Summary

CyrusOne is a Dallas-headquartered colocation datacenter operator with 55+ facilities across North America and Europe. The company was founded in 2000, spun out of Cincinnati Bell, and became a publicly traded REIT (ticker: CONE) in January 2013. In March 2022, KKR and Global Infrastructure Partners (GIP) took the company private in an all-cash buyout valued at approximately $15 billion ($90.50/share). GIP was subsequently acquired by BlackRock in October 2024, making CyrusOne jointly held by KKR and BlackRock’s infrastructure platform. Though the tags reflect its REIT heritage, CyrusOne is now a private infrastructure company — a fact material to understanding its financing, governance, and disclosure posture.

CyrusOne’s AI infrastructure strategy centers on a purpose-built design platform called Intelliscale, launched in August 2023. Where legacy colocation facilities average 5–10 kW/rack, Intelliscale is engineered for 12–300 kW/rack using hybrid cooling (air, rear-door heat exchanger, direct-to-chip liquid, and immersion), single-story layouts optimized for dense deployment, and modular scalability. The design targets a footprint roughly 25% the size of conventional datacenters for equivalent compute density. By 2025, Intelliscale had moved from product announcement to deployment at scale, with major hyperscale campus projects in Texas and European expansion underway.

CyrusOne’s 2025 operational posture is defined by a “power + land + interconnect” bundling strategy: securing power purchase agreements, powered-land deals adjacent to generation sources, and grid-side partnerships — rather than waiting years for new transmission interconnects — to compress time-to-power for hyperscale customers. The company secured nearly $12 billion in debt financing in 2025, reflecting both the scale of its expansion and the capital intensity of AI-grade infrastructure development. That growth was interrupted in November 2025 by a high-profile cooling failure at its Aurora, Illinois facility that halted CME Group derivatives trading globally for more than 10 hours — a reputational and operational setback that prompted cooling system upgrades and a new Chief Business Officer hire.

Key Facts

  • Founded: 2000 (originally as Cincinnati Bell’s datacenter division)
  • IPO: January 2013 (NASDAQ: CONE)
  • Taken private: March 25, 2022; acquired by KKR and Global Infrastructure Partners for ~$15B ($90.50/share)
  • Current ownership: KKR (50%) + BlackRock/GIP (50%) — GIP acquired by BlackRock, completed October 2024
  • HQ: Dallas, TX
  • Type: Private (formerly public REIT)
  • CEO: Eric Schwartz (appointed September 2022; previously Chief Strategy and Development Officer and EMEA President at Equinix)
  • Revenue (at time of going private): ~$1.2B annualized
  • Portfolio: 55+ data centers in North America and Europe
  • US concentration: Dallas-Fort Worth (14+ facilities, largest single market), plus Northern Virginia, Phoenix, San Antonio, Houston, Chicago/Aurora IL
  • European portfolio: 270+ MW IT capacity across Frankfurt, London, Amsterdam, Madrid, Paris, Dublin
  • Intelliscale platform: Launched August 2023; supports 12–300 kW/rack; single-story layout; hybrid cooling; modular construction
  • AI facility scale: 21 MW to 100+ MW per Intelliscale building
  • Key 2025 Texas projects: DFW7 (200 MW, Fort Worth / Chisholm Grid BESS partnership with Eolian); DFW10 (400 MW, Bosque County / Calpine Thad Hill powered-land deal); DFW17 cluster (Whitney, TX)
  • Key 2025 European project: FRA7, Frankfurt/Griesheim — Intelliscale hyperscale facility; E.ON 61 MW local power generation partnership
  • 2025 financing: ~$12B total — $7.9B warehouse credit facility (Morgan Stanley, TD Securities, KKR Capital Markets); $1.8B revolving credit (Wells Fargo); $687M CMBS loan on DFW1
  • Sustainability target: Climate Neutral by 2030 (SBTi-validated); 100% renewable in Europe; 51% globally (2024)
  • CME outage: November 28, 2025 — Aurora IL facility cooling failure (human error; winterization procedure failure) caused 10+ hour halt of CME Globex derivatives trading globally

