WholeReach · industry directory

Data Centers

The companies that define the data centers industry, each read on the same nine-check website rubric — a 0–100 score with the exact fixes — alongside the concepts, the developments and the questions that shape the field.

See the directoryGet your free score

The companies

18 companies in the data centers industry, each graded on the nine things that decide whether a buyer or an AI assistant can find and trust them online. Open any name for the full write-up and the specific fixes.

70+ solid   45–69 gaps   under 45 costing customers

Sites we couldn’t reach

19 businesses whose website blocked our scanner — common when a site filters datacenter traffic, and not a sign the site is broken. A blocked site is never scored zero.

Who defines the data centers market

The firms, funds and institutions that set the terms in data centers — context, not local listings; never scored.

Data Centers guides

Concepts

Critical load vs IT load

IT load is the power drawn by servers, storage and network gear. Critical load is everything the UPS and generators must carry, including IT plus the mechanical and control systems that keep it alive. Leases are written against one or the other, and confusing them misprices a deal.

MW of critical capacity

The unit the industry actually trades in. Watch which megawatt is quoted: design capacity on paper, IT capacity sellable to tenants, and installed or energized capacity ready to power a rack today are three different numbers, and announced campus totals are usually the first.

kW per rack / rack density

Power drawn by a single cabinet. Traditional enterprise colocation ran 3-8 kW; modern AI training racks run an order of magnitude higher. Density, not floor area, determines whether a hall can host a workload at all.

PUE (power usage effectiveness)

Total facility energy divided by IT equipment energy, popularized by The Green Grid. A PUE of 1.0 would mean every watt reaches the servers. It is the industry's headline efficiency metric and also its most gamed one — always ask about annualized, whole-facility, at-load measurement.

WUE (water usage effectiveness)

Litres of water consumed per kilowatt-hour of IT energy. Evaporative cooling trades electricity for water, so a very low PUE can hide a very high WUE — a live tension in drought-exposed Texas.

CUE (carbon usage effectiveness)

Kilograms of CO2 equivalent per kilowatt-hour of IT energy. Unlike PUE it depends on the grid mix and the hours you consume, which is why 24/7 matching and CUE tend to be discussed together.

N+1 and 2N redundancy

N is the capacity needed to carry the load. N+1 adds one spare unit; 2N duplicates the entire system on independent paths; 2N+1 adds a spare to each. Redundancy is what you pay for when you buy uptime, and it multiplies both capex and footprint.

Uptime Institute Tier I-IV

The reliability classification buyers speak in: Tier I basic capacity, Tier II redundant components, Tier III concurrently maintainable, Tier IV fault tolerant. Uptime certifies Design Documents, Constructed Facility and Operational Sustainability separately — a 'Tier III design' is not a certified Tier III building.

Concurrent maintainability

The Tier III property: any capacity component or distribution path can be taken out of service for planned maintenance without dropping IT load. It is about planned work, not failures.

Fault tolerance

The Tier IV property: a single unplanned failure anywhere in the power or cooling system does not affect IT load, with compartmentalization and continuous cooling. Far more expensive, and genuinely required by fewer workloads than sales decks imply.

UPS topology

How the uninterruptible power supply is configured — double conversion (online), line interactive, eco or multi-mode, plus distributed-redundant and block-redundant architectures, and battery versus rotary flywheel. The topology sets both efficiency and the failure modes you inherit.

Static transfer switch (STS)

A solid-state switch that moves a single-corded load between two independent power sources in milliseconds, letting single-power-supply equipment sit on a dual-path design without dropping.

Generator runtime and fuel contracts

Standby generators are sized for the critical load, but the real questions are on-site fuel storage hours, refueling contracts with guaranteed priority delivery, load-bank testing cadence, and air permit limits on runtime hours.

