How Colocation Providers Find and Reclaim Stranded Power Capacity

The Problem

One in four data centers operates at under 40% of its available UPS capacity, according to the Uptime Institute’s 2024 Global Data Center Survey.

Many colos currently leverage manual processes to find stranded power, pulling a BMS export from last Tuesday, a spreadsheet three people update inconsistently, and a nameplate-based power budget built on assumptions nobody had tested in three years. By the time the team could confirm real available capacity across the full power chain, the tenant had signed with someone else.

That’s the stranded power problem the colocation providers are facing in 2026, and the potential revenue lost could be millions.

Why Is Stranded Power Hard to Find?

Operators hunting for stranded power usually start at the meter. But a facility-level meter only shows that headroom exists, not where, why, or whether it’s recoverable. Stranded power accumulates across five distinct points in a colocation environment, and each one compounds the others.

1. Nameplate provisioning gaps – Traditional cabinets provisioned at 10 kW may routinely draw only 3–4 kW, leaving contracted capacity unused. Across hundreds of cabinets, that gap can add up to megawatts of stranded power at the same time AI tenants are requesting deployments of 30–50 kW per rack or more. Colocation providers need real-time consumption data to determine whether unused capacity can be safely reclaimed and consolidated to support those higher-density requirements.
2. Phase imbalance at the branch circuit level – A circuit rated for 30 amps may be delivering 20 amps of usable capacity because load is concentrated on one leg. Headroom capacity exists electrically but without branch-circuit-level monitoring, it’s unreachable and invisible to the meter.
3. Ghost power draw – According to research by Anthesis Group and Stanford’s Jonathan Koomey, up to 30% of servers are “comatose” — powered on, drawing load, consuming cooling, and doing nothing productive. Manual audits could surface idle draw at one facility but not scale across 20 locations.
4. Cooling-constrained racks – Power is available at the circuit, but the rack has exhausted its thermal headroom. Cooling infrastructure built for 8 kW average density cannot dissipate the heat generated by a 30-kW AI rack. Every kilowatt above that thermal ceiling becomes structurally stranded in capacity.
5. Redundancy headroom held flat across the portfolio – N+1 maintained uniformly across 40 sites burns capacity that mature, stable zones don’t need. Identifying where full redundancy is genuinely required, and proving it with data, recovers real megawatts without touching uptime.

Schedule a personalized demo with our technical team to discover how much stranded capacity is hidden across your data center portfolio.

What Does Full-Chain Visibility Actually Require?

Recovering stranded capacity requires more than another meter or dashboard. It requires visibility across the entire power chain, because each measurement tier answers a different operational question, and leaves a different blind spot.

• Facility and utility feed level gives the headline gap. It doesn’t give the cause, the location, or whether the gap is recoverable.
• UPS and switchgear level shows load on each power path and whether N+1 headroom reflects actual risk or a design document nobody has revisited since commissioning.
• PDU and floor distribution level is where phase imbalance and nameplate gaps become visible. For a provider making SLA commitments by circuit, revenue-grade metering at ANSI C12.20 accuracy at this tier isn’t optional, it’s the data that defends the contract.
• Branch circuit level is where tripping risk and hidden headroom surface. Each breaker, each phase. Legacy facilities carry recoverable capacity here that never appears at the facility or PDU level.
• Rack and BMC level closes the gap between what’s been sold and what’s being drawn — identifying ghost servers and nameplate over-provisioning at the device level.

The challenge, then, is not collecting more data. It is connecting the data that already exists. BMS data resides in one system, PDU metrics in another, and SNMP traps in yet another. Branch-circuit data arrives in formats that have never been standardized across the portfolio. Turning those fragmented inputs into a single, real-time operational picture is not a spreadsheet exercise.

Once that picture exists, stranded capacity becomes measurable, locatable, and actionable, giving colocation providers defensible capacity they can confidently sell and convert into revenue.

How to turn hidden power capacity into revenue?

1. Right-sized power budgets measured draw. Replace nameplate-based provisioning with measured peak draw plus a calibrated diversity factor. At a 500-cabinet facility running at 40% average utilization, this recovers hundreds of kilowatts of provably available capacity — without changing a single piece of physical infrastructure.
2. Rebalance phases to recover circuit headroom. Correcting phase imbalance at the branch circuit level recovers headroom that was tripping breakers before circuits were electrically full. It starts with monitoring that identifies which circuits are worth addressing.
3. Decommission ghost load. Automated detection of powered-on, near-zero-utilization devices frees power, cooling, and space simultaneously. Across a multi-site portfolio, this requires real-time telemetry — not the next manual floor walk.
4. Implement dynamic redundancy management. Releasing N+1 headroom in zones where full redundancy isn’t genuinely required puts reserved capacity into sellable inventory — a material advantage over operators holding full headroom everywhere by default.

Any one of these recovers’ capacity at a single site. What makes them matter is what happens when you run them across a portfolio.

The Commercial Case

At one facility, stranded power is an operational problem. At 20 or 40 sites, it’s a revenue and risk question.

Every site carries SLA exposure. Every gap in real-time visibility is a potential tenant dispute, a missed capacity sale, or regulatory exposure in markets where utility contracts increasingly include “use it or lose it” clauses for underutilized power allocations.

Recovering 2% to 3% of stranded capacity across a 100 MW portfolio is 2 to 3 MW of sellable inventory — without new grid connections, without new construction, and without waiting five to seven years for a substation upgrade.

At CBRE’s reported average of $184 per kilowatt per month for 250 to 500 kW deployments, 1 MW of recovered stranded capacity is worth approximately $2.2 million in annual recurring revenue.

