Most companies are making AI infrastructure decisions based on what they need right now. That’s the mistake quietly setting up the next wave of enterprise AI failures.
A recent conversation with Alexander Troshin of AMD, published on Forbes, lays out a problem few leadership teams have fully confronted: the data center built for traditional IT workloads is not the data center that can support AI at scale. Companies now run three different computing demands at once: legacy IT systems, active AI workloads, and a fast-growing population of autonomous AI agents, all inside infrastructure designed for none of that.
The tension isn’t really about technology. It’s about time horizon. Infrastructure decisions made for today’s needs are already outdated by the time they’re finished being built, because AI adoption inside most enterprises is moving faster than a typical multi-year infrastructure planning cycle was ever built to handle. A three-year IT roadmap written for last year’s AI usage is not a roadmap. It’s a placeholder.
The Numbers Behind the Warning
This isn’t a theoretical concern. Gartner’s June 2026 forecast puts global data center electricity consumption at 565 terawatt-hours for the year, up 26% from 447 TWh in 2025, with AI-optimized servers driving an 84% jump in power draw over the same period. Gartner also projects AI hardware will overtake conventional servers in total energy consumption by 2027, a crossover analysts did not expect this soon.
The International Energy Agency’s Energy and AI report frames the longer trend: global data center electricity use is projected to roughly double, from around 415 TWh in 2024 to nearly 945 TWh by 2030. Cooling is a direct driver of that curve. Gartner forecasts electricity used specifically by cooling systems to climb 22.6% in 2026 alone, a direct result of denser, hotter AI hardware racks replacing older equipment.
The number that matters here isn’t the raw TWh figure. It’s the growth rate. Traditional enterprise computing demand has largely leveled off. AI-specific infrastructure demand has not, and nothing in the current data suggests it will.
Why This Matters Beyond IT
Energy and cooling costs used to sit on the facilities team’s budget line. They’re becoming a CFO and boardroom problem. AI workloads draw power at a different order of magnitude than the systems most data centers were originally built around, and that cost curve doesn’t level off as adoption grows. It builds on itself, year over year.
A company scaling its AI agent deployment without scaling its power and cooling capacity at the same rate isn’t just facing a technical bottleneck. It’s facing a budget line growing faster than the finance team planned for.
This pattern isn’t isolated to infrastructure. The same imbalance shows up in AI agent security, where agent fleets have roughly doubled in a single quarter while monitoring systems stayed flat. The underlying problem is identical: deployment scaling faster than the systems meant to support it. Infrastructure is simply where that same problem shows up physically, in power draw and cooling capacity instead of security coverage.
The Planning Gap Few Companies Are Closing
Infrastructure planning built only for current workloads guarantees a rebuild within a few years, typically at a far higher cost than planning ahead would have required. Boards that treat this as a pure IT decision, handed entirely to a data center or facilities team, are the ones most likely to hit a capacity wall mid-scale-up, at the exact moment the disruption is hardest to absorb.
The better question for a CEO or founder isn’t whether current infrastructure handles what the company runs today. It’s whether the infrastructure roadmap accounts for what the company will run in three years, once AI agents move from pilot project to core operations. That question belongs in the same conversation as the AI budget itself, not a separate one handled quietly by IT after the fact. A roadmap nobody outside IT has reviewed isn’t a strategy. It’s an assumption.
Four Questions to Bring to Your Next Leadership Meeting
- What is our current power and cooling capacity, and what share of it is already committed to AI workloads? Leadership teams without a ready answer to this already have their answer: the planning gap exists.
- Who owns the three-to-five-year infrastructure roadmap, and has it been checked against our AI adoption plans directly? If the infrastructure team and the AI strategy team built separate plans without comparing notes, that’s a coordination gap worth closing now, not after the next capacity review.
- What happens to our cost structure if AI agent deployment doubles again next year? Given that fleets have already shown this exact pattern elsewhere in the enterprise landscape, this is a near-term planning input, not a hypothetical stress test.
- Are we building for liquid cooling, or still relying on legacy air cooling as AI hardware density rises? This is a real engineering decision with real cost consequences, and it’s one finance leaders are rarely asked to weigh in on, even though the resulting costs land on their desk eventually.
The Takeaway
AI-ready infrastructure isn’t a bigger version of what companies already have. It’s a different kind of planning built around a moving target instead of a fixed one. Companies building for where AI adoption is headed, not where it stands today, are the ones that won’t be rebuilding under pressure, and under budget strain, a few years from now.
📩 Have questions about what AI-ready infrastructure planning looks like for your organization? Digital Success Hub helps CEOs and founders think through AI strategy decisions before they become expensive to reverse.
Email: info@digitalsuccesshub.org