Delays, Not Cancellations: Used Gear as Strategy

The debate over whether the AI data center boom is about to burst got most of the attention this quarter. The more consequential story is quieter: the buildout is not being cancelled, it is being delayed. Demand for AI capacity remains exceptionally strong, but permitting, power, and long-lead equipment are stretching project timelines well beyond what operators planned. That gap between demand and deliverable capacity changes the economics of existing hardware. When you cannot build new capacity on schedule, the gear already in your racks, and the quality used gear available on the secondary market, stops being a compromise and becomes a strategy. Here is the 2026 picture, and what it means for how operators should think about their installed base.

TL;DR

The data center buildout is supply-constrained, not demand-constrained, and that reframes the value of existing and used hardware. The essentials:

Infrastructure spending is still surging. The worldwide data center physical infrastructure (DCPI) market grew 27% year-over-year to $14 billion in Q2 2026, extending a run of above-20% growth, according to Dell’Oro Group.
The constraint is permitting and power, not demand. Dell’Oro’s research director framed it directly: the loud debate was whether the buildout is being cancelled; the quieter, more consequential one is whether it will be permitted. Demand is not in doubt; timelines are.
Delays, not cancellations, is the story. Power, long-lead equipment, grid interconnection, and community opposition are all stretching project schedules, which Dell’Oro now regards as the most serious near-term brake on the market.
Timelines are now harder to predict than demand. The timing of permits and grid interconnections is genuinely difficult to call, even as the hyperscalers’ spending plans leave little doubt about the underlying demand.
This changes the value of existing assets. When new capacity cannot arrive on schedule, sweating existing hardware longer and sourcing quality used equipment to extend or expand capacity becomes a deliberate strategy, not a fallback.

The reframe for operators: in a market where demand outpaces the ability to build, the binding constraint is time-to-capacity. Hardware that is already operating, or that can be deployed immediately from the secondary market without waiting on a permit or an interconnection queue, has a strategic value that a pure new-build mindset misses. The buildout delays are turning the installed base and the secondary market into levers, not afterthoughts.

What Dell’Oro Actually Reported

The data point anchoring this is Dell’Oro Group’s Q2 2026 Data Center Physical Infrastructure report, released October 14, 2026. The headline numbers describe a market still growing fast.

Metric	Figure	Source
DCPI market revenue, Q2 2026	$14 billion	Dell’Oro Group
Year-over-year growth	27%	Dell’Oro Group
Growth streak	Above 20% for multiple consecutive quarters	Dell’Oro Group
Fastest-growing segment	Thermal management, up more than 40% Y/Y	Dell’Oro Group
Fastest-growing customer segment	Other Cloud (neoclouds and AI developers)	Dell’Oro Group
Market center of gravity	North America	Dell’Oro Group

The growth itself is not the news; the market has been expanding above 20% for several quarters as AI capacity demand runs ahead of supply. The modest slowdown from the prior quarter reflected a tougher year-over-year comparison, not softer demand. What made this report notable was the framing Dell’Oro put around those numbers.

As Alex Cordovil, Research Director at Dell’Oro Group, put it, the quarter’s loudest debate was whether the AI data center buildout is being cancelled; the quieter and more consequential one is whether it will be permitted. He went further: the hyperscalers’ spending plans leave little doubt about demand, but power, long-lead equipment, and community opposition are all stretching project timelines, and Dell’Oro now regards siting and permitting as the most serious near-term brake on the market. The timing of permits and grid interconnections, he noted, is harder to call than the demand behind them.

That is the whole story in one line: the risk to the buildout is not that the demand evaporates. It is that the capacity cannot be delivered on schedule.

Delays, Not Cancellations: What Is Actually Slowing the Buildout

The “bubble” narrative frames the risk as demand collapsing. The data points the other way: demand is robust, and the brakes are all on the supply side. Several constraints are stretching timelines at once.

Permitting and Siting

Permitting has emerged as the single most-cited near-term constraint. Securing approvals for large new data center campuses takes time, and the process has lengthened as the scale and number of projects has grown. In one illustrative move, Texas ordered an audit of data center projects in the ERCOT interconnection queue, delaying the first batch of large-load studies. When the approvals process itself becomes the bottleneck, even fully-funded, high-demand projects slip.