Because the colocation providers winning high-density AI tenant contracts aren’t the ones with the most capacity on paper. They’re the ones who can prove available capacity — by rack, by circuit, by site — in a single conversation. Schedule a personalized demo with our technical team to discover how much stranded capacity is hidden across your data center portfolio.

How Modius OpenData Can Help You Find Stranded Capacity

Stranded capacity is rarely visible in one system. Power data may sit in the BMS, circuit loads in intelligent PDUs, environmental conditions in separate monitoring tools, and equipment demand in server-level systems. Modius OpenData® brings those fragmented inputs together into a single, real-time operational view across racks, rooms, sites, and portfolios.

By collecting data in real-time and normalizing data from different vendors, devices, and protocols, OpenData helps colocation teams compare provisioned capacity with actual demand and identify where capacity is being lost to overprovisioning, phase imbalance, idle equipment, cooling constraints, or overly conservative operating assumptions. This gives operators the context to determine not only where headroom exists, but whether it can be safely reclaimed and sold.

OpenData helps colocation providers:

• Validate available capacity with real-time operating data – Compare contracted or nameplate capacity with actual peak and historical demand rather than relying on estimates or isolated meter readings.
• Locate stranded capacity across the full power chain – Trace available and constrained capacity from the facility and UPS level down to PDUs, branch circuits, racks, and connected equipment.
• Identify conditions preventing capacity from being used – Surface phase imbalance, abnormal consumption, idle load, thermal limitations, and other conditions that may leave power unavailable or underused.
• Evaluate capacity consistently across multiple sites – Normalize data from disparate systems and vendors into a common portfolio-level view, making it easier to compare sites and identify the best location for new deployments.
• Support higher-density tenant requirements – Give sales, capacity-planning, and operations teams defensible data showing where reclaimed capacity can support new AI and other high-density workloads.

The result is a more accurate inventory of sellable capacity. Instead of telling a prospective tenant how much power appears to be available on paper, colocation teams can show what is available by site, circuit, and rack — and demonstrate the operational evidence behind it. OpenData is designed to help operators increase sellable capacity, reduce operational risk, and manage complex multi-tenant environments at scale. Schedule a personalized demo with our technical team to discover how much stranded capacity is hidden across your data center portfolio.

Frequently Asked Questions

What is the difference between provisioned capacity and actual power draw in a colocation facility?

Provisioned capacity is what’s been contracted with tenants and allocated in power budgets — typically based on nameplate server ratings. Actual draw is what equipment is genuinely consuming in real time. The gap between the two, which commonly runs 40–60% of provisioned capacity at the facility level, is where stranded power lives.

Modius OpenData® continuously collects and correlates real-time power data across the facility, helping operators compare contracted capacity with actual consumption and identify potentially stranded capacity.

How do colocation providers calculate a safe power oversubscription ratio?

There’s no universal number. A defensible oversubscription ratio comes from measured diversity data — actual peak draw across tenants over time — plus a safety margin calibrated to tenant mix, redundancy design, and SLA exposure. Oversubscription ratios built on nameplate assumptions rather than real consumption data are guesses, not capacity plans.

Modius OpenData provides historical and real-time demand data across tenants, equipment, and sites, giving operators the evidence needed to establish more defensible oversubscription limits.

What does it take to find stranded power across multiple colocation sites?

A real-time operational intelligence platform that unifies data from metered PDUs, branch circuit monitors, BMS systems, and server BMC telemetry into a single cross-site view. The requirement isn’t just monitoring at individual sites — it’s normalizing across different PDU vendors, BMS protocols, and SNMP implementations into a portfolio-level picture that updates continuously, not on a reporting cycle.

Modius OpenData collects and normalizes telemetry from disparate vendors, systems, and protocols into a unified, multi-site view, allowing teams to evaluate available capacity consistently across their portfolio.

Why does nameplate provisioning cause stranded power in multi-tenant data centers?

Colocation contracts are written against the nameplate power rating of equipment tenants plan to deploy. Nameplate ratings represent maximum theoretical draw, not actual operating load — which commonly runs 30–60% below nameplate. Power capacity gets reserved on paper for load that never materializes, and it can’t be resold without real-time proof that the provisioned allocation isn’t being used.

Modius OpenData measures actual equipment and tenant demand over time, helping operators replace nameplate assumptions with defensible consumption data when evaluating available capacity.

Can stranded power be reclaimed without new physical infrastructure?

In most cases, yes. Right-sizing power budgets to measured draw, rebalancing phases, decommissioning ghost load, and implementing dynamic redundancy management are all software and data exercises that recover sellable capacity from infrastructure already in place. Physical upgrades are required specifically for the cooling-constrained problem created by high-density AI workloads — but the measurement and contract-level levers don’t require breaking ground.

Modius OpenData identifies underused capacity, imbalanced loads, abnormal consumption, and other operational conditions that may be corrected before operators invest in additional physical infrastructure.

References:

Uptime Institute. 2024 Global Data Center Survey. 2024. https://datacenter.uptimeinstitute.com/rs/711-RIA-145/images/2024.GlobalDataCenterSurvey.Report.pdf

Koomey, Jonathan, and Jon Taylor. Zombie/Comatose Servers Redux. Anthesis Group, 2015. https://info.anthesisgroup.com/hubfs/Website%20PDFs/Comatose-Servers-Redux.pdf

Data Center Knowledge. “Why AI Data Center Projects Face Years of Delays After Approval.” 2025. https://www.datacenterknowledge.com/energy-power-supply/why-ai-data-center-projects-face-years-of-delays-after-approval

CBRE. North America Data Center Trends H2 2024. 2025. https://www.cbre.com/insights/reports/north-america-data-center-trends-h2-2024