What It Is / How It Works

Colocation REIT Business Model

CyrusOne was built on the colocation REIT model: own and operate datacenter real estate, lease physical space and power to enterprise and hyperscale customers on multi-year contracts, and pass through the tax efficiency of a REIT structure to investors. The core product is a guaranteed allocation of datacenter capacity — measured in megawatts, square footage, and cross-connect infrastructure — at a specified SLA. Customers bring their own hardware; CyrusOne provides the power, cooling, physical security, and connectivity.

Three service tiers defined CyrusOne’s commercial structure as a public company and carry forward under private ownership: (1) hyperscale colocation — large, dedicated deployments of 21 MW to 100+ MW for cloud providers (AWS, Google Cloud, Microsoft Azure are publicly confirmed customers); (2) enterprise colocation — retail cage and cabinet deployments for Fortune 1000 companies, typically smaller allocations with more managed services; and (3) build-to-suit — custom-designed facilities developed to a specific customer’s specification, then leased back long-term. Approximately 200 Fortune 1000 companies are CyrusOne customers, and the company has claimed coverage of more than 60% of the Fortune 500.

Going private in 2022 removed the quarterly earnings pressure and public disclosure requirements of REIT status, while providing KKR and BlackRock’s infrastructure capital for the $10B+ expansion program the AI wave demands. The REIT structure itself is no longer operative (CyrusOne is now structured as a private LLC/partnership entity under its PE owners), though the operational model — build, own, and lease — is unchanged.

Intelliscale: AI-Native Facility Design

The defining product evolution of the post-KKR era is Intelliscale, CyrusOne’s AI-purpose-built datacenter platform announced in August 2023. Traditional enterprise colocation buildings are engineered for 5–10 kW/rack — a figure derived from air-cooled x86 servers with moderate TDP. Next-generation AI accelerators (NVIDIA H100, H200, GB200; AMD MI300x, MI350x) draw 700W to 1,000W+ per GPU, and rack-level deployments of 4–8 GPUs per 1U server produce thermal loads that air cooling cannot reject above roughly 20–25 kW/rack without extraordinary airflow. Intelliscale addresses this by redesigning around the AI thermal profile from the ground up.

Key architectural elements of Intelliscale:

Power density range: 12 kW to 300 kW per cabinet, enabling the same building to host both legacy enterprise air-cooled workloads and next-generation liquid-cooled AI clusters in distinct zones. The upper bound of 300 kW/rack is designed for immersion-cooled chassis and future dense GPU configurations; the 50–100 kW/rack range is the near-term target for current direct-to-chip deployments.

Hybrid cooling architecture: Rather than committing to a single cooling modality, Intelliscale supports air cooling, rear-door heat exchanger (RDHx), direct-to-chip liquid (warm water to GPU cold plates), and immersion cooling within the same facility. This allows customers to bring mixed hardware generations and workloads into a single building without stranded cooling infrastructure.

Single-story layout: Intelliscale buildings are single-story to maximize floor-to-ceiling height, simplify structural loading for heavy liquid cooling infrastructure, and allow craning of large modular cooling distribution units directly onto the slab. Multi-story conventional colos impose floor loading constraints and elevator logistics that complicate dense liquid-cooled hardware installation.

Space efficiency: CyrusOne claims Intelliscale facilities occupy approximately 25% of the footprint of conventional datacenters for equivalent compute density. This is primarily a function of eliminating raised floor plenum, reduced aisle spacing enabled by in-row or direct-to-chip cooling, and the absence of large air handlers that consume floor area in traditional designs.

Modular and scalable: The Intelliscale design is modular at the mechanical and electrical layer — using industrially scaled CDUs and cooling distribution manifolds — allowing capacity to be added incrementally as customer demand scales, and allowing design adjustments during construction without full redesign.