Power and Grid Interconnection

Power availability and the time to connect to the grid are the other dominant constraint, a theme that runs through every fast-growing data center market. Interconnection queues stretch years, new generation and transmission take years to build, and the largest AI campuses require power at a scale that the grid cannot always deliver on the operator’s timeline. The result is that a project can be ready to build with the capital in hand and still wait on the power to run it.

Long-Lead Equipment

Even the physical equipment is a constraint. Certain categories of data center infrastructure, from large transformers to specialized cooling and power-train components, carry long lead times, and supply, rather than demand, has become the binding constraint on some of them. Dell’Oro noted, for instance, that busway remained the preferred overhead power train for AI clusters with supply, not demand, as the limiting factor. When the components themselves are backordered, the build waits.

Community Opposition

Finally, local opposition has grown into a real factor in project timing, particularly in the United States, where community pushback over power costs, water use, and land has become a genuine variable in whether and when projects proceed. This is the social layer of the permitting constraint, and it adds uncertainty that is difficult to price into a timeline.

The common thread across all four is time. None of these constraints suggests the demand is not there. All of them stretch the gap between wanting capacity and having it running.

Why Delays Change the Economics of Existing Hardware

Here is the shift that matters for infrastructure strategy, and the one the buildout-delay story implies without stating. When new capacity cannot be delivered on schedule, the value of capacity you already have, or can deploy immediately, goes up.

Time-to-Capacity Becomes the Binding Constraint

In a demand-constrained market, the question is whether there is enough demand to justify building. In a supply-constrained market, which is what Dell’Oro is describing, the question is how fast capacity can be brought online at all. That reframes every asset decision around time-to-capacity. A server that is already running, or a rack of quality used hardware that can be deployed next week, delivers capacity on a timeline that a new build waiting on a permit or an interconnection study simply cannot match. When time is the scarce resource, immediately-available hardware carries a premium that a pure new-build accounting misses.

Sweating Existing Assets Becomes Deliberate

Operators facing stretched timelines for new capacity have a rational incentive to extend the service life of the hardware they already run. Keeping existing infrastructure productive longer, rather than retiring it on the default refresh schedule, bridges the gap until new capacity arrives. This is not a failure to modernize; it is a deliberate response to a market where the replacement capacity cannot be guaranteed to arrive on time. Sweating assets, done well, is a timeline strategy.

Quality Used Gear Becomes a Sourcing Strategy

The same logic extends to acquisition. When new equipment carries long lead times and new capacity carries permitting and power delays, quality used hardware from the secondary market becomes a genuine sourcing strategy rather than a budget compromise. Tested, certified, immediately-available used gear, networking, compute, storage, and the supporting infrastructure, can be deployed without the lead times that constrain new equipment. For capacity expansions, lab and test environments, redundancy, and bridging builds, the secondary market offers something the new market cannot in a constrained environment: speed. The operator who can source quality used gear quickly can add capacity while the new-build queue is still waiting on approvals.

The Residual-Value Picture Strengthens

There is a corollary for operators retiring hardware. If used gear is becoming a strategic sourcing channel for buyers, the hardware that operators retire holds its value better, because demand for quality secondary-market equipment is supported by exactly the buildout constraints slowing new supply. The residual value in a retiring fleet is strengthened, not weakened, by a market where immediately-available capacity is scarce. That makes informed asset recovery, capturing that residual value through proper remarketing rather than treating retired gear as surplus, more rewarding in a constrained environment.

What Operators Can Reasonably Do

The buildout-delay environment suggests a set of practical moves, all oriented around treating existing and secondary-market hardware as strategic levers rather than afterthoughts.

Plan capacity around time-to-deployment, not just cost. In a constrained market, how fast capacity can be brought online matters as much as its sticker price. Weigh the immediate availability of existing and used hardware against the real timeline for new builds, including permits and interconnection.
Extend service life deliberately where it bridges a gap. Rather than retiring hardware on a default schedule, assess whether sweating specific assets longer bridges the gap until delayed new capacity arrives. Make it a decision, not a drift.
Treat the secondary market as a sourcing channel. For capacity expansions, redundancy, test environments, and bridging builds, quality tested used gear can deliver capacity on a timeline new equipment cannot. Build a relationship with a disposition and recovery partner who can source it.
Capture residual value on what you do retire. In an environment where secondary-market demand is strong, the hardware you retire holds value. Recover it through informed remarketing rather than default disposal, and plan retirement to preserve that value.
Keep disposition and acquisition connected. The same market dynamics that make used gear a smart acquisition make retired gear a valuable asset. An operator who sees both sides, what to keep, what to source used, and what to retire for value, is positioned to turn the buildout constraints into an advantage.

