeconomic analyst and financial writer at Universal Asset Owners, specialising in institutional intelligence for long-duration capital allocators. Previously a Strategic Intelligence Analyst for the French Ministry of Economy and the French Embassy in Brazil. Full bio →
Source Notes
- Alphabet Inc. 2025 Annual Report / Form 10-K: Capital expenditures of $52.5 billion in 2024 and $91.4 billion in 2025; discussion of technical infrastructure investments including servers, network equipment and data centers.
- International Energy Agency, Electricity 2026 — Executive Summary: U.S. electricity demand projected to grow nearly 2% annually through 2030, with around half of the total increase driven by rapid data-center expansion.
- International Energy Agency, Energy and AI — Energy demand from AI: Global data-center electricity consumption projected to double to around 945 TWh by 2030; data-center electricity consumption projected to grow around 15% annually from 2024 to 2030.
- Central Statistics Office Ireland, Data Centres Metered Electricity Consumption 2024: Data centers consumed 6,969 GWh in 2024, equal to 22% of total metered electricity consumption; urban households accounted for 18%.
- METI, FY2026 Initial Budget Summary: AI-and-semiconductor-related budget of JPY 1.239 trillion in FY2026 versus JPY 332.8 billion in FY2025.
- Ministry of Finance Japan, FY2026 METI Budget Highlights: AI robots / physical AI and AI foundation-model development budget item of JPY 387.3 billion.
- Statistics Bureau of Japan, Statistical Handbook of Japan 2025: Japan working-age population of 73.73 million in 2024, equal to 59.6% of the total population.
- Xinhua, “China moves forward with its AI Plus initiative”: AI Plus described as extensive and in-depth integration of AI across industries, the economy and society.
- Meritech Software Pulse, 01-May-2026: Public software valuation context: only eight companies above 10x NTM revenue and 73% below 5x.