Facility scale: Individual Intelliscale buildings are sized from 21 MW to over 100 MW IT load — the hyperscale range. This is not a retail colocation product; it is a wholesale AI campus product.

Power + Land + Interconnect Strategy

The AI datacenter buildout cycle of 2024–2026 has revealed grid interconnect as the primary constraint on datacenter deployment speed in most major US and European markets. New transmission interconnection requests in Northern Virginia, Silicon Valley, and most European capitals face queues of 3–7+ years. CyrusOne’s response — articulated by CEO Eric Schwartz — is a “power + land + interconnect” bundling model that treats existing grid assets as the scarcest input and builds facility strategy around them:

  • Powered-land deals: The Calpine DFW10 partnership (400 MW total; 190 MW Phase 1 July 2025, +210 MW Phase 2 November 2025) places a CyrusOne campus adjacent to the Thad Hill Energy Center gas peaker plant in Bosque County, Texas. Power is sourced via a dedicated agreement, bypassing ERCOT transmission queues. Target: operational Q4 2026.

  • BESS co-location: The Eolian/DFW7 partnership in Fort Worth (200 MW) co-locates the CyrusOne campus at an existing 100 MW battery energy storage site (Chisholm Grid), inheriting the site’s existing high-voltage transmission interconnect while Eolian upgrades its battery plant. This model accelerated projected delivery by 1–2 years versus a greenfield interconnection approach. Ground broken April 2025; initial capacity targeted 2026.

  • Local generation in grid-constrained European markets: The E.ON IQ Energy Center partnership for FRA7 in Frankfurt/Griesheim uses 61 MW of onsite gas engine generation (4.5 MW modules; hydrogen-capable) to bypass Frankfurt’s constrained grid for the initial phase, with waste heat routed to absorption chillers for cooling the datacenter. This allows FRA7 to reach 126 MW IT capacity versus the 81 MW the constrained grid alone could support.

Financing Model Under Private Ownership

Without public REIT equity markets, CyrusOne under KKR/BlackRock finances its buildout primarily through project-level and corporate debt. The $12B in financing capital raised in 2025 reflects the standard private infrastructure development playbook: warehouse credit facilities fund in-progress development; CMBS securitizations monetize stabilized assets (DFW1 CMBS at $687M); revolving credit supports working capital. The sustainability-linked pricing on both the warehouse and revolving facilities (tied to GHG reduction targets) reflects the ESG mandate of BlackRock’s infrastructure platform.

Notable Developments

  • 2026-01: Robert Johnson appointed Chief Business Officer, overseeing operational excellence and customer outcomes — a direct response to the CME outage and reputational reset needed post-incident.
  • 2025-11: Calpine DFW10 Phase 2 announced — additional 210 MW, bringing total campus to 400 MW (Bosque County, TX). Target Q4 2026 operations. (Calpine press release)
  • 2025-11-28: CME Group Globex trading halt — cooling failure at CyrusOne Aurora, IL facility caused by human error (cooling tower winterization procedure not followed ahead of freezing temperatures). Inside temps exceeded 100°F (38°C); >10 hour outage affecting global derivatives markets (gold, oil, interest rates, equity futures). Goldman Sachs paused a $1.3B CyrusOne bond sale in the wake of the incident. (Data Center Dynamics) (Fortune)
  • 2025-07: CyrusOne secures nearly $12B in financing — $7.9B warehouse facility (Morgan Stanley, TD Securities, KKR Capital Markets) + $1.8B revolving credit (Wells Fargo) + $687M CMBS on DFW1. (Data Center Frontier)
  • 2025-07: Calpine DFW10 Phase 1 announced — 190 MW powered-land agreement adjacent to Thad Hill Energy Center in Bosque County, TX. (Calpine press release)
  • 2025-06: E.ON IQ Energy Center partnership announced for FRA7 Frankfurt/Griesheim — 61 MW onsite gas engine generation (H2-capable); expected online 2029. Extends FRA7 IT capacity from 81 MW to 126 MW. (CyrusOne press release) (Data Center Dynamics)
  • 2025-04: Ground broken on DFW7, Fort Worth — 200 MW campus at Chisholm Grid BESS site (Eolian partnership); targeting 2026 delivery. (CyrusOne press release)
  • 2024-10: GIP acquired by BlackRock for $12.5B, completing in October 2024 — CyrusOne’s co-owner GIP now sits within BlackRock’s infrastructure platform.
  • 2023-08: Intelliscale platform launched — AI-purpose-built datacenter design supporting 12–300 kW/rack via hybrid cooling; single-story layout; 25% smaller footprint vs. conventional colos. (BusinessWire)
  • 2022-09: Eric Schwartz named CEO — first post-private-equity CEO; Equinix veteran; restructured strategy around hyperscale AI demand.
  • 2022-03: KKR and Global Infrastructure Partners complete $15B take-private acquisition of CyrusOne. CONE delisted April 2022. (GIP press release)
  • 2013-01: CyrusOne IPO on NASDAQ (CONE); converted from Cincinnati Bell subsidiary to publicly traded REIT.

