Most searches for the top colocation data center providers return the same ranked list: Equinix, Digital Realty, NTT, CoreSite, CyrusOne, Iron Mountain, QTS. The list is not wrong. It is just answering a question that no longer decides outcomes.
In 2026, the question is not who is biggest. The question is who can deliver power, in your market, on your timeline.
That is a different question. And the answer rearranges the shortlist in ways the conventional rankings do not capture.
North American colocation vacancy has collapsed to roughly 1%, and 92% of capacity currently under construction is already precommitted before a single cabinet is installed.Vacancy rates have held at approximately 1% for a second consecutive year, while nearly two-thirds of new capacity is being developed outside established hubs such as Northern Virginia and Silicon Valley. The supply side has not caught up, and it will not catch up soon. Utility moratoriums and extended interconnection queuesare forcing operators to look beyond primary markets, with state-level incentives in Ohio, Texas, and Idaho reinforcing the shift to secondary geographies.
A short context section, because the rest of the argument depends on it.
The colocation market has structurally inverted. After nearly 15 years of declining rates, the industry has shifted witha global average rate increase of 17% over the last 5 years, according to datacenterHawk intelligence. The deflationary tenant-friendly environment of the 2010s has given way to a supply-constrained seller’s market.
Three forces converged to make this happen. AI workloads created demand the industry had not modelled. The global data center colocation market is expected togrow by 18.6% on an annual basisto reach US$128.6 billion in 2026, with the structural demand driver having fundamentally shifted to AI infrastructure. Hyperscaler preleasing accelerated, with most new capacity committed years before delivery. And power, not real estate, became the binding constraint. Average grid-connection wait times in primary markets now exceed four years.
In Canada, the pattern is the same with a regional twist. The Canada data center colocation market is poised for substantial growth, with an annual increase of 16.5% projected to reach USD 4.22 billion by 2026. Toronto faces tightened vacancy due to hyperscale demand absorption, while Montreal enjoys cost advantages despite its smaller enterprise base. Toronto dominates the existing Canadian market with a power capacity of more than 378 MW, which is around32% of Canada’s existing power capacity, with development land in the GTA now commanding a premium because of grid scarcity.
Any shortlist built on 2022 assumptions will fail in this market. The next five sections explain why.
The premise was true in 2018. It is misleading in 2026.
Equinix, Digital Realty, NTT, QTS, CoreSite, CyrusOne, and Iron Mountain remain the largest names in the market. Equinix, Digital Realty, QTS Data Centers, CoreSite, Cyxtera, and Iron Mountain together account for roughly 45–50% of installed colocation capacity nationwide. But installed capacity is not available capacity. Equinix and Digital Realty maintain dominant positions in retail and wholesale colo respectively, and both are largely sold out of meaningful blocks in primary markets.
What matters for a tenant signing a contract in 2026 is the operator’s near-term delivery pipeline in the specific submarket you need, and the power commitment behind it. A provider with 300 facilities globally is irrelevant if their Toronto, Ashburn, or Silicon Valley positions are at 99% leased and the next phase is three years away. Two providers with smaller global footprints but a powered shell coming online in your timeline are the better shortlist.
The right first question is not “who is biggest.” It is “who has megawatts deliverable in my market within my window?” That answer rarely matches the brand-recognition shortlist.
A decade of marketing trained buyers to weight interconnection ecosystems heavily. Network density, cross-connect counts, on-ramps to major clouds. These features still matter for latency-sensitive retail colocation deployments.
They matter less than they used to for the workloads driving 2026 leasing volume. AI training, GPU inference, and hyperscale enterprise deployments are increasingly power-bound, not network-bound. Operators are now prioritizing proximity to available power and transmission infrastructure over traditional network density alone, with constraints around power availability, land scarcity, and community pushback forcing operators to look beyond core clusters into secondary and tertiary submarkets.
The implication is structural. A provider’s interconnection density should still be weighted for retail and latency-sensitive workloads. For AI, GPU, and hyperscale enterprise workloads, interconnection is a tiebreaker, not a primary criterion. Tenants who built their shortlist around the network exchange leaders are screening the wrong variable for the workload they are actually deploying.
