Infrastructure Costs—The Hidden Multiplier Behind Rising AI Budgets

The CFO community has fixated on subscription and token costs while the real driver of AI expense growth hides in plain sight: infrastructure. Managing AI infrastructure costs is like operating a factory: the machinery (AI models) gets the spotlight, but the power grid and cooling systems (data centers) determine the true cost of operations. CFOs who optimize the energy supply chain will stay ahead of competitors still focused only on software pricing.

For context on how subscription costs interact with these infrastructure expenses, see our analysis of AI Subscriptions vs. Tokens: Why Costs Rise Even as Prices Fall.

Bloomberg NEF projects that US data center power demand will more than double by 2035, rising from 35 gigawatts in 2024 to 78 gigawatts. But energy consumption growth will be even steeper—nearly tripling from current levels by 2035. The math is stark: AI workloads can require 4 to 40 times more energy per square foot than traditional computing, with some new facilities demanding up to 2,000 megawatts of power—equivalent to a small city.

Before diving into industry trends, see what these costs might mean for your organization:

Use our AI Energy Cost Calculator to model your organization's specific energy expenses and regional cost variations.

The Energy Economics CFOs Can't Ignore

Unlike software licensing, energy costs compound across multiple layers. Understanding these dynamics is critical for strategic planning:

AI Query Impact:

• A single AI query through ChatGPT requires 10x more electricity than a Google search
• Data centers already account for more than 4% of US electricity use
• That figure could reach 12% by 2028, with AI comprising up to 40% of global data center power demand by 2026

Power Usage Effectiveness (PUE) has become a critical financial metric:

• The most efficient data centers achieve PUE ratings of 1.2 or lower (only 20% energy goes to non-computing functions)
• AI-intensive facilities often struggle with higher PUE ratings due to thermal demands of GPU-heavy workloads
• Every 0.1 increase in PUE translates to roughly 10% higher energy costs—a direct hit to operational margins

Cooling represents the largest hidden cost multiplier:

• Traditional data centers dedicate up to 40% of total power consumption to cooling systems
• AI workloads generate significantly more heat, often requiring liquid cooling systems
• Advanced cooling strategies report PUE improvements of 0.3 to 0.5 points, translating to 25-40% reductions in total energy costs

Regional Cost Variations Create Strategic Opportunities

Geographic positioning has become a crucial cost management strategy, with regional differences creating 25-30% cost advantages:

Lowest Cost Regions:

• Pacific Northwest: Hydroelectric power in Washington state provides rates around $0.08/kWh vs. $0.12 national average
• Texas: Wind energy abundance offers stable pricing around $0.09/kWh with strong grid reliability

Moderate Cost Regions:

• Southeast: Coal and nuclear baseload power provides predictable pricing around $0.11/kWh
• Midwest: Mixed energy sources offer balanced cost and reliability

Highest Cost Regions:

• California: Renewable mandates and grid constraints push rates to $0.22/kWh or higher
• Northeast: Grid capacity limitations and transmission costs drive rates to $0.16/kWh average

Strategic Implications: Organizations can achieve 20-30% cost reductions by strategically placing AI workloads in low-cost regions while maintaining compliance and performance requirements.

The Cloud Cost Pass-Through Reality

Enterprise AI costs aren't limited to direct subscriptions—they include infrastructure costs cloud providers pass through via higher service pricing. The math is stark: hyperscalers' $300 billion investment in AI infrastructure in 2025 will drive systematic price increases across all cloud services, directly impacting your AI budget regardless of which models you choose.

Hyperscaler Infrastructure Investment:

• Microsoft, Google, and Amazon are building data centers that consume 2-5x more power than previous generation facilities
• The largest planned facilities require up to 2 gigawatts of power capacity—doubling to quadrupling existing mega-scale data centers
• Construction timelines average seven years from planning to operation, creating supply constraints that drive up pricing

Energy-as-a-Service Models:

• Some organizations partner with energy providers for on-site generation, battery storage, and demand response systems
• Hybrid approaches can reduce grid dependency by 30-50%
• Revenue streams through wholesale electricity market participation during peak pricing periods

Just as chasing the latest AI models can inflate costs without corresponding value, overbuilding infrastructure without clear ROI can lock organizations into decades of escalating expenses. The discipline lies in strategic positioning, not maximum capacity.

