On-premise vs colocation vs hyperscale cloud: data center sustainability performance compared

Why It Matters

Global data center electricity consumption reached an estimated 460 TWh in 2025, roughly 1.8% of worldwide electricity demand, and the International Energy Agency projects that figure could exceed 800 TWh by 2030 driven largely by AI workloads (IEA, 2025). Where and how enterprises house their compute infrastructure has direct, measurable consequences for carbon emissions, water withdrawal, and electronic waste. An on-premise server room in a temperate office building operates at a fundamentally different efficiency profile than a purpose-built colocation facility or a hyperscale campus engineered by Google, Microsoft, or Amazon. Yet many enterprises still default to legacy on-premise deployments without evaluating the sustainability trade-offs. This comparison provides the data, cost benchmarks, and decision criteria needed to align data center strategy with net-zero commitments.

Key Concepts

Power Usage Effectiveness (PUE). The ratio of total facility energy to IT equipment energy. A PUE of 1.0 means every watt powers compute; overhead (cooling, lighting, power conversion) adds to the denominator. The global average PUE across all data centers sits at approximately 1.58, but hyperscale operators routinely achieve 1.08 to 1.12 (Uptime Institute, 2025).

Water Usage Effectiveness (WUE). Liters of water consumed per kilowatt-hour of IT energy. Evaporative cooling systems are efficient for heat rejection but consume significant water. Google reported an average WUE of 1.1 L/kWh across its fleet in 2024, while on-premise facilities using conventional chillers may not report WUE at all (Google Environmental Report, 2025).

Carbon-Free Energy (CFE) matching. The percentage of electricity consumption matched by carbon-free sources on an hourly, location-specific basis. Google achieved 64% 24/7 CFE across its global fleet in 2024 and targets 100% by 2030 (Google, 2025). Microsoft committed to 100% renewable energy by 2025 on an annual matching basis and is pursuing hourly matching through energy attribute certificates (Microsoft Sustainability Report, 2025).

Scope 1, 2, and 3 emissions. On-premise data centers generate Scope 1 emissions (diesel backup generators), Scope 2 (purchased electricity), and Scope 3 (embodied carbon in servers). Colocation and cloud shift Scope 2 and parts of Scope 3 to the provider, but the emissions still exist within the value chain.

Embodied carbon. The carbon emitted during manufacture, transport, and installation of IT hardware. Servers typically carry 500 to 1,200 kg CO2e of embodied carbon, and extending server life from three to five years reduces annualized embodied emissions by approximately 40% (Dell Technologies, 2024).

Stranded capacity. On-premise facilities average 20% to 30% utilization rates, meaning 70% to 80% of provisioned capacity sits idle while still drawing power for cooling and standby (McKinsey, 2025). Hyperscale operators target 60% to 80% utilization through workload orchestration.

Head-to-Head Comparison

Cost Analysis

Capital expenditure. Building a new on-premise data center costs $8 to $12 million per megawatt of IT load, inclusive of land, building, power infrastructure, and cooling (JLL, 2025). Colocation eliminates facility CapEx; enterprises pay monthly per-kilowatt fees ranging from $120 to $200/kW/month in Tier III facilities in major markets. Hyperscale cloud eliminates both facility and server CapEx, converting costs to operational expenditure billed per compute-hour, storage-gigabyte, or reserved instance.

Energy costs. On-premise operators pay retail electricity rates (US average $0.13/kWh commercial in 2025) multiplied by their PUE. A 1 MW on-premise facility at PUE 1.8 consumes 1.8 MW total, costing roughly $2.05 million per year. The same IT load in a hyperscale facility at PUE 1.1 consumes 1.1 MW, and hyperscale operators negotiate wholesale power purchase agreements at $0.04 to $0.07/kWh, yielding annual energy costs of $385,000 to $675,000 per megawatt of IT load (Bloomberg NEF, 2025). That represents a 60% to 80% energy cost reduction.

Carbon costs. Under the EU Emissions Trading System, carbon costs reached approximately EUR 65/tonne in early 2026. A 1 MW on-premise facility in a carbon-intensive grid (500 g CO2e/kWh) at PUE 1.8 emits roughly 7,900 tonnes CO2e annually, generating a carbon liability of EUR 513,000. The same workload on a hyperscaler running 90% carbon-free energy emits approximately 480 tonnes, reducing the liability to EUR 31,000. Even without a direct carbon tax, internal carbon prices adopted by over 2,000 companies globally make this differential financially material (CDP, 2025).

Total cost of ownership (TCO) over five years. Gartner estimates that enterprises migrating 80% of on-premise workloads to hyperscale cloud reduce five-year infrastructure TCO by 30% to 45%, with sustainability-related savings (energy, carbon, e-waste compliance) contributing 8 to 12 percentage points of that reduction (Gartner, 2025).