In 2014, locking in a 10-year colocation contract at a fixed escalation was prudent. It hedged against rising rates and gave the operator confidence to build out for you.
In 2026, the same instinct works against you.The deflationary environment thatdefined the 2010s has given way to a supply-constrained seller’s market with significant pricing power, with North America’s average vacancy rate sitting below 2% and many tier 1 markets experiencing sub-1% vacancy. Rates have continued climbing through 2025 and 2026 in primary markets.
The risk has flipped. A long-term contract at today’s market rate locks you in to peak pricing if AI demand normalizes, while a multi-year ramp commitment can leave you holding contracted megawatts you do not yet need. Term length, escalation structure, expansion rights, and partial termination provisions all need to be negotiated against a different risk profile than the playbook the operator’s standard form contemplates. Most tenants are still using the 2018 playbook. Most operators are not.
SOC 2, ISO 27001, PCI DSS, HIPAA, NIST 800-53, TIA-942 Rated 3. The certification list reads the same across every major provider’s marketing page, and it should. These are now table stakes for any provider serving regulated industries.
Treating compliance as a meaningful differentiator in 2026 is a 2015 mental model. Every provider on the conventional shortlist meets the certifications most enterprises need. The actual differentiator is whether the SLA covers the failure modes that matter in dense AI deployments, including power capacity caps, cooling delivery, ramp commitments, and remediation for delayed delivery on a build-to-suit. Standard SLAs were written for retail colocation, and they treat power as effectively unconstrained.
Read the SLA against the failure mode you are actually worried about. If your concern is a 30 kW per rack AI deployment, your SLA needs language on cooling and power density that does not appear in the standard form. The provider’s badge wall is not the answer.
A provider with 200 facilities in 30 countries sounds like geographic diversity. For most enterprise tenants, the practical relevance is narrow. Your workloads sit in two or three specific submarkets, and the diversity of the operator’s global portfolio is not the resilience story for you.
The 2026 resilience question is different. In 2025, Columbus (Ohio), San Antonio, and Boise have seen increased land acquisition and permitting activityfrom major colo developers including QTS Realty and Iron Mountain. State-level incentives in Ohio, Texas, and Idaho are reinforcing this shift, and secondary US markets will gain disproportionate share of new colo supply. The relevant diversity is whether your provider has capacity in the secondary market that will absorb your overflow when Northern Virginia, Silicon Valley, or downtown Toronto run out. That is a different question than counting flags on a corporate map.
For a Canadian tenant, the question becomes more specific again. Toronto absorbed most of its powered capacity. Cologix is the largest domestic colocation operator by the number of Canadian facilities, with Equinix operating in Toronto and Vancouver, and Montreal’s hydroelectric advantage is increasingly the overflow market for GTA-based enterprises whose Toronto requirements cannot be filled in their timeline. A US-headquartered provider with extensive global reach but limited Canadian power positions does not solve a Canadian tenant’s problem.
A shortlist that actually serves the tenant in 2026 inverts the conventional process. Start with the workload, the submarket, and the timeline. Then test which operators can credibly deliver against those constraints, with the SLA structure to match.
The questions that belong on the shortlist:
That shortlist looks nothing like the brand-recognition list. It often surfaces operators that do not appear in the top of conventional rankings. It excludes operators whose marketed footprint is large but whose deliverable capacity in your market is functionally zero. And it makes the contracting conversation a power conversation first, a real estate conversation second.
This is the shift in selection methodology that separates a 2026 shortlist from a 2022 one. Engaging a tenant-rep data centre consultant early, before requirements are circulated to providers, is how that methodology gets executed without giving up negotiating leverage.
Not sure where you stand on data center capacity availability in your market? Get a free assessment of your current position before your next RFP.
The CFO question is not which provider has the strongest brand. It is whether your data center commitment is sized correctly, structured correctly, and timed correctly for a market where capacity is scarce, pricing has inverted, and contract terms favour the operator’s standard form.
