Hyperscalers outspend US megaprojects on AI infrastructure

We have entered the era of the "permanent megaproject" where private capital has replaced the state as the primary architect of national infrastructure.

• –Cumulative AI capex for 2023-2025 is estimated to exceed $600 billion, eclipsing the ~$500B inflation-adjusted cost of the entire US Interstate Highway System.
• –Individual datacenter projects are scaling to 500MW and $10 billion, matching the price tag and complexity of modern nuclear reactors.
• –Unlike historical infrastructure like rail or roads, GPU-based compute has a rapid 5-10 year depreciation cycle, requiring a continuous, multi-trillion dollar investment loop to maintain.
• –This concentrated spending creates a "brute-force" moat that effectively locks out international competitors and consolidates US technological hegemony.
• –The scale of investment mirrors the 19th-century railroad booms, carrying significant risk of over-expansion if the AI returns on investment fail to materialize quickly.