Earn with Nexo
Half of AI datacenter spending goes to non-chip expenses, and crypto miners are paying attention
Morgan Stanley projects $3 trillion in global AI datacenter spending by 2029, with infrastructure costs dwarfing chip purchases and former crypto miners racing to fill the gap.
Morgan Stanley estimates that worldwide spending on AI-related datacenters will hit roughly $3 trillion by 2029. Here’s the part that should reframe how you think about the entire sector: approximately half of that spending goes to construction, not hardware. The concrete, the cooling systems, the power infrastructure, the networking equipment.
Chips are the minority line item
Goldman Sachs projects that by 2026, chips will represent only about 25% of total AI datacenter spending. The remaining 75% flows into physical infrastructure development. In English: for every dollar spent on the brains of an AI datacenter, three dollars go toward keeping those brains housed, cooled, and powered.
Cooling systems alone can account for up to 40% of electricity demand in a datacenter.
The top hyperscalers, Amazon, Microsoft, Google’s parent Alphabet, and Meta, are projected to spend somewhere in the range of $587 billion to $670 billion on AI infrastructure capital expenditures in 2026.
Former crypto miners are already pivoting
Former crypto miners are converting their facilities for AI compute workloads, and the deals are not small. IREN landed a $9.7 billion contract with Microsoft. TeraWulf entered a $9.5 billion joint venture with Google.
A crypto mining facility already has the power infrastructure, the cooling capacity, and the permitting in place. Converting it for AI workloads is dramatically cheaper and faster than building from scratch. When 75% of your datacenter cost is infrastructure rather than chips, having that infrastructure already built is an enormous competitive advantage.
What this means for crypto investors
Platforms like Akash Network, Render, and Bittensor are positioning themselves as alternatives to traditional cloud services, using token incentives to aggregate existing compute resources without construction permits, water rights negotiations, or multi-year build timelines.
Decentralized compute is still a fraction of the market compared to what Amazon Web Services or Google Cloud command. The technology works for certain workloads, particularly inference and rendering tasks, but training large foundation models still requires the kind of concentrated, low-latency infrastructure that only purpose-built facilities can provide.
The IREN and TeraWulf deals demonstrate that the market is willing to pay enormous premiums for ready-to-deploy infrastructure. The energy demands of AI datacenters will intensify competition for power resources, which directly affects proof-of-work mining economics.