Utility interconnection and the large-load queue

The formal process for connecting a very large new load to the transmission or distribution system: studies, capacity confirmation, facility upgrades and an energization date. In ERCOT the large-load process is the gate every Texas project passes through, and queue position now determines project viability more than site quality.

Load ramp schedule

The agreed curve of how quickly a campus draws its contracted megawatts — often stepped over years. Utilities plan generation and transmission against it, and tenants pay for capacity they have not yet filled.

Curtailable and interruptible load

Capacity a data center agrees to reduce on demand, in exchange for faster interconnection, lower rates or both. Once considered incompatible with data centers, it has become a mainstream way to get connected sooner in constrained grids.

Demand response participation

Programs that pay large loads to shed or shift consumption during scarcity events. Data centers participate via generators, batteries, workload shifting or shutting down flexible compute, turning a cost center into a grid asset.

Behind-the-meter generation

Generation sited on the customer's side of the utility meter — gas turbines, reciprocating engines, fuel cells, solar, storage and prospectively small modular reactors — used to bypass interconnection delays or firm up supply. Regulatory treatment varies sharply by state.

PPA and VPPA

A power purchase agreement contracts for physical delivered energy; a virtual PPA is a financial contract-for-difference that transfers renewable energy certificates without moving electrons. VPPAs are how most data center renewable claims are actually made.

24/7 carbon-free energy vs annual matching

Annual matching buys enough clean generation over a year to offset total consumption. 24/7 CFE requires clean supply in every hour on the same grid. The gap between the two is the honest measure of a sustainability claim.

Substation build and transmission cost allocation

A large campus usually needs a dedicated substation and often transmission upgrades. Who pays — the developer, the utility, or ratepayers through the rate base — is the central regulatory fight in every fast-growing data center market.

Air cooling and economization

Moving heat with air, using outside air or water-side economizers to run chillers less when ambient conditions allow. Cheap and well understood, but it hits a practical ceiling around the low tens of kilowatts per rack.

CRAC vs CRAH

A CRAC is a computer room air conditioner with its own refrigerant compressor. A CRAH is a computer room air handler using chilled water from a central plant. CRAHs dominate at scale; CRACs suit smaller or distributed rooms.

Hot aisle / cold aisle containment

Physically separating supply and return air with doors, roofs or chimneys so hot exhaust cannot mix with cold supply. The single highest-return efficiency retrofit in most legacy halls.

ASHRAE thermal envelope

The recommended and allowable temperature and humidity ranges for IT equipment published by ASHRAE Technical Committee 9.9, with equipment classes A1-A4 and newer high-density and liquid-cooling classes. It is the reference every operator cites when raising setpoints to cut cooling energy.

Chilled water plant and dry coolers

Chillers, pumps, piping and cooling towers producing chilled water for CRAHs or CDUs. Dry coolers reject heat to ambient air without evaporating water — more electricity, far less water, and increasingly chosen in water-stressed regions.

CDU (coolant distribution unit)

The heat exchanger and pumping skid that isolates the facility water loop from the technology loop feeding cold plates, controlling flow, pressure and temperature. It is the interface component that makes a hall liquid-ready.

Direct-to-chip liquid cooling

Cold plates mounted on CPUs and GPUs carry heat away in liquid while air handles the rest of the server. The mainstream path to high-density AI racks because it retrofits into existing halls more readily than immersion.

Immersion cooling

Submerging entire servers in dielectric fluid, single-phase or two-phase. Very high heat capture and excellent efficiency, but it demands different racks, service procedures, fluid handling and vendor warranty support.

Rear-door heat exchanger

A liquid-cooled coil replacing the rack's rear door, capturing heat as air leaves the cabinet. A pragmatic middle step that raises achievable density without re-plumbing the servers themselves.

Water consumption and reclaimed water

Evaporative and adiabatic cooling consume water; closed-loop and air-cooled designs largely do not. Reclaimed, non-potable or effluent water sourcing is now a standard permitting condition in Texas and a routine community question.

