A trillion dollar forecast with four names on it
Dell'Oro Group's newest forecast puts worldwide data center capital expenditure on a path to 1.7 trillion dollars by 2030. That figure alone would be a significant story about industry growth. What makes it a story about market structure is the concentration behind it: Amazon, Google, Meta, and Microsoft are expected to represent roughly half of that total spending, and the four companies had already combined for nearly 600 billion dollars in data center capex at the start of 2026.
Baron Fung, senior research director at Dell'Oro Group, summed up why that concentration persists: "Hyperscalers continue to invest aggressively, supported by large cash reserves and long-term market share focus." That is the key mechanism. These four companies can fund capex out of operating cash flow and balance sheet strength that almost no other category of buyer, including most enterprises and even many well capitalized neoclouds, can match at the same scale.
AI servers are reshaping what the money buys
The forecast's more specific claim is about composition, not just total dollars. Accelerated servers built for AI workloads could account for more than two thirds of total data center infrastructure spending by 2030, up from a much smaller share just a few years ago. That is a shift in what capex actually buys: less general purpose compute, more specialized AI accelerated hardware with shorter useful lifespans and faster depreciation cycles than the server generations most enterprise budgets were originally built around.
For any enterprise still planning data center or cloud consumption budgets using depreciation assumptions from the pre AI server era, this forecast is a signal to revisit those assumptions. A market where two thirds of new infrastructure spending goes to accelerated servers is a market where the underlying hardware refresh cycle is compressing, and vendors pricing cloud services around that hardware will pass the faster depreciation through to customers eventually, whether through pricing or through pressure to adopt newer instance types faster than planned.
Everyone else is stuck behind three specific headwinds
Dell'Oro's forecast is explicit about why enterprise data center investment outside the big four hyperscalers remains constrained: tariffs, monetary policy, and unresolved uncertainty about AI return on investment. Each of those is a real, specific barrier rather than generic caution about a hot market cooling off. Tariffs raise the landed cost of server hardware and networking equipment for buyers without a hyperscaler's purchasing scale to negotiate around them, a disadvantage that compounds every time a hardware generation turns over faster than budgets were planned for.
Monetary policy affects the cost of capital for any enterprise financing a data center build or a major cloud commitment with debt, a cost structure the four hyperscalers are largely insulated from given their cash positions. AI return on investment uncertainty is the softest of the three but arguably the most consequential for budget approval: finance leaders are harder to convince to fund new infrastructure for AI workloads whose payback period remains unproven, even as the hyperscalers keep building capacity on the assumption that demand will eventually catch up to supply.
What concentrated capex means for pricing power
When half of a 1.7 trillion dollar global market sits with four companies, those companies set the pace for hardware vendors, power purchase agreements, and construction markets alike. Nvidia, AMD, and the broader AI hardware supply chain increasingly design product roadmaps around what the four hyperscalers will buy, not around what the long tail of enterprise buyers wants. That has downstream effects on what hardware options are even available to a mid sized enterprise building its own infrastructure.
It also means enterprises buying cloud capacity rather than building it are, in effect, riding on the hyperscalers' capex decisions rather than making independent ones. That is not inherently a problem, since it is exactly the value proposition of renting rather than owning infrastructure. It does mean enterprise cloud cost forecasting should track hyperscaler capex and server refresh trends as a leading indicator of future pricing, rather than treating current cloud pricing as a stable baseline.
The build versus buy question gets starker
This forecast sharpens the build versus buy decision rather than settling it. Building proprietary data center capacity now means competing for the same constrained hardware supply, financing, and construction capacity that the hyperscalers are absorbing at a scale few enterprises can match, while facing the same tariff and funding headwinds that are constraining the rest of the non hyperscaler market and showing no sign of easing before the decade is out.
Buying capacity from a hyperscaler means accepting dependence on four companies' capital allocation decisions, but it also means inheriting their purchasing scale, their negotiating leverage with chip vendors, and their ability to absorb the AI server refresh cycle without passing the full cost through immediately. For most enterprises outside of a narrow set of regulatory or latency driven exceptions, the capex concentration Dell'Oro describes is itself an argument for buying rather than building.
The roadmap takeaway
A 1.7 trillion dollar global data center market with half the spending concentrated in four companies is not a market enterprise IT budgets can meaningfully influence from the outside, no matter how large a single enterprise's own infrastructure budget happens to be. The practical response is to plan around that concentration rather than against it: track hyperscaler capex trends as a leading indicator for cloud pricing, revisit hardware depreciation assumptions given the shift toward AI accelerated servers, and treat any new on premises build as a decision made in direct competition with the best funded buyers in the industry for the same constrained hardware supply.
The enterprises that will manage this transition well are the ones that stop evaluating data center capex as an isolated line item and start reading it as a signal about where the broader market's leverage sits. Right now, that leverage sits with four companies, and Dell'Oro's own forecast says it is only going to consolidate further through the end of the decade, which makes this a trend worth revisiting at every annual planning cycle rather than a one time data point.