Use Cases and Best Fit

On-premise remains viable when data sovereignty regulations mandate physical control (e.g., certain defense, healthcare, or government workloads), when ultra-low-latency edge requirements preclude round-trip to external facilities, or when legacy applications cannot be refactored for cloud deployment. The European Central Bank, for instance, maintains on-premise infrastructure for core payment systems where regulatory compliance and latency requirements override sustainability optimization. Organizations choosing on-premise should invest in liquid cooling, hot/cold aisle containment, and renewable energy procurement to close the PUE gap.

Colocation fits when enterprises need physical proximity, custom hardware configurations, or multi-provider connectivity without building their own facility. Equinix, the world's largest colocation provider with 260+ data centers across 72 metros, achieved 96% renewable energy coverage in 2024 and offers customers granular carbon accounting through its Equinix Sustainability Dashboard (Equinix, 2025). Digital Realty, another major provider, committed to net-zero Scope 1 and 2 emissions by 2030 and operates 300+ facilities globally. Colocation is particularly suited for hybrid architectures where some workloads require dedicated hardware while others burst to cloud.

Hyperscale cloud delivers the strongest sustainability profile for general-purpose compute, analytics, AI training, and elastic workloads. AWS announced that its global operations reached 100% renewable energy matching in 2024, one year ahead of its 2025 target, and operates over 500 renewable energy projects worldwide (AWS, 2025). Google has operated on 100% annual renewable matching since 2017 and is pursuing 24/7 carbon-free energy at every facility. Microsoft Azure provides a Carbon Optimization dashboard that lets customers track and reduce per-workload emissions. Enterprises with aggressive net-zero targets and cloud-compatible workloads achieve the fastest sustainability improvements by migrating to hyperscale providers.

Decision Framework

1. Audit current infrastructure. Measure PUE, utilization rates, energy source, and WUE for existing on-premise facilities. If PUE exceeds 1.6 and utilization sits below 25%, the sustainability case for migration is strong.

2. Classify workloads by sensitivity. Map each workload against data sovereignty requirements, latency thresholds, and compliance mandates. Sovereign or ultra-low-latency workloads may need to remain on-premise or in local colocation; all others are candidates for cloud migration.

3. Evaluate provider sustainability commitments. Compare prospective colocation and cloud providers on PUE, renewable energy percentage, WUE, Scope 3 reporting, and circular economy programs. Request provider-specific carbon data rather than relying on marketing claims.

4. Model carbon impact. Calculate annualized emissions under each deployment model using provider-published emission factors. Include embodied carbon from hardware refresh cycles and backup generator fuel.

5. Assess financial incentives. Factor in carbon pricing (explicit or internal), energy cost differentials, and stranded asset risk for aging on-premise facilities. Many enterprises find that the combination of lower energy costs and avoided carbon liability makes migration cost-neutral or better within 18 to 24 months.

6. Plan the transition. Adopt a phased migration starting with development/test environments and non-sensitive workloads. Use colocation as a stepping stone for workloads that need physical proximity but can benefit from shared cooling and renewable energy procurement.

Key Players

Established Leaders

• Amazon Web Services (AWS) — 100% renewable energy matching achieved in 2024; 500+ renewable energy projects globally
• Google Cloud — 64% 24/7 CFE fleet-wide; pioneered hourly carbon-free energy accounting methodology
• Microsoft Azure — Carbon Optimization dashboard; $50B+ invested in data center capacity 2024-2026 with sustainability integration
• Equinix — Largest colocation provider; 96% renewable energy; 260+ facilities in 72 metros
• Digital Realty — 300+ facilities; net-zero Scope 1/2 target by 2030; Science Based Targets validated

Emerging Startups

• Crusoe Energy — Uses stranded natural gas and flared methane to power modular data centers, reducing net emissions
• Submer — Immersion cooling technology reducing PUE to 1.03; deployed in colocation and enterprise settings
• Nautilus Data Technologies — Water-cooled floating data centers eliminating water consumption for cooling
• Deft (formerly ServerCentral) — Mid-market colocation with 100% renewable energy and carbon-neutral operations
• Lancium — Clean-energy-powered computing infrastructure co-located with renewable generation assets

Key Investors/Funders

• Brookfield Renewable Partners — Multi-billion-dollar data center energy infrastructure investments
• DigitalBridge — $80B+ AUM in digital infrastructure including sustainable data center platforms
• Global Infrastructure Partners — Major investor in data center and renewable energy co-development
• TPG Rise Climate — Climate-focused PE fund backing low-carbon data center technologies

FAQ

Does moving to the cloud actually reduce total emissions, or just shift them? Migration to a major hyperscale provider genuinely reduces absolute emissions in most scenarios. The efficiency gap is structural: hyperscale operators achieve server utilization rates three to five times higher than typical on-premise environments, run at PUE values 30% to 45% lower, and procure renewable energy at scale that individual enterprises cannot match. The IEA (2025) estimates that if all enterprise workloads currently running in on-premise facilities at PUE >1.6 migrated to hyperscale, global data center energy demand would drop by 25% to 35%, even accounting for rebound effects from increased usage.