Retail colocation

Selling by the cabinet, cage or partial rack with power measured in kilowatts, bundled with interconnection and remote hands. Higher margin per kW, shorter terms, and sold to enterprises and network operators.

Wholesale colocation

Leasing dedicated halls or suites by the megawatt on long terms, with the tenant handling its own IT fit-out. Lower margin per kW, far larger deals, and the format hyperscalers and AI companies actually buy.

Build-to-suit and hyperscale self-build

A developer builds a campus to a single tenant's specification under a long pre-lease, or the hyperscaler buys the land and builds it themselves. Self-build gives control and lowest long-run cost; leasing gives speed, which in a power-constrained market often wins.

Retail colo cabinet pricing

Typically a monthly recurring charge per cabinet for space plus a committed kW of power, with metered power billed as pass-through or included to a cap, then cross-connects, remote hands and bandwidth on top. Comparing quotes requires normalizing all of it to dollars per kW.

Cross-connect and interconnection revenue

Fees for physical circuits between tenants inside a facility. Individually small, collectively the highest-margin recurring revenue in retail colocation, and the reason dense carrier ecosystems are so defensible.

Meet-me room (MMR)

The neutral room where carriers, networks and tenants physically interconnect. Its carrier count and diversity are the clearest measure of a facility's network value.

Carrier neutrality

The operator sells no transport of its own and lets any network in on equal terms. It is what makes a meet-me room worth joining, and it distinguishes independent colocation from telco-owned facilities.

Cloud on-ramp

A private, dedicated port into a public cloud — AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect, Oracle FastConnect — available inside the facility. On-ramp presence is a primary site-selection criterion for hybrid architectures.

Latency and edge

Round-trip time between user or device and compute, bounded by distance and hops. Edge deployments exist to defend a latency budget or to keep data local; where neither applies, 'edge' is usually a marketing label on a regional facility.

Powered shell

A completed building with utility power, substation and structure delivered, but without mechanical and electrical fit-out. Sold or leased to tenants who want to install their own MEP, and a common way to shorten time to energization.

Land banking

Acquiring and holding sites with credible future power for later development. In practice developers are banking interconnection positions and utility relationships as much as dirt.

Entitlement and zoning for data centers

Rezoning, special use permits, site plan approval, noise ordinances, setbacks, water and wastewater commitments and traffic studies. Entitlement risk has risen sharply as local opposition organized, and it now belongs on the schedule alongside interconnection.

Chapter 380/381 abatement (Texas)

Texas Local Government Code Chapter 380 lets municipalities and Chapter 381 lets counties offer economic development agreements, commonly structured as partial property tax rebates over a term. These local agreements, not state incentives, are where most Texas data center tax negotiation actually happens.

Texas data center sales tax exemption

Texas exempts qualifying data center equipment from state sales and use tax for certified projects that meet statutory thresholds for facility size, capital investment, permanent job creation and wage level, with a longer term available to larger qualifying large data center projects. Certification runs through the Texas Comptroller — always confirm current thresholds against the Comptroller's published rules rather than a summary.

Long-lead equipment

Large power transformers, medium-voltage switchgear, generators, chillers and UPS modules with order-to-delivery times now measured in many months to years. Procurement is placed before design is finished, and equipment slots have become tradeable assets.

Commissioning Levels 1-5

The staged proof that a facility works: factory acceptance (L1), component verification on site (L2), pre-functional and point-to-point checks (L3), functional performance of each system (L4), and integrated systems testing of the whole facility under simulated load and failure (L5). Level 5 is where design assumptions get falsified before a tenant moves in.

SLA and uptime credits

The contractual availability commitment and the service credits owed when it is missed. Credits are almost always capped at a share of the monthly recurring charge, so they are a signal of confidence rather than real compensation for a business outage — read the exclusions.

