Research compiled March 2026
From AI chips to gigawatt campuses — understanding the infrastructure powering the AI era, and how to build a career that leads to operating one.
The convergence of AI demand, sovereign computing ambitions, and the training-to-inference shift has turned data centers from boring infrastructure into the most capital-intensive sector in tech history. In 2025, the top hyperscalers spent $413 billion on data center capex — up 84% from $224B in 2024. By 2026, that number will approach $600-690 billion.
The scale is staggering: Meta is building a 5 GW campus in Louisiana. The Stargate project spans multiple states with 7 GW planned. Saudi Arabia is investing $77 billion in data center infrastructure. South Korea committed $75 billion to sovereign AI.
The industry faces a massive timing mismatch: nearly $4 trillion in cumulative capex is projected through 2030, while cumulative AI revenue in that period is expected to be under $2 trillion. The bet is that post-2030 revenue acceleration will justify today's buildout. Power constraints — not capital — are the natural throttle preventing a dot-com-style overbuild.
Getting from zero to operating a data center is more accessible than it appears. RunPod started with $50K in GPU rigs and hit $120M ARR. CoreWeave pivoted from crypto mining to a $23B+ IPO. The common thread: start with software on leased infrastructure, then vertically integrate into facilities as you scale. Finance, energy procurement, and real estate skills matter as much as technical depth.