How should enterprises account for cloud emissions in their Scope 3 reporting? Cloud computing emissions fall under Scope 3, Category 1 (purchased goods and services) or Category 8 (upstream leased assets) of the GHG Protocol. AWS, Google Cloud, and Microsoft Azure all provide customer-facing carbon footprint tools that report emissions attributable to specific accounts. The GHG Protocol's forthcoming guidance on cloud computing emissions (expected 2026) will standardize allocation methodologies. Enterprises should request location-based and market-based emission factors from providers and integrate these into their corporate carbon inventories.

What about water consumption? Are hyperscale data centers making the water crisis worse? Water usage is a legitimate concern, particularly for facilities using evaporative cooling in water-stressed regions. Google consumed approximately 6.1 billion gallons of water across its data centers in 2024 (Google Environmental Report, 2025), prompting commitments to replenish 120% of water consumed by 2030. Microsoft has a similar water-positive pledge. On-premise facilities using air-cooled chillers consume less water but more electricity. The trade-off is site-specific: in water-scarce regions, air cooling or closed-loop liquid cooling is preferable despite higher energy use. Leading colocation providers like Equinix are deploying adiabatic and liquid cooling systems that reduce WUE below 0.5 L/kWh.

Is colocation a meaningful middle ground, or should enterprises go directly to cloud? Colocation serves a distinct and valuable role for workloads that require dedicated hardware, custom network configurations, or physical proximity to specific interconnection points. Financial services firms running high-frequency trading, healthcare organizations with on-premise medical device integrations, and enterprises with large existing hardware investments benefit from colocation's shared cooling, power, and renewable energy procurement without refactoring applications. For new workloads and cloud-native applications, hyperscale cloud is typically the faster path to sustainability improvement.

How quickly are hyperscale operators moving toward 24/7 carbon-free energy? Progress is accelerating but uneven. Google's fleet-wide 24/7 CFE score improved from 61% in 2022 to 64% in 2024, with individual campuses (Finland, Oregon) exceeding 90%. Microsoft has signed over 13 GW of renewable energy capacity since 2024, the largest corporate clean energy portfolio in history (Microsoft, 2025). The main bottleneck is not commitment but grid infrastructure: connecting new renewable generation to data center load requires transmission upgrades that take 3 to 7 years in most markets. Battery storage, long-duration energy storage, and nuclear power purchase agreements (as pursued by both Google and Microsoft with SMR developers) are the emerging solutions.

Sources

• International Energy Agency. (2025). Data Centres and Data Transmission Networks: Tracking Report 2025. IEA, Paris.
• Uptime Institute. (2025). Global Data Center Survey 2025: PUE Trends and Efficiency Benchmarks. Uptime Institute.
• Google. (2025). 2024 Environmental Report: Carbon-Free Energy, Water Stewardship, and Circular Economy. Google LLC.
• Microsoft. (2025). 2024 Sustainability Report: Progress Toward Carbon Negative, Water Positive, Zero Waste Goals. Microsoft Corporation.
• AWS. (2025). Amazon Sustainability Report 2024: Renewable Energy and Data Center Efficiency. Amazon Web Services.
• Equinix. (2025). 2024 Sustainability Report: Global Renewable Energy Coverage and Carbon Accounting. Equinix Inc.
• McKinsey & Company. (2025). The Green Data Center: Pathways to Sustainable IT Infrastructure. McKinsey Digital.
• Bloomberg NEF. (2025). Global Corporate PPA Pricing and Data Center Energy Procurement Trends. BNEF.
• CDP. (2025). Putting a Price on Carbon: The State of Internal Carbon Pricing 2025. CDP Worldwide.
• Gartner. (2025). Cloud Migration TCO Analysis: Infrastructure Cost and Sustainability Impact. Gartner Research.
• Dell Technologies. (2024). Embodied Carbon in Server Infrastructure: Lifecycle Assessment and Extension Benefits. Dell Technologies.
• JLL. (2025). Data Center Construction Costs: Global Benchmarking Report 2025. Jones Lang LaSalle.