The default path is to issue an RFP to the conventional shortlist, evaluate on a familiar scoring matrix, and sign a long-term contract at quoted rates. The cost of that default in 2026 is paying peak market pricing on a contract structure that does not flex with workload uncertainty, with delivery exposure if your selected provider’s pipeline slips. Operators are aware of this dynamic. Most tenants are not.
A well-run procurement process starts from workload and submarket constraints, builds a shortlist against deliverable capacity, runs a competitive process that puts pricing pressure on the operator before exclusive negotiations begin, and structures the contract to preserve flexibility on the variables that actually matter. The provider name on the final contract is the last decision, not the first.
Your questions answered
They belong on the long list. They are not automatically the right shortlist for any given requirement.
These operators continue to lead in installed capacity and global brand recognition. Equinix, Digital Realty, NTT Data Centers, KDDI, and QTS accounted for roughly 20–25% of the global share in 2025. But the shortlist that wins in 2026 is built from deliverable megawatts in your submarket on your timeline, and that shortlist often includes regional operators or specialty hyperscale developers that do not lead the conventional rankings.
Start the search by defining the workload and the submarket. The provider names follow from that, not the other way around.
Eighteen to twenty-four months earlier than the playbook recommended five years ago.
Enterprises that once planned infrastructure 6 to 12 months ahead are now securing capacity 18 to 24 months before deployment. For larger commitments, particularly anything involving build-to-suit elements or significant power blocks, the planning horizon stretches to three or four years.
The tenants who treat colocation procurement as a near-term decision are arriving late to a market where the capacity they need is already preleased. Start the process well before you think you need to.
It applies to both, with different intensity.
Retail colocation in the sub-megawatt range still has more available inventory than the hyperscale segment, and the procurement timeline is shorter. But the same pricing pressure is reaching retail tenants in primary markets, and the gap between operators’ marketed inventory and their actually deliverable inventory has widened across both segments. Approximately 73% of pipeline developments are preleased, and vacancy is expected to remain below 5% through 2027.
The methodology is the same. The urgency scales with the size and power density of the deployment.
The distinction has not changed fundamentally. Colocation means you own the hardware and rent space, power, and cooling. Cloud means you rent the compute itself and the provider owns the underlying infrastructure.
What has changed is that the economic case for repatriating workloads from public cloud back to colocation has strengthened for certain workloads. A portion of US enterprises that migrated workloads to public cloud are selectively returning compute and storage to colo environments on cost and control grounds. Financial services firms and healthcare providers are among the most active segments executing hybrid strategies.
Heavy compute, GPU-driven, predictable-load, and cost-sensitive workloads now look better in colocation than they did three years ago. Variable and bursty workloads still favour cloud.
Probably. The market has shifted enough that contracts signed before 2023 were priced against very different supply assumptions. We routinely see incumbent tenants paying above current market for their existing footprint and below current market for their expansion blocks, simultaneously.
Renewals also tend to favour the operator because the tenant’s switching cost is real and the operator knows it. A competitive process at renewal, even if you ultimately stay in place, is the single highest-leverage action available to a colocation tenant. Data centre tenant representation at renewal is the same discipline as at a first contract, and the savings opportunity is typically larger.
Canada is smaller, more concentrated, and more power-advantaged in Quebec.
Toronto is tight on available megawatts and increasingly looking to peripheral GTA submarkets for new capacity. Montreal offers a meaningful cost advantage on power because of hydroelectric supply, and is increasingly the overflow market for Toronto-based enterprises with AI or GPU workloads where latency to Toronto users is not strictly required.
For a Canadian-headquartered tenant, evaluating providers means evaluating their Canadian power positions specifically, not their global footprint. The brand-name US providers with limited Canadian capacity are not the right answer for a Canadian deployment, even if they dominate the broader rankings.
For a corporate tenant, nothing direct. ENCOR’s data centre consulting practice is compensated by the operator at the close of the transaction, the same fee structure that applies in conventional commercial real estate brokerage. The tenant pays no out-of-pocket fee for representation.
The economic question is the opposite of “can we afford an advisor.” It is whether the absence of one is costing more than the cost of engaging one. In a market where the operator’s information asymmetry is structurally greater than it was five years ago, the answer is usually clear.