What’s happening

• AI training and inference demand has driven data center absorption to levels the industry has never planned for, converting a steady enterprise build cycle into a capacity race. The practical consequence for buyers is scarcity: vacancy in the major US markets has fallen to record lows and large blocks of contiguous, near-term capacity are frequently pre-leased before a building is topped out. 2026-08 source ↗
• Grid interconnection has replaced land, fibre and construction labor as the binding constraint. ERCOT's large-load interconnection process is now the single gate every Texas project passes through, and queue position, study results and an achievable energization date determine viability more than any site attribute. Announced interconnection requests in ERCOT far exceed what will actually be built, so the queue is a demand signal, not a forecast. 2026-08 source ↗
• Texas legislators moved to formalize how very large loads connect and behave, adding requirements around disclosure of duplicate requests, curtailment obligations during grid emergencies, and cost responsibility for transmission. The net effect is that flexibility — the willingness to be curtailed or to run behind-the-meter generation — has become a bargaining chip for faster interconnection. 2026-08 source ↗
• Large power transformers, medium-voltage switchgear, generators and chillers now carry lead times measured in many months to years. Developers place equipment orders before design completion and before tenants are signed, which shifts risk upstream, favors operators with balance sheets and standing supplier relationships, and makes 'we already own the transformers' a genuine competitive claim worth verifying. 2026-08
• Site selection has inverted. Where fibre routes and tax treatment once led, the first question is now how many megawatts a utility can deliver and in what year. Secondary and tertiary markets with generation and transmission headroom — West Texas, the Panhandle, Temple and the I-35 corridor north of Austin — are drawing projects that a decade ago would have gone to established metros. 2026-08
• On-site and behind-the-meter generation has moved from contingency to strategy. Gas turbines and reciprocating engines are being deployed to bridge interconnection delays, battery storage is being used for ride-through and grid services, and multiple developers have announced interest in nuclear including small modular reactors. Treat all nuclear-adjacent announcements as long-dated intent, not near-term supply. 2026-08 source ↗
• Water use has become a first-order permitting and political issue in drought-exposed Texas. Operators are shifting toward closed-loop, air-cooled and dry-cooler designs and toward reclaimed or non-potable sourcing, and WUE disclosure has moved from voluntary marketing to something municipalities ask for during entitlement. 2026-08
• Community opposition has organized and is winning. Rezonings have been rejected, moratoria and interim development standards adopted, and noise, water draw and ratepayer cost allocation have become standard hearing topics across Texas and nationally. Entitlement risk now belongs on the project schedule next to interconnection, and developers who lead with transparency, noise studies and community benefit terms clear hearings that opaque ones do not. 2026-08
• The economics of incentives are under open debate. The Texas data center sales and use tax exemption and local Chapter 380 and 381 abatements are being weighed against jobs-per-megawatt ratios that are low by industrial standards, and school district and county officials are negotiating harder terms — clawbacks, minimum wage floors, water commitments — than they did five years ago. 2026-08 source ↗
• Liquid cooling has crossed from exotic to expected for high-density AI racks. Direct-to-chip cold plates are the mainstream path because they retrofit into existing halls, immersion holds an efficiency edge with harder operational trade-offs, and CDUs, manifolds, structural loading and facility water loops are now standard items in new-build specifications rather than options. 2026-08
• A large and growing share of legacy colocation inventory physically cannot host modern AI racks. Halls designed around single-digit kilowatts per cabinet lack the power density, floor loading, and water or piping infrastructure to support liquid cooling, which is creating a two-tier market: retrofit-capable assets that re-lease at a premium and stranded ones that do not. 2026-08
• The gap between announced and energized capacity is the most important number in the market and the least reported. Press-released gigawatts routinely combine multi-phase, multi-year campuses contingent on interconnection studies, equipment delivery and tenant demand. Buyers, journalists and investors should distinguish land controlled, power contracted, permits issued, steel erected and megawatts actually energized — and treat any single headline figure as the first of those, not the last. 2026-08
• Central Texas has developed a distinct north-corridor identity along I-35 and US-79. Taylor drew Samsung's semiconductor investment and the accompanying utility and infrastructure buildout, ERCOT's own campus sits in Taylor, and Temple, Round Rock, Pflugerville and Georgetown have all seen data center and large-load interest as developers look for land and power within reach of Austin's talent pool but outside Austin Energy's constrained service territory. Verify each project's status individually — several are announced or entitled rather than under construction. 2026-08 source ↗
• Texas now hosts the full spectrum of the AI buildout in one state: the Abilene campus developed by Crusoe with Lancium as land and power partner and financed through a Blue Owl Capital and Primary Digital Infrastructure joint venture, delivering capacity leased by Oracle for OpenAI's Stargate program; Microsoft's long-established San Antonio campuses; the mature Dallas-Fort Worth cluster anchored by CyrusOne, Digital Realty, Aligned, Compass and DataBank; Austin and Houston colocation from Data Foundry and Element Critical; and speculative Panhandle megaprojects such as Fermi America's Project Matador near Pantex, which is under construction and not operating. Each phase energizes on its own schedule. 2026-08 source ↗

