Data centers are the new oil wells
The most valuable physical infrastructure being built today isn't pipelines or refineries. It's data centers. Private equity firms are no longer just leasing rack space, they're building entire campuses from scratch, complete with dedicated power plants. Blackstone, BlackRock, and sovereign wealth funds are writing checks that rival upstream oil and gas investments. In 2025 alone, global data center deals hit a record $61 billion, and capital expenditure in the sector reached $770 billion, surpassing upstream oil and gas spending for the first time. This isn't a tech story. It's an infrastructure story, and it follows a pattern we've seen before.
The oil parallel
The comparison to petroleum isn't rhetorical flourish. The structural dynamics are remarkably similar. In the oil economy, whoever controlled extraction and refining infrastructure controlled the industry. OPEC's power didn't come from owning oil in the ground, it came from controlling the infrastructure that turned crude into something usable. The same logic now applies to compute. NVIDIA controls roughly 92% of the discrete GPU market. The hyperscalers, Amazon, Google, Microsoft, and Meta, are projected to spend nearly $700 billion on AI infrastructure in 2026 alone. Amazon leads with $200 billion in projected capex, Google follows at $175 to $185 billion, and Meta estimates $115 to $135 billion. These numbers don't represent software budgets. They represent concrete, steel, and silicon at a scale that rivals the construction of national energy grids. And just like oil, the infrastructure is concentrating into fewer hands. A consortium including NVIDIA, Microsoft, BlackRock, and xAI acquired Aligned Data Centers in a deal reaching $40 billion. McKinsey projects total investment in the sector could reach $7 trillion by 2030.
Energy is the real bottleneck
The AI scaling story has shifted. The constraint is no longer algorithms or even chips. It's power. Data centers consumed roughly 415 terawatt-hours of electricity globally in 2024, about 1.5% of total global electricity consumption. The IEA found that data center electricity use surged 17% in 2025, with AI-focused facilities growing even faster. In the United States alone, data centers used 183 TWh in 2024, equivalent to Pakistan's entire annual electricity demand, and that figure is projected to grow 133% by 2030. Capital deployment is now driven primarily by power availability, not network connectivity. Investment strategies have pivoted to regions that can guarantee large-scale, reliable electricity. Morgan Stanley described this as a "race to solve the AI power bottleneck," with energy markets scrambling to meet demand that grows faster than anyone predicted. The irony is thick. Oil majors are now profiting from AI twice: deploying AI to optimize extraction, then selling natural gas electricity directly to data centers. Meta alone is funding 10 new gas-fired power plants for its Louisiana data center campus. Google is building a massive natural gas facility in North Texas. The AI industry's demand for power is so intense that it's creating a feedback loop with fossil fuel producers.
Geopolitics of compute
Data center locations are becoming strategic assets, and governments know it. Singapore has opened applications for at least 200MW of new data center capacity, requiring that 50% of proposed capacity be powered by eligible green energy pathways. Bridge Data Centres announced a S$3 to 5 billion investment to position Singapore as Asia Pacific's leading AI hub. The Middle East and Nordic countries are competing for similar positioning, leveraging cheap energy and favorable climates. But geography cuts both ways. S&P Global notes that hyperscalers continue expanding globally despite geopolitical tensions, adding six new regions in Asia-Pacific in 2025 alone. AWS opened Bangkok as its first US-based hyperscaler presence in Thailand. The calculus is familiar: just as oil-producing nations leveraged their reserves for geopolitical influence, countries with abundant energy and favorable regulations are now leveraging their capacity to attract compute infrastructure. The recent US-Iran conflict has made this even more concrete. Data centers, once considered neutral commercial assets, are now being evaluated as potential military targets. The physical security of digital infrastructure has become a national security concern, further blurring the line between technology strategy and foreign policy.
The oligopoly problem
Only a handful of organizations can afford to play this game. When a single data center campus costs tens of billions of dollars and requires its own power generation, the barriers to entry are functionally insurmountable for most companies. This creates a dynamic similar to what happened in oil: natural oligopolies emerge, and everyone else becomes a customer. Startups will rent compute from fewer, more powerful landlords. The hyperscalers aren't just cloud providers anymore, they're infrastructure monopolists whose physical assets determine who can build what. Private credit and structured debt are flooding into the sector, and insurers are creating specialized teams and bespoke policies just to underwrite these facilities. CNBC reported that the sheer scale of capital is "stress testing" the insurance industry itself. When the financial infrastructure has to reshape itself around your asset class, you've moved beyond technology and into something more fundamental.
Jevons paradox and the demand spiral
There's a tempting assumption that efficiency improvements will slow this build-out. DeepSeek's R1 model demonstrated that competitive AI performance was possible at a fraction of the training cost, briefly wiping $600 billion from NVIDIA's market cap. The logic seemed sound: cheaper AI should mean less infrastructure demand. But Jevons paradox suggests the opposite. In 1865, economist William Stanley Jevons observed that improving steam engine efficiency paradoxically accelerated coal consumption. More efficient engines made coal profitable in more applications, and total demand exploded. The same dynamic is playing out with AI compute. Cheaper inference doesn't reduce demand, it unlocks it. More companies can afford to train models. More teams can experiment. More products integrate AI features. Each efficiency gain expands the addressable market, and the underlying infrastructure has to grow to match. A Nature Cities study confirmed this pattern in urban data center energy systems: algorithmic efficiency gains in metropolitan data centers may enlarge, not shrink, the total energy footprint of AI. This is the Jevons spiral applied to data centers. Every dollar saved on compute gets reinvested in more compute.
What this actually means
The data center boom is not a bubble in the traditional sense. Bubbles are characterized by assets that lack underlying utility. Data centers are the opposite: they're the physical substrate on which an increasing share of economic activity depends. But the concentration of this infrastructure raises serious questions. When five companies control the majority of global compute capacity, and when access to that compute determines who can build competitive AI products, we're looking at a power structure that echoes the oil cartels of the 20th century. The difference is that oil was geographically fixed. You drilled where the oil was. Compute infrastructure is theoretically portable, but practically constrained by power availability, regulatory environments, and the massive capital requirements that limit who can build it. For anyone building technology today, the implication is straightforward: the landlords are getting more powerful, and the rent is going up.