For Smaller Organizations

Smaller enterprises can sidestep infrastructure cost traps through targeted strategies:

Leverage Serverless and Shared Resources:

• Use serverless AI platforms that absorb infrastructure complexity and spread costs across multiple tenants
• Select cloud providers with data centers in low-cost regions like Texas (wind) or Washington (hydroelectric)
• Negotiate fixed-rate contracts that limit exposure to infrastructure cost pass-throughs

Strategic Provider Selection:

• Prioritize vendors with transparent pricing and low-PUE data centers (ratings below 1.3)
• Focus on providers offering geographic flexibility to optimize for cost vs. compliance requirements
• Consider multi-tenant solutions where infrastructure investments are shared across user bases

Cost Control Mechanisms:

• Implement usage monitoring to avoid unexpected spikes during high-demand periods
• Build contract provisions that cap annual infrastructure cost increases to defined percentages
• Evaluate total cost of ownership, not just subscription pricing, when making vendor decisions

Practical CFO Implementation Framework

1. Model Energy Costs Proactively

• Assume 20-40% annual increases in data center energy costs through 2030, driven by supply constraints
• Include cooling, backup power, and transmission costs in total calculations—not just compute energy
• Model these increases as a percentage of total AI spend to maintain budget discipline

2. Demand Vendor Transparency

Require detailed infrastructure cost visibility from all AI service providers:

• Request PUE ratings, energy sourcing strategies, and geographic deployment details for all major contracts
• Vendors with PUE ratings below 1.3 and renewable energy commitments offer better long-term cost predictability
• Build contract provisions that limit infrastructure cost pass-throughs to defined annual percentage increases

3. Optimize Geographic and Strategic Positioning

Treat data center location as a strategic cost lever:

• Evaluate AI workload placement based on regional energy costs, not just latency requirements
• Consider hybrid cloud strategies that balance cost optimization with performance and compliance needs
• Data centers in renewable energy-rich regions often provide 20-30% lower long-term operating costs

Multi-Year Planning in an Inflationary Environment

The math is stark on systematic cost pressures that will persist regardless of AI model efficiency improvements:

Capacity Pricing Escalation:

• PJM Interconnection's capacity auction prices rose nearly 500% for 2025/2026, affecting 13 states
• Capacity charges can represent 20-30% of total electricity costs for high-demand users
• Goldman Sachs predicts 3.3 billion cubic feet per day of new natural gas demand by 2030—a 3.5% daily increase

Infrastructure Investment Timeline Mismatches:

• Data center construction averages seven years from planning to full operation
• AI demand is growing faster than infrastructure capacity can be added
• Supply constraints create systematic upward pricing pressure that compounds annually

The Strategic Implications

Infrastructure costs represent the largest unmanaged component of AI budgets. Unlike software subscriptions that can be cancelled or renegotiated, energy and data center commitments create multi-year operational obligations that compound over time.

CFOs who model these costs proactively—including energy inflation, cooling requirements, and geographic optimization—will avoid the budget surprises that derail AI initiatives. More importantly, they'll convert higher infrastructure investments into competitive advantages through better performance, reliability, and strategic positioning.

The bottom line: This isn't just about managing expenses—it's about positioning for a decade where AI infrastructure becomes as critical to business operations as traditional manufacturing or logistics infrastructure. The organizations that treat AI infrastructure as a strategic asset class, not just an operational expense, will convert these higher costs into sustainable competitive advantages.

Those that continue focusing only on software pricing while ignoring infrastructure planning will face escalating costs without corresponding strategic value—a disadvantage that compounds with every budget cycle.

Infrastructure costs for AI are like running a high-performance factory: the subscription price for the machinery gets attention, but the power, cooling, and facility costs determine whether the operation generates profit or losses over time.

References

1. Bloomberg NEF. "Power for AI: Easier Said Than Built." July 2025. https://about.bnef.com/insights/commodities/power-for-ai-easier-said-than-built/
2. Deloitte. "Can US infrastructure keep up with the AI economy?" July 2025. https://www.deloitte.com/us/en/insights/industry/power-and-utilities/data-center-infrastructure-artificial-intelligence.html
3. Diversegy. "Energy Prices 2025: US & Global Market Forecast." October 2024. https://diversegy.com/energy-prices-market-forecast-2025/
4. Diversegy. "Mitigating Rising Energy Capacity Prices in 2025." February 2025. https://diversegy.com/mitigating-rising-energy-capacity-prices-2025/
5. ENR. "AI-Fueled Data Center Boom Sets Energy Delivery's New Course." July 2025. https://www.enr.com/articles/61083-power-hungry-ai-fueled-data-center-boom-sets-energy-deliverys-new-course
6. HGA. "Four Practical Approaches to Planning Energy-Efficient Data Centers." April 2025. https://hga.com/four-practical-approaches-to-planning-energy-efficient-data-centers/
7. Lumenalta. "Understanding the cost to setup an AI data center (updated 2025)." January 2025. https://lumenalta.com/insights/understanding-the-cost-to-setup-an-ai-data-center-updated-2025
8. MIT Sloan. "AI has high data center energy costs — but there are solutions." November 2023. https://mitsloan.mit.edu/ideas-made-to-matter/ai-has-high-data-center-energy-costs-there-are-solutions
9. NREL. "High-Performance Computing Data Center Power Usage Effectiveness." April 2025. https://www.nrel.gov/computational-science/measuring-efficiency-pue
10. Nutanix. "Data Center and Cloud Cost Control in Enterprise AI Era." May 2025. https://www.nutanix.com/theforecastbynutanix/business/data-center-and-cloud-cost-control-in-enterprise-ai-era