Colocation in Phoenix: operators, ecosystem and what the power constraint means

Start here: two operators that are not in this market#

Ask a search engine or an AI assistant who provides colocation in Phoenix and you will very likely be told about Equinix. This is worth correcting first, because it is the most common wasted week in a Phoenix search.

The reason this error is so persistent is structural rather than careless: the largest operators generate most of the published writing about colocation, so they dominate any answer assembled from that writing, whether or not they operate in the metro being asked about. The longer version of that problem is worth reading if you are relying on AI answers for shortlists.

What Phoenix does have is a deep, unusually varied operator set — a downtown carrier hotel, several very large suburban campuses, a strong mid-tier, and one of the largest single hyperscale campuses in the western US.

The operator landscape#

Every facility below is drawn from our own reconciled Phoenix directory, verified against operator-published material as of 26 August 2026. Square footage and megawatt figures are the operator's own unless marked as a third-party estimate. None of these figures says anything about what is available — see the note at the end of this page on why we do not publish that.

Retail and interconnection — where most buyers under a megawatt will land#

Wholesale and hyperscale#

Operators we have corrected out of this market#

These appear in third-party listings and search results and should not be on a Phoenix bid list. We keep them visible rather than deleting them, because the correction is more useful to you than a silently shorter table.

• DataBank — apparent exit. DataBank's own current facility directory (checked 26 August 2026) lists no Phoenix site anywhere in its West region, while third-party aggregators still carry a downtown Phoenix building as active and separately flag a second Phoenix site as decommissioned. Read together this is the signature of a quiet market exit that the listings have not caught up with. We do not have a DataBank confirmation either way, so we flag rather than delete — but do not treat DataBank as having active Phoenix capacity without asking them directly.
• DataBank at 2121 S Price Rd, Chandler — withdrawn outright. No verified source places DataBank at that address; it belongs to Digital Realty's PHX15. This looks like an aggregator conflating two operators at one address, and it is the clearest single example of why address-keyed listings mislead.
• 365 Data Centers — exited around 2016–17. The chain of title is fully traceable: Switch and Data to Equinix in 2010, Equinix divesting sixteen US sites to 365's predecessor in 2012 with Phoenix among them, and 365 exiting a few years later. Its only current Arizona presence is a network node inside another operator's building, which is not colocation supply.

Operators present in the metro that our verified corpus does not yet cover#

These are named because omitting them would make this guide quietly narrower than the market. They came from an earlier open-web pass and have not been through the same verification as everything above, so we state presence and nothing more: QTS (two Phoenix sites, one sharing the downtown carrier hotel), Vantage Data Centers (Goodyear), STACK Infrastructure (Phoenix and Avondale), EdgeCore (Mesa), Stream Data Centers (Goodyear), H5 Data Centers (Chandler), Edged (Ahwatukee), and Aligned (Phoenix). Most are wholesale or hyperscale developers. Treat this list as a prompt to ask, not as verified detail.

Network and cloud ecosystem#

Phoenix's interconnection story concentrates around two poles, and the distinction matters when you are choosing a building.

120 East Van Buren Street is the downtown carrier hotel and the metro's densest interconnection point. It hosts Digital Realty's PHX10 and a QTS facility in separate suites. If your requirement is carrier choice and cross-connect density, this is the centre of gravity, and other operators' buildings are measured by how well they reach it — Evocative, for instance, publishes dark fibre from its Union Hills facility to this building specifically.

The Sky Harbor cluster around 615 N 48th Street and East Van Buren is the second pole, holding Iron Mountain's AZP-1 and AZP-2 and Csquare's PHX1. The Iron Mountain campus is joined by a 100G fibre ring spanning AZP-1, AZP-2 and Scottsdale's AZS-1, which makes those three buildings function as a single deployment for resilience design.

Confirmed cloud on-ramps#

This is where Phoenix rewards precision, because the market's marketing is looser than its facts. Two facility-level confirmations exist in our corpus:

• EdgeConneX PHX01 (Tempe) — AWS Direct Connect and Azure ExpressRoute confirmed as peering locations at this building.
• phoenixNAP's Phoenix facility — the first AWS Direct Connect location in Arizona (2017), and an Azure ExpressRoute peering location. Note the ownership transition above.

Everything else you will be told about cloud connectivity in this metro is either reachability over an operator's own backbone, or access through an interconnection partner such as Megaport, PacketFabric or Console Connect. Both are legitimate and widely used; they differ from a same-building on-ramp in latency, in hop count and in who you call when it breaks. Ask specifically whether the on-ramp for your cloud region is physically in the building you are being quoted, or reached over a network.

For carrier presence, Evocative publishes AT&T, Lumen (Level 3), Cogent, Verizon, Zayo and Cox as on-net at its Union Hills facility — one of the few operator-published carrier lists in our Phoenix set. Everywhere else, ask for the on-net list for the specific building rather than the operator's portfolio claim.

The power constraint, and what it actually means for you#

Phoenix's defining market condition is no longer land or fibre. It is generation and transmission capacity, and it now shapes every conversation about growth in this metro.