The unifying idea is that buildout delays reward operators who think about their installed base and the secondary market as strategic resources. In a market where new capacity is slow and demand is strong, the hardware already available, in your racks or on the secondary market, is one of the few levers that moves on your timeline rather than the permitting queue’s.

Frequently Asked Questions

Is the AI data center buildout slowing down?

Not in demand terms. According to Dell’Oro Group’s Q2 2026 report, the data center physical infrastructure market grew 27% year-over-year to $14 billion, extending a multi-quarter run of above-20% growth. What is slowing is the pace at which new capacity can actually be delivered. Dell’Oro’s research director framed the real debate as not whether the buildout is being cancelled, but whether it will be permitted. Demand remains exceptionally strong; permitting, power, long-lead equipment, and community opposition are stretching project timelines. The accurate description is delays, not cancellations: the buildout is supply-constrained, not demand-constrained.

What is causing data center buildout delays in 2026?

Four constraints are stretching timelines simultaneously. Permitting and siting have emerged as the single most-cited near-term constraint, with approvals processes lengthening as project scale and number grow. Power availability and grid interconnection are the other dominant brake, with interconnection queues stretching years. Long-lead equipment, from transformers to specialized power-train and cooling components, carries supply constraints that delay builds even when funded. And community opposition over power, water, and land has become a real factor in project timing, particularly in the US. None of these reflects weak demand; all of them stretch the gap between wanting capacity and having it running.

What is the data center physical infrastructure (DCPI) market?

The data center physical infrastructure market covers the power, cooling, racks, and supporting physical systems that data centers require, as distinct from the servers and networking gear inside them. Dell’Oro Group tracks it quarterly. In Q2 2026, the market grew 27% year-over-year to $14 billion in manufacturer revenue, with thermal management (led by direct liquid cooling) the fastest-growing segment at more than 40% growth, reflecting the cooling demands of high-density AI hardware. The market has grown above 20% for several consecutive quarters as AI capacity demand runs ahead of supply.

Why does a supply-constrained market make used data center gear more valuable?

Because the binding constraint becomes time-to-capacity rather than cost. When new capacity cannot be delivered on schedule due to permitting, power, and equipment delays, hardware that is immediately available carries a strategic premium. Quality used gear from the secondary market can be deployed without the long lead times that constrain new equipment, so it becomes a genuine sourcing strategy for capacity expansions, redundancy, test environments, and bridging builds, rather than a budget compromise. The same dynamic strengthens the residual value of hardware operators retire, because secondary-market demand is supported by exactly the constraints slowing new supply.

Should operators extend the life of existing data center hardware?

In a buildout-delay environment, extending service life deliberately can be a sound strategy rather than a failure to modernize. When new replacement capacity faces uncertain timelines due to permitting and power constraints, sweating existing assets longer bridges the gap until that capacity arrives. The key is to make it a deliberate decision based on where it genuinely bridges a gap, rather than an unplanned drift. Operators should weigh the reliability and efficiency of specific existing assets against the real, permit-and-interconnection-inclusive timeline for their replacements, and extend life where the math favors it.

Does buying used data center equipment mean compromising on quality?

Not when it is sourced properly. Quality used data center hardware, tested, certified, and sourced through a reputable disposition and recovery partner, can deliver reliable capacity on a timeline new equipment cannot match in a constrained market. The shift in 2026 is that buildout delays have changed the calculation: used gear is increasingly chosen for its immediate availability and strategic fit, not merely its lower cost. For capacity expansions, redundancy, lab and test environments, and bridging builds while new capacity is permitted, quality used equipment is a deliberate strategic choice rather than a compromise.

How do buildout delays affect the value of retired data center hardware?

They strengthen it. If quality used gear is becoming a strategic sourcing channel because new capacity and new equipment both face delays, then demand for secondary-market hardware is supported by the same constraints slowing new supply. That means the hardware operators retire holds its value better, and the residual value in a retiring fleet is strengthened rather than eroded. For operators, this makes informed asset recovery, capturing that value through proper remarketing rather than treating retired equipment as surplus, more rewarding in a constrained market than in a loose one.