Key People

Eric Schwartz — CEO

  • Appointed: September 2022
  • Background: 16+ year career at Equinix; served as Chief Strategy and Development Officer; previously President of Equinix EMEA, where he built the region’s market-leading position through investment and acquisitions; also responsible for Equinix’s xScale hyperscaler joint venture program
  • Notes: Schwartz is the architect of Intelliscale and the power-first site selection strategy. His Equinix background gives him direct operational and commercial experience with hyperscale colocation at scale — the customer set CyrusOne is targeting for AI buildout.

Krupal Raval — EVP, Chief Strategy Officer

  • Role: EVP and Chief Strategy Officer; responsible for CyrusOne’s US development programs and strategic engagement with hyperscale customers
  • Notes: The operational lead for the Texas expansion pipeline (DFW7, DFW10, DFW17); hyperscale customer relationships at the capacity planning level

John Hatem — EVP, Chief Operating Officer

  • Role: EVP and Chief Operating Officer; responsible for day-to-day facility operations globally
  • Notes: The Aurora/CME incident falls under operational responsibility; the post-incident cooling upgrades and Robert Johnson hire as CBO reflect changes in the operational hierarchy

Robert Johnson — EVP, Chief Business Officer

  • Appointed: January 2026 (post-CME outage)
  • Role: EVP and Chief Business Officer; responsible for operational excellence and customer outcomes
  • Notes: The timing of this hire — immediately after the CME incident — signals KKR/BlackRock’s response to the reputational risk from the Aurora outage. CBO is an unusual title in datacenter operators; it suggests a role bridging operations and customer success.

Robert M. Jackson — EVP, General Counsel

  • Role: EVP, General Counsel, and Secretary
  • Notes: Leads legal for a significant real estate and infrastructure development portfolio; relevant for complex land, power, and construction contracting at the DFW10/DFW7 scale

Key People — Last Reviewed: 2026-04-02

Supply Chain Position

CyrusOne is a facility owner-operator positioned at the colocation layer of the AI infrastructure stack. Unlike Crusoe (which extends both upstream to power and downstream to cloud/inference), CyrusOne’s core model is real estate and power delivery. The shift toward Intelliscale and powered-land deals extends CyrusOne’s reach slightly upstream (toward power sourcing partnerships) but does not cross into cloud services.

