Diesel gensets commanded 81.24% of the global data-center generator market in 2025 per Mordor Intelligence [S1], with another commercial market report recording roughly 52% [S5], confirming diesel's continued dominance despite dual-fuel, natural-gas, and HVO-ready alternatives.
That dominance is concentrated in the segments that matter most to engineers: the 2 MW+ and hyperscale brackets, where U.S. data-center diesel nameplate nearly tripled from 20 GW in 2018 to 55 GW in 2024 [S3][S6], and where large frames (1,250 to 3,250 kW standby) now carry 52-plus-week lead times in 2026 versus the 16 to 24 weeks that was normal before 2020 [S4].
Market share: where diesel actually sits in 2025
Diesel's reported share of the data-center generator market varies 52% to 81.24% across three 2026-published datasets, a spread driven by scope (new unit shipments versus installed base versus revenue) rather than disagreement on direction [S1][S5][S9]. Mordor splits 2025 share by capacity, with sub-1 MW units taking 45.79% of revenue and the above-2 MW segment forecast to grow at a 5.61% CAGR through 2031 [S1], which is the bracket that hyperscalers are pulling hardest. By facility tier, Tier 3 sites held 52.86% of 2025 spend, while Tier 4 is projected to grow at 5.71% CAGR through 2031 [S1]. By operator type, colocation led at 53.14% in 2025, with hyperscalers expanding fastest at 5.87% CAGR through 2031 [S1].
The geographic split reinforces the AI-cluster story: North America held 40.34% of 2025 revenue, but Asia-Pacific is projected to grow fastest at 6.09% CAGR through 2031 [S1]. MarketsandMarkets sizes the overall data-center generators market at USD 8.57 billion in 2026, rising to USD 9.79 billion by 2031 at a 2.7% CAGR [S2]; Mordor's 2026 baseline is lower at USD 7.88 billion, growing to USD 9.84 billion by 2031 at a 4.55% CAGR [S1], a 1.5 to 1.7 percentage-point gap that reflects the same scope assumptions behind the share spread.
The large-unit bracket: why above-2 MW is where the action is
A 100 MW AI campus typically installs 100+ MW of standby generation, realised as dozens of 2 to 3 MW gensets per site, and hyperscaler procurement has structurally shifted the demand curve toward large frames [S4]. Rack power density has moved from the 5 to 10 kW range to as high as 100 kW for AI clusters, multiplying per-site backup requirements, and the buildout shows up directly in nameplate: U.S. data-center diesel capacity rose from 20 GW in 2018 to 55 GW in 2024, with Virginia alone having permitted more than 10,500 generator units totalling 27 GW by the end of 2025 [S3][S4][S6].
Operator procurement is also pulling forward into OEM order books: Caterpillar's backlog reached roughly USD 63 billion, and large-genset production slots at major OEMs are reported as allocated into 2028 [S4]. That capacity squeeze is the practical reason a buyer sourcing today is looking at 52-plus-week delivery on 1,250 to 3,250 kW standby diesel in 2026, against a 16 to 24 week pre-2020 norm [S4]. For sizing context, the Mordor 2025 dataset shows large data centres at 51.68% of market share with hyperscale growing at 5.46% CAGR through 2031 [S1], which is the same demand profile that is filling OEM backlogs.
Lead times and price escalation by frame size
Lead-time stratification by frame size in 2026 is the cleanest signal of where the constraint sits: 25 to 400 kW standby diesel at 12 to 26 weeks, 500 to 1,000 kW at 26 to 52 weeks, and 1,250 to 3,250 kW at 52-plus weeks, all up sharply from the 4 to 24 week pre-2020 baseline [S4]. The bottleneck cascades beyond the genset itself: medium-voltage switchgear is running up to 72 weeks versus a roughly 24 week pre-2020 norm, and large automatic transfer switches are at 20 to 40 weeks versus 6 to 12 weeks [S4].
Pricing has followed the queue. New-unit prices have escalated well above general inflation since 2021, with annual increases and raw-material surcharges now routine on quotes [S4]. The used market has repriced accordingly: low-hour large gensets that historically sold at 40 to 50% of new now command 60 to 75% of new when new-unit delivery is a year out [S4], which is essentially a market price for time-to-power. For buyers with downtime risk, a load-bank-tested used unit delivered this month is frequently the lowest total-cost option, even before accounting for paralleling flexibility [S4].