What should operators do differently in a delays-not-cancellations market?

Treat existing and secondary-market hardware as strategic levers. Plan capacity around time-to-deployment, not just cost, weighing the immediate availability of existing and used gear against real new-build timelines. Extend the service life of specific assets deliberately where it bridges a gap until delayed capacity arrives. Use the secondary market as a genuine sourcing channel for expansions, redundancy, and bridging builds. Capture residual value on retired hardware through informed remarketing. And keep disposition and acquisition connected, since the same dynamics that make used gear smart to buy make retired gear valuable to recover. The installed base and the secondary market move on the operator’s timeline, not the permitting queue’s.

Is this a sign the AI bubble is bursting?

The Dell’Oro data suggests the opposite of a demand collapse. Infrastructure spending grew 27% year-over-year, demand for AI capacity continues to outpace supply, and the hyperscalers’ spending plans leave little doubt about demand. The risk Dell’Oro identifies is not that demand disappears but that capacity cannot be delivered on schedule because of permitting, power, and equipment constraints. That is a supply-side timing problem, not a demand-side bust. Whether the broader AI investment cycle is sustainable is a separate and genuinely debated question, but the specific signal in this quarter’s infrastructure data is constrained supply meeting strong demand, not the reverse.

The Bottom Line

The loudest debate about the AI data center buildout this quarter was whether it is being cancelled. The more consequential story, as Dell’Oro Group’s Q2 2026 data makes clear, is that it is being delayed. Infrastructure spending grew 27% year-over-year to $14 billion, demand for AI capacity continues to outrun supply, and the hyperscalers’ commitments leave little doubt about the underlying demand. What is stretching is the timeline: permitting, power, grid interconnection, long-lead equipment, and community opposition are all pushing project schedules out, and the timing of permits and interconnections is now harder to predict than the demand behind them. Delays, not cancellations.

For infrastructure operators, that distinction reframes the value of hardware. In a demand-constrained market, the question is whether to build. In a supply-constrained one, the question is how fast capacity can be brought online at all, and that makes time-to-capacity the binding constraint. Hardware already running, and quality used hardware that can be deployed immediately from the secondary market, delivers capacity on a timeline that a new build waiting in a permitting or interconnection queue cannot match. Sweating existing assets longer becomes a deliberate bridging strategy. Sourcing quality used gear becomes a legitimate acquisition channel rather than a compromise. And the residual value of retired hardware strengthens, because secondary-market demand rests on the very constraints slowing new supply. When the buildout is gated by time rather than demand, the installed base and the secondary market stop being afterthoughts and become strategy.

How ROC Telecom Helps

ROC Telecom is an R2v3, RIOS, NIST 800-88, and ITAR-compliant ITAD specialist, positioned on exactly the side of this market that buildout delays make more valuable: the installed base and the secondary market:

Specialist asset recovery and remarketing across routing, switching, optical transport, compute, and storage, capturing the residual value in retired hardware that a constrained secondary market rewards
Deep domestic buyer and supply network that connects retiring operators with buyers sourcing quality used gear, on both sides of a market where immediate availability carries a premium
GPU and AI infrastructure recovery for the high-value hardware moving through compressed refresh cycles, with speed-to-remarketing that protects value
NIST 800-88 data destruction with per-asset serialized Certificates of Destruction and full chain-of-custody documentation, so hardware can move into the secondary market securely
Data center decommissioning with 48-hour rapid-response mobilization, reclaiming value and space efficiently
R2v3 Appendix E materials recovery for genuine end-of-life hardware, with zero landfill

15+ years of ITAD experience, $25M+ in client capital recovered, 45M+ pounds diverted from landfill.

Request a Free Asset Recovery and Sourcing Assessment

Whether you are retiring hardware and want to capture its value, or sourcing quality used gear to add capacity while new builds wait on permits and power, a specialist can help. Tell us what you are retiring or looking for, and we will discuss recovery value, available inventory, and secure, certified disposition. No commitment, no spam.

Prefer to talk directly? Call 585-406-1249 or email info@roctelecom.com.

"*" indicates required fields

Related reading:

Delays, Not Cancellations: Why Buildout Bottlenecks Make Used Data Center Gear a Strategy

What Dell’Oro Actually Reported