Layer CyrusOne’s role
Power sourcing Partnerships, not direct ownership: Calpine powered-land (DFW10), Eolian BESS co-location (DFW7), E.ON onsite generation (FRA7); ERCOT grid PPAs for most US sites
Facility developer Owner-developer for all CyrusOne campuses; construction executed by contractors (Clune Construction confirmed for Fort Worth campus)
Facility operator Long-term owner-operator; responsible for power delivery, cooling, physical security, SLA compliance
Cooling infrastructure Hybrid: air, RDHx, direct-to-chip liquid, immersion; no proprietary cooling hardware — CyrusOne integrates vendor equipment (Vertiv, Schneider Electric) within the Intelliscale design
Colocation services Hyperscale (21–100+ MW), enterprise, build-to-suit
Key customers AWS, Google Cloud, Microsoft Azure (hyperscale); ~200 Fortune 1000 companies; CME Group (financial services)
Hardware supplied by None — customers bring their own GPU hardware; CyrusOne is hardware-agnostic
Capital structure KKR (50%) + BlackRock/GIP (50%); project-level debt from warehouse facility; CMBS for stabilized assets

⚑ Hyperscaler concentration risk: CyrusOne’s largest revenue contributors are likely the hyperscalers (AWS, Google, Microsoft) who take 21–100+ MW blocks. If any major hyperscaler pauses its colocation expansion in favor of owned facilities or shifts its geographic mix away from CyrusOne’s footprint, utilization risk is significant. Post-private-equity, CyrusOne does not disclose customer revenue concentration.

⚑ Operational execution risk at scale: The CME outage in November 2025 revealed a gap between CyrusOne’s rapid expansion pace and its operational discipline at existing facilities. The Aurora failure was a procedural/human error issue (winterization), not an infrastructure design failure — but at scale, the frequency of such incidents increases proportionally to headcount and facility count. The CBO hire and cooling upgrades are responses, but the root cause is execution management during high-growth operational scaling.

⚑ Power access in constrained markets: CyrusOne’s Intelliscale value proposition requires delivering 30–300 kW/rack, which demands grid power density most legacy datacenter campuses cannot retrofit. The powered-land and BESS co-location strategies solve this for new greenfield sites, but European grid constraints (Frankfurt, London) compress addressable capacity. The E.ON FRA7 partnership is the template for overcoming this; its replication to London and Amsterdam markets is feasible but involves regulatory and permitting complexity in each market.

⚑ Financing and debt capacity: ~$12B in debt capital raised in 2025 on a company valued at ~$15B at acquisition is a high leverage ratio. CyrusOne’s debt service depends on maintaining high occupancy and long-term lease commitments from hyperscaler customers. The Goldman pause on the $1.3B bond sale post-CME, while temporary, illustrates how a single operational incident can affect capital market access for a debt-funded infrastructure company.

Claim Verification

Claim: 300 kW/rack support via Intelliscale

Status: Company-stated; technically credible for immersion cooling architectures; not independently validated at production scale

Supporting:

  • 300 kW/rack is achievable with immersion cooling (single-phase or two-phase dielectric) — the physics support it as the cooling medium (oil/fluid) has much higher thermal capacity than air or even direct-to-chip water
  • CyrusOne’s stated range (12–300 kW) correctly encompasses the full spectrum from legacy air-cooled (12 kW) through direct-to-chip liquid (50–150 kW) to immersion (up to 300+ kW)
  • The 300 kW figure aligns with specifications from immersion cooling vendors like GRC and Submer, as well as NVIDIA’s stated support envelope for future rack-scale architectures

Refuting / questioning:

  • No public reference customer has been named for 300 kW/rack deployments in CyrusOne facilities; it is possible the upper bound reflects engineering design rather than deployed operational experience
  • The range is very wide — a facility designed primarily for 12–25 kW/rack legacy air-cooling is substantially different from one designed for 300 kW/rack immersion; the Intelliscale claim covers both, which may overstate the transformation from conventional colocation
  • High-density deployments require significant structural upgrades (floor loading, CDU manifold infrastructure) that may not be present in all Intelliscale-labeled buildings

Summary: The technical claim is credible; the 300 kW/rack figure should be understood as the design ceiling enabled by immersion cooling, not the average deployed density. Actual AI workload deployments today are primarily in the 25–100 kW/rack range (direct-to-chip for H100/H200 systems).