The scale of the imbalance is the part worth internalising. Arizona Public Service peaks at roughly 8,200 MW system-wide, and has reported cumulative data-center interconnection requests of roughly 30,000 MW — around 4,500 MW already committed and roughly 19,000 MW in active discussion. Arizona's Energy Promise Task Force has reported that combined proposed data-center load across APS and SRP could nearly triple those utilities' current demand.

Both utilities have responded with gatekeeping rather than expansion promises: queue triage, a proposed rate class for extremely high load factors, a "bring your own power" policy, and — at SRP — a large-customer process requiring up-front funding of utility upgrades and minimum billing commitments. Separately, Arizona enacted a three-year pause on new data-center tax-incentive applications running from July 2026 to June 2029. That is a tax measure, not a construction ban, and the two are frequently conflated in coverage.

What this changes about how you buy here#

• Existing, energised capacity is worth more than campus square footage. An operator with committed feed capacity already in place is in a materially different position from one whose expansion depends on a new utility interconnection. Ask which one you are buying.
• "We have 169 MW on this campus" is not an availability statement. No public source states how much of any Phoenix campus is unallocated, and campus size tells you nothing about what you can have.
• Ask about the utility interconnection position directly for anything that depends on new power: is the capacity already energised, is it in queue, and where in the queue.
• Expect longer lead times on growth than the market's marketing implies, and negotiate expansion rights at signature rather than assuming you can add later.

When Phoenix is the right metro — and when it is not#

Phoenix fits when#

• You need western US presence without California costs or California seismic exposure. This is the metro's oldest and still strongest argument, and it is why much of the existing footprint is here.
• You want a disaster-recovery site paired with a California or Pacific Northwest primary — far enough for genuine geographic separation, close enough for acceptable latency.
• Your requirement is in the roughly two-to-forty-cabinet range and you want real competition. The Phoenix mid-tier is unusually deep, and several of these operators will compete hard for a deployment that a global operator would price indifferently.
• You need very large contiguous capacity and are prepared to engage wholesale operators directly — the Chandler and West Valley campuses are large by any standard.

Phoenix does not fit when#

• Your growth path depends on adding significant power in-metro over the next few years. The constraint above is real and it is the single best reason to look elsewhere. Ask any operator quoting you here what their expansion position actually is.
• You need the deepest possible carrier and peering ecosystem. Phoenix has a genuine carrier hotel and reasonable density, but it is not Ashburn, Dallas or One Wilshire, and a network-density-driven requirement should be tested against those markets.
• You need an operator with a global footprint under a single contract. The two operators most associated with that model are not in this market.
• Water usage is a hard procurement constraint and you have not yet asked how a specific facility cools. Approaches vary widely between buildings here, and this is a live scrutiny issue in Arizona.

When the big brands genuinely are the right answer#

Neutrality cuts both ways, and this guide would be dishonest if it only argued downward. Digital Realty's PHX10 is the correct answer when downtown interconnection density is the actual requirement — it is the densest point in the metro and nothing in the mid-tier substitutes for it. Iron Mountain is the correct answer when physical security posture, compliance-heavy operations or a genuine third-party Tier certification is a gating requirement; very few operators in any market hold real Uptime certification, and AZP-1 does. And CyrusOne's Chandler campus is the correct answer for a multi-megawatt requirement that needs contiguous scale, provided you can reach them directly and accept a non-retail commercial process.

The argument this page makes is not that large operators are wrong. It is that they should be on your list because they fit, rather than because they are the names that surface first.

Pricing context for Phoenix#

We do not publish per-operator Phoenix pricing, and it is worth being precise about why: no operator in this market publishes a rate card. Pricing across every operator in our corpus — global, REIT, mid-tier and regional alike — is quote-only and requires a sales engineer. There is no list price to check against and none to negotiate down from. The specific dollar figures circulating online for this metro trace back to community forum posts and aggregator pages, several of them years old, and quoting them would hand you false precision.

For orientation only: CBRE's H2 2025 North American data-center report put retail colocation asking rates at roughly $196 per kW per month for deployments in the 250–500 kW range, against a record-low 1.6% vacancy rate across North America. That is a continental blend of very different markets and an asking rate rather than a transacted one — useful for calibration, not as a number to expect.

Two Phoenix-specific commercial observations that are more actionable than any number. First, the tier spread here is genuinely wide: the metro holds premium interconnection space, compliance-premium space, and mid-tier operators positioned well below both, which means a bid run across tiers in Phoenix produces more spread than one run across three global operators. Second, in a constrained-power market, terms other than the headline rate — expansion rights, the utility interconnection position, escalators on renewal — carry more value than they would in a soft market, and are frequently more negotiable than the rate itself.

The full process, including the line items to demand so quotes come back comparable, is in how to run a competitive colocation bid.

What this page deliberately does not tell you#

It does not tell you who has space.

Every capacity figure on this page describes a building — how large it is, how much power it was built to carry. None of them describes what is unallocated, and the difference between those two things is the entire question you are actually asking. Colocation availability decays in weeks: an operator that had eight cabinets last quarter may have none now, and one that was full may have had a customer leave.

There is no honest way to publish that on a page. It is answerable only by asking every relevant operator, at the moment you need to know, on your specific requirement — and writing down what each one said and when. That is most of what a colocation search actually consists of, and it is the work this page is designed to make faster rather than to replace. The sample reports show what the answers look like when they come back.