New vs. Used Data Center Equipment: The Real Cost Math in the AI Buildout

Every capital plan for a new data center starts with a number for equipment cost and a number for schedule. In 2026, both numbers are wrong more often than they are right, because new equipment is quoting later and pricier than it did even a year ago, and the gap between "new" and "used" on paper is no longer the gap it used to be in practice. Before you commit a budget line to new or used, it is worth doing the math the way a procurement team should: on total time to energized capacity, not on the sticker.

Why new is backed up this far

The AI buildout has pulled every category of heavy electrical equipment into the same queue. On Eaton's first quarter 2026 earnings call, Chief Operating Officer Paulo Sternadt told investors that the company's total data center backlog had grown to 228 gigawatts, equivalent to 12 years of backlog at 2025 build rates, up from 11 years at the prior update, with data center orders up 240 percent year over year. That is one supplier's order book, and it is instructive: demand did not create a one-time bump, it created a standing queue that keeps growing faster than it can be worked down.

Transformers show the same pattern with harder numbers. Wood Mackenzie's second quarter 2025 survey found standard large power transformers averaging 128 weeks for delivery and generator step-up transformers averaging 144 weeks, with some orders extending to four years. The same data put power transformer prices up 77 percent since 2019. A transformer ordered today, at today's price, may still not be the transformer standing on your pad in 2027.

Switchgear and distribution equipment are not far behind. According to a surplus and reconditioned equipment buyer's guide published by PowerGen Enterprises, new distribution transformers are currently quoting 50 to 100-plus weeks and new switchgear 40 to 80 weeks, with circuit breakers frequently on manufacturer allocation rather than open order. The same guide reports that tested surplus equipment, when it carries documentation, sells at 30 to 60 percent of new pricing and can ship within days rather than months.

Generators tell a more nuanced version of the story. A generator shortage analysis published by CS Diesel Generators lays out current lead times by size class: 12 to 26 weeks for 25 to 400 kW units, 26 to 52 weeks for 500 to 1,000 kW units, and 52-plus weeks for 1,250 to 3,250 kW units, each roughly double to triple what the same size classes quoted before 2020. The same analysis notes that low-hour used gensets, which once traded at 40 to 50 percent of new-unit price, now command 60 to 75 percent of new when new delivery is a year or more out, while tested used units already in inventory can ship in one to four weeks. Scarcity has pushed up the price of used equipment too. It just has not erased the schedule advantage.

Servers are its own case, driven by memory

IT hardware is not backlogged the way heavy electrical equipment is, but new server pricing is moving fast for a different reason: memory. Counterpoint Research, cited by Network World, projects that DRAM prices will have already risen roughly 50 percent year to date in 2025, with another 30 percent increase expected in the fourth quarter of 2025 and a further 20 percent in early 2026, driven by manufacturers redirecting fabrication capacity toward AI accelerators. The same report states that DDR5 64GB RDIMM modules commonly used in enterprise servers could cost twice as much by the end of 2026 as they did in early 2025, and that Samsung has already raised list pricing on 32GB DDR5 modules from 149 dollars to 239 dollars, a 60 percent increase.

That memory cycle is exactly why refurbished enterprise servers are pulling ahead on value this year rather than just on price. A refurbished-hardware buying guide from PC Server and Parts puts typical savings for a refurbished server at 40 to 70 percent less than a new equivalent configuration, and the mechanism is straightforward: a refurbished unit usually carries memory populated before the current price run-up, so the buyer is not paying 2026 memory prices for capacity bought months ago. For workloads that do not require the newest generation of chips, a refurbished fleet bought now can be meaningfully cheaper than new capacity ordered for delivery next quarter.

The real cost math, not the sticker price

None of this means used is automatically the right call. It means the comparison has to include variables that a simple new-versus-used price table leaves out:

Time to energized capacity. A new transformer at a fair price that arrives in 2028 is not cheaper than a used one at a premium that arrives in six weeks, if the project's revenue clock is already running. Model the cost of the delay, not just the cost of the unit.

Documentation as a condition of price. The surplus and used market only delivers on its discount if the equipment comes with test data: insulation resistance and turns-ratio results for transformers, primary or secondary injection results for breakers, load-bank results and service logs for generators. Undocumented used equipment is a different, riskier asset than documented used equipment, and it should be priced and underwritten differently.

Compatibility with the rest of the plant. Voltage class, protection scheme, and control system compatibility matter more on a retrofit than on a greenfield build. A used unit that needs a controller swap or a protection study to integrate can eat its own price advantage.

Warranty gap. New equipment typically carries full OEM warranty. Reputable refurbishers and resellers offer limited warranties, and some used equipment is sold as-is. Price the risk difference, not just the dollar difference.

Where the premium is actually worth paying. For a single critical path item on a hard energization date, paying up for a documented, tested used unit that ships this month can be the entire project's schedule insurance. For a non-critical expansion with a flexible date, waiting for new capacity at a better price can make more sense.

Structure the decision like a procurement, not a shopping trip

The buyers getting this right are not choosing new or used as a blanket policy. They are specifying the requirement, capacity, voltage, condition tier, documentation standard, and required-by date, and then letting the market answer with whatever combination of new, surplus, and refurbished equipment actually meets it. That requires visibility into both channels at once, which is harder to get from a single OEM account rep or a single broker's inventory than it sounds.

The bottom line

New equipment lead times and prices are both moving against the buyer in 2026, across transformers, switchgear, generators, and increasingly the memory inside every server. Used and refurbished equipment is not automatically cheaper anymore in every category, but it is almost always faster, and speed has a dollar value on a project with a revenue date attached. Do the math on total time to energized capacity, insist on documentation before price, and treat new and used as two supply channels competing for the same requirement rather than two different philosophies.

---

Sourcing data center equipment on a deadline? SupplyVault AI is a structured marketplace for the data-center secondary market. Post a procurement request with capacity, voltage, condition tier, and timeline, and get matched to documented inventory from verified sellers, alongside new-equipment options, with deal rooms, contracts, and payment tracking on one platform. Full access is free for 30 days → supplyvaultai.com/buyers