The Secondary Market Is the Shock Absorber of the AI Data-Center Boom

Global data center capital spending is on track to exceed 1 trillion dollars in 2026, according to Dell'Oro Group, which raised its forecast after the first quarter came in stronger than expected. The four largest U.S. cloud providers, Amazon, Google, Meta, and Microsoft, increased their own data center capex by 78 percent year over year in that same quarter. That is not a market growing at a normal pace. It is a market absorbing a demand shock, and the physical supply chain underneath it was never sized for one.

The result is not a simple shortage. It is volatility, and volatility is a different problem that needs a different kind of fix.

The primary supply chain cannot flex on this timescale

Transformers and switchgear are built in factories that take years to expand and certify, and demand arrived faster than any factory could respond to it. Wood Mackenzie analyst Ben Boucher told Reuters that generator step-up transformer lead times surpassed 160 weeks by the first quarter of 2026, up from an average of 143 weeks in 2024. High-voltage circuit breakers climbed to 125 weeks in the second half of last year, compared with 77 weeks in 2023. Wood Mackenzie's own projection explains why relief is not close: U.S. data center capacity is set to hit 110 gigawatts in 2030, up from around 24 gigawatts today, and data centers' share of the electrical equipment market could swell to 40 percent under accelerated buildout scenarios, from just under 2 percent in 2020. Boucher put the developer's dilemma plainly: equipment availability is becoming the biggest concern, because time to market is what everyone is actually competing on now.

The grid adds its own multi-year queue on top of the equipment queue. An October 2026 analysis from Carbon Direct found that interconnection requests in PJM wait an average of 40 months, ERCOT averages 20 months, and the national average is 25 months, against a FERC target of 8 to 11 months for the same process. Active data-center load-growth projects in PJM and ERCOT specifically are waiting 36 to 48 months, and more than 2.2 terawatts of generation and storage are sitting in interconnection queues nationally. A transformer that ships on schedule still cannot be energized until the grid connection clears, and the grid connection is often the longer wait of the two.

Scarcity and cancellation are happening at the same time

Here is the part that makes this a volatility problem rather than a straightforward capacity problem. Bernstein's research estimates that 35 to 40 percent of announced global data center capacity is at risk of delay or cancellation, with the pace of cancellations expected to accelerate into 2027. Bernstein's analysts describe power availability, not capital, as now the primary gating factor on new sites, and they note that interconnection queues of 3 to 4 years in markets like Northern Virginia, Frankfurt, and London are already pushing operators toward secondary markets and retrofits. Construction costs per megawatt have risen 20 to 25 percent since 2023 on the same report, and transformer and switchgear lead times of 80 to 100 weeks are baked into that cost increase.

So the same twelve months that produce a 78 percent jump in hyperscaler capex also produce a wave of projects that slip their schedule or get cancelled outright. Developers respond to multi-year lead times by ordering defensively, reserving transformers and switchgear at site control, before permits, sometimes before final design, because waiting for certainty means losing the queue position. When a project then gets re-scoped, delayed, or cancelled, as Bernstein's own numbers say a third to nearly half of announced capacity will be, that equipment does not evaporate. It sits, reserved and sometimes already delivered, with no project left to consume it. The same boom that starves some sites of equipment is quietly stranding equipment meant for others.

A second supply is already flowing the other way

There is a parallel source of usable equipment that has nothing to do with cancelled orders: equipment coming off line because it is being replaced, not because it failed. The global market for data center decommissioning services was worth 12.95 billion dollars in 2026 and is projected to reach 19.94 billion dollars by 2032, a compound annual growth rate of 7.37 percent, according to decommissioning contractor Powerhouse Systems. Facilities being upgraded or consolidated take generators, UPS systems, and switchgear lineups offline well before those assets are worn out, for the same reason hyperscalers are refreshing GPU and CPU fleets faster than traditional IT replacement cycles ever ran.

Put the two flows together and the shape of the problem becomes clear. On one side, projects that cannot get new equipment fast enough to hit an energization date measured against an interconnection queue that itself runs 20 to 48 months. On the other, equipment stranded by cancelled orders and equipment released by decommissioning and refresh cycles, both arriving at the same time, for unrelated reasons. Neither flow is small, and neither is going away while the primary supply chain remains capacity-constrained for years rather than quarters.

Why "shock absorber" is the right description, not just a metaphor

A shock absorber does not remove the bumps in the road. It converts a sharp jolt into something the vehicle can actually ride through. That is precisely the function a structured secondary market performs for a capex cycle moving this fast: it takes equipment released in one place, by a cancelled project or a decommissioning schedule, and routes it to a site that needs capacity now and cannot wait out a 160-week transformer order or a 40-month interconnection queue. Without that routing mechanism, both halves of the problem get worse at once. Stranded equipment sits as a write-down. Starved projects slip further and lose pre-sold AI capacity to a competitor who found capacity somewhere else.

The industry mostly still treats the secondary market as a backup plan, something to check after the OEM order is placed and the lead time comes back worse than hoped. That is backwards for a cycle this volatile. A market moving 78 percent year over year on the demand side, against a supply side measured in years rather than months, needs a reallocation mechanism built into the plan from the start, not bolted on after a schedule slips. The operators who treat the secondary market as a primary sourcing channel, not a last resort, are the ones positioned to absorb the next shock instead of being flattened by it.

Whether you are sourcing equipment against a deadline the primary supply chain cannot meet, or sitting on reserved or decommissioned gear with no project left to use it, that reallocation happens at supplyvaultai.com/marketplace.