Claim: Intelliscale occupies “25% of the space of typical data centers”

Status: Company-stated; plausible for specific high-density comparisons; not a general comparison across all workloads

Supporting:

  • At 100–300 kW/rack vs. 5–10 kW/rack legacy, the compute per square meter is indeed 10–30x higher in dense AI facilities
  • Single-story layout, elimination of raised floor plenums, and in-row/direct-to-chip cooling do meaningfully reduce aisle spacing requirements vs. hot/cold aisle air-cooled designs

Refuting / questioning:

  • “Typical datacenter” is not defined; the 25% figure compares a max-density Intelliscale deployment against the median of all enterprise colocation space, which includes low-density legacy configurations
  • For hyperscale air-cooled facilities (which are already highly optimized for density), the differential would be smaller
  • Support infrastructure (electrical switchgear, PDUs, CDUs, dry coolers) does not shrink proportionally with compute density improvements; mechanical room footprint is relatively fixed per MW

Summary: Directionally accurate for the comparison between legacy enterprise colocation and high-density AI facilities, but the 25% figure is a favorable-case framing. A more precise claim would specify the density comparison points.

Claim: 300 kW/rack DFW10 powered-land with Calpine delivering 400 MW by Q4 2026

Status: Public commitments from both CyrusOne and Calpine; construction underway; timeline is aggressive but not implausible

Supporting:

  • Phase 1 (190 MW) announced July 2025; Phase 2 (+210 MW) announced November 2025 — both are signed agreements with Calpine, not aspirational targets
  • “Under construction” confirmed in multiple press releases as of July 2025; Q4 2026 is a ~15–17 month build timeline from announcement, which is tight but achievable for a facility adjacent to existing generation infrastructure
  • Powered-land model removes the grid interconnection bottleneck that typically extends timelines

Refuting / questioning:

  • Texas hyperscale campuses at 400 MW scale have limited precedent for sub-18-month execution; Crusoe’s Abilene Phase 1 at 206 MW took ~12 months from ground-break, but that was a single phase
  • Phase 2 commitment in November 2025 suggests Phase 1 was not yet complete — stacking 400 MW of delivery into a single Q4 2026 window would require parallel construction of both phases, which adds execution complexity
  • Commissioning risk: large cooling systems (especially liquid-cooled AI-density facilities) have historically been commissioning-heavy; DFW10’s cooling profile has not been publicly disclosed

Summary: The commitments are real; Q4 2026 for 400 MW is achievable with executed construction, but should be treated as a best-case target rather than a firm delivery date until Phase 1 is confirmed operational.

Claim: CME outage caused by “human error” / winterization failure

Status: Confirmed by CyrusOne to Bloomberg; consistent with the technical failure mode described

Supporting:

  • Cooling tower winterization (draining towers ahead of freeze to prevent burst pipes) is a standard preventive maintenance procedure; failure to execute is a known and well-documented failure mode in cold-climate facilities
  • The temperature differential described (outside sub-freezing; inside >100°F) is consistent with a total cooling loss scenario where heat rejection was lost but compute loads continued running
  • CyrusOne publicly acknowledged the human error cause via Bloomberg

Refuting / questioning:

  • “Human error” framing places responsibility on individuals rather than systemic process failures (inadequate procedures, training, or oversight) — the two are not mutually exclusive
  • The outage duration (10+ hours) suggests either delayed detection or slow restoration — both are operational process gaps in addition to the root cause failure

Summary: The stated cause (human error / winterization procedure failure) is credible and consistent with evidence. The deeper question for investors and customers is whether CyrusOne’s process controls are adequate for a facility serving systemically important financial market infrastructure (CME Group). The cooling capacity additions and CBO hire are necessary but do not by themselves address process and training gaps.

Sources