Fuel-flexibility and the non-diesel alternatives
Natural gas, bi-fuel, hydrogen, and HVO-ready units are the four alternatives that procurement teams are formally evaluating, and they line up differently against diesel on three decision criteria: permitting difficulty, pickup performance, and Scope 1 emissions. NG unit queues are often shorter than diesel in 2026 and permitting is easier in strict air districts, but the trade-offs are pipeline dependence during regional disasters and derated pickup performance versus diesel [S4].
Mordor's 2026 dataset frames the longer arc: hydrogen- and HVO-ready units are projected at a 5.22% CAGR through 2031, the above-2 MW segment at 5.61% CAGR through 2031, and the Tier 4 segment at 5.71% CAGR through 2031 [S1], so the clean-fuel story is layered on top of the large-frame, high-tier story rather than replacing it. For operators weighing flow-meter and pressure-transmitter upgrades alongside a fuel-flex retrofit, the practical decision is whether dual-fuel capability earns its capex against the 52-week-plus large-frame diesel queue and the diesel-only permitting friction in California, Virginia, and the Northeast.
Siting and environmental constraints on diesel growth
The Better Data Center Project's March 2026 report quantifies what diesel dominance looks like at the community level: more than 10,500 generator units permitted in Virginia by the end of 2025, totalling 27 GW, which is the power consumption equivalent of more than 20 million U.S. homes, in a state with fewer than 4 million homes [S3]. Oregon's permitted and proposed data-center diesel fleet reached 6 GW, double the average power draw of all the state's homes [S3]. These are not abstract numbers; they drive local air-permitting rules, noise ordinances, and fuel-storage spill containment requirements that engineers have to design around even before the genset order is placed.
MarkNtelAdvisors' 2026 US dataset shows the Northeast region holding roughly 32% of the US data-center power generator market share, driven by hyperscale clustering in northern Virginia [S7]. Together these figures frame why the same buyer is also evaluating battery-first architectures, pressure-sensor enabled microgrid controllers, and gas gensets as parallel options rather than wholesale diesel replacements.
Procurement playbook for 2026 large-frame orders
Five operational moves are documented in the 2026 procurement literature and are worth specifying into a bid package: order at design stage rather than construction stage, negotiate caps on price escalation clauses, request firm slot commitments, consider paralleling smaller frames to substitute for a single large unit, and watch cancelled-order inventory for nearly-new equipment [S4]. On a 100 MW AI campus the paralleling logic is straightforward: two 500 kW units can replace one 1,000 kW unit and shorten the queue, at the cost of additional switchgear and an N+1 redundancy gain a single large unit cannot offer [S4].
Buyers also need to derate for altitude, ambient temperature, and harmonic loads at the spec stage rather than at commissioning, because a 2 MW nameplate does not deliver 2 MW at 1,000 m elevation in 40°C ambient. The secondwatt.com 2026 procurement guide flags these derate calculations alongside Tier 4 permitting sequencing as the two most common causes of schedule slip on 2026 large-frame orders [S8]. For engineers integrating generator controls with building management or PLC sequencing, the 52-plus-week large-frame queue means control-panel and industrial-valve specification should be locked at least two quarters ahead of the genset delivery slot, not concurrent with it.
Trackable signals through 2027
Three data points are worth watching through the next four quarters: the U.S. Northeast's share of the data-center power generator market (32% in the MarkNtelAdvisors 2026 dataset, the bellwether for hyperscale build-out continuation) [S7], Caterpillar's backlog number in its next 10-K (USD 63 billion as of mid-2026 [S4]) which is the cleanest read on hyperscaler forward demand, and the ratio of new large-frame lead times to switchgear lead times, currently 52-plus weeks versus 72 weeks [S4], since switchgear going shorter than gensets would signal the constraint rotating back to the engine itself. The DC microgrids in industrial plants 2026 spec snapshot is a useful reference for the paralleling architecture, and the secondary copper 2026 refined mix update tracks one of the input cost lines feeding the raw-material surcharges now appearing on genset quotes.