Questions people ask

What is colocation and how is it different from cloud?

colocation sells space, power and cooling for hardware you own; cloud sells compute you rent — explain that most enterprises now run both and the choice is per workload

What does a megawatt of critical capacity actually mean?

distinguish IT load from total facility load and explain why a facility described in megawatts is describing what it can deliver to racks, not what it draws

What is PUE and what is a good number?

define the ratio and explain that climate, load factor and measurement boundary make cross-site comparison unreliable; cite Uptime Institute survey data for any benchmark rather than asserting one

What are Uptime Institute Tiers and what do they certify?

explain concurrent maintainability and fault tolerance, and the difference between a certified facility and a marketing claim of 'Tier III design'

What is the difference between N+1 and 2N redundancy?

spare component versus mirrored system; tie it to what maintenance can be done without dropping load

What is a meet-me room and what is a cross connect?

explain carrier neutrality, why the MMR is where the network value sits, and that cross connects are a recurring revenue line for the operator

What is rack density and why is kW per rack now the headline number?

explain that AI hardware concentrates load per cabinet, so a hall's total megawatts matter less than what a single rack position can take

What is liquid cooling, and how is direct-to-chip different from immersion?

cold plates and CDUs retrofit into existing halls; immersion is more efficient with harder operational trade-offs — explain the facility water loop as the real change

How much water does a data center use?

explain WUE and that consumption depends on cooling design and climate — closed-loop and air-cooled designs use little; never quote a gallons figure for an unnamed facility

What is the ERCOT large load interconnection process?

describe the study stages and the transmission operator's role, and that queue position and study results decide viability more than any site attribute

What is a powered shell?

a building with power and utilities delivered but no data center fit-out; explain who finishes it and why the term is used loosely in marketing

What is the difference between retail colocation, wholesale and build-to-suit?

by cabinet, by hall or by building — explain how contract length, capital responsibility and operating control differ at each step

What is a cloud on-ramp and why does it affect site choice?

private connectivity into a hyperscaler from inside the facility; explain the latency and egress-cost argument in plain terms

What is behind-the-meter generation and why is everyone talking about it?

on-site generation serving load without full reliance on the grid connection; explain it as a bridge to interconnection and note the regulatory questions it raises

How long does it take to get power for a new data center in Texas?

the answer is set by the interconnection study, network upgrades and equipment delivery for the specific point of interconnection; give the stages and refuse to state a lead time

That campus was announced with a big megawatt number — is it real?

teach the ladder — land controlled, power contracted, permits issued, steel erected, megawatts energized — and treat any headline figure as the first rung, not the last

Watch

Run a data centers company?

Get your free website score and the exact list of what to fix — no charge, no pitch. Claim your listing or ask for the full report.