What Hyperscaler Capex Means for Mid-Market MEP
Big Tech's $725B 2026 data center capex is concentrating among large MEP firms. Here is what the mid-market needs to compete for that work.
TL;DR. The four big hyperscalers (Amazon, Google, Meta, Microsoft) will spend roughly $725 billion on capex in 2026, with most of it flowing into data center buildout. According to ABC, contractors with active data center work are running 12+ month backlogs while firms below $30M in revenue are reporting their lowest backlog levels ever. The work is concentrating fast. For mid-market MEP specialty contractors in the $20-100M revenue band, the strategic question for 2026-2027 is not whether to chase data center work, but which version of the opportunity to chase: hyperscale-direct, frontier-market spillover, or behind-the-meter and substation adjacencies.
If your mid-market mechanical or electrical contracting business is watching backlog tighten in 2026, it is worth understanding why. The same year hyperscalers committed to roughly $725 billion in capex (per Q1 2026 earnings calls), the Associated Builders and Contractors reported the widest backlog gap on record between large data-center-exposed contractors and everyone else. Firms above $100 million in annual revenue are at their highest backlog levels since 2021. Firms below $30 million are at their lowest. The capex tide is not lifting all boats. It is lifting a few very specific ones, and a $40-80M MEP shop has decisions to make about which side of that gap to be on.
The numbers: $725 billion is the floor, not the ceiling
The four major hyperscalers each raised their 2026 capex guidance in Q1 earnings. Microsoft committed to $190 billion (up from analyst estimates of $147 billion, with CFO Amy Hood telling investors $25 billion of the increase is component cost inflation). Amazon held at $200 billion. Google guided to $180-190 billion. Meta raised its range to $125-145 billion. Combined: roughly $725 billion in announced capex, almost all of it data center infrastructure.
That is the headline number. The fuller picture is bigger. Moody's Ratings now projects hyperscaler capex at $785 billion in 2026 when Oracle and CoreWeave are added, and approaching $1 trillion by 2027. The hyperCAPEX cohort took 25 years to spend their first trillion dollars cumulatively. They will likely spend the next trillion in 18 months.
For context: the combined 2025 capex of the four big hyperscalers was $416 billion, up 66% from $251 billion in 2024. The 2026 numbers represent another 74% increase year-over-year. This is not a normal capex cycle. It is the largest sustained infrastructure buildout in modern American industrial history, and it is mostly going into mechanical and electrical systems: power distribution, transformers, cooling, fire suppression, structured cabling, controls.
That work has to be built by someone. The interesting question for mid-market operators is: by whom.
Where the money is actually flowing
Look at the public earnings of the three largest specialty MEP contractors with explicit data center exposure. The numbers tell a consistent story.
EMCOR Group (NYSE: EME) reported $4.63 billion in Q1 2026 revenue, up 19.7% year-over-year. Inside that headline number, the Electrical Construction segment grew 33% with the Network & Communications sub-sector (where data centers sit) up nearly 50%. The Mechanical Construction segment grew 29%, with Network & Communications up 86% as data center cooling and AI liquid-cooling requirements drove the work. EMCOR's RPOs hit a record $13.25 billion at the end of 2025, up $3.15 billion year-over-year.
Comfort Systems USA (NYSE: FIX) is the more dramatic story. Q1 2026 revenue jumped 56.5% to $2.87 billion. Adjusted EBITDA rose 150%. Backlog hit $12.45 billion, up 80.7% year-over-year, having nearly doubled in twelve months. Free cash flow swung from negative $109 million in Q1 2025 to positive $242 million in Q1 2026. Management explicitly attributes the surge to technology sector demand, specifically data centers. The company acquired J&S Mechanical in late 2025 as a strategic bet on data center HVAC. According to MetricDuck's DuPont analysis, FIX is running 40.8% ROIC versus Quanta Services' 9.7% despite Quanta having four times the revenue. The data center concentration is paying off.
IES Holdings reported Q2 FY26 Communications segment revenue of $367.7 million, up 35% on data center demand. Infrastructure Solutions grew 64% to $192.4 million on data center custom engineered solutions.
These are the firms eating the meal. None of them are mid-market. EMCOR is a Fortune 500 company. FIX has $12.45 billion in backlog. IES has 11,000+ employees. The capex flowing through them is real, the growth is real, and the operating margins (FIX at 17%, EMCOR's electrical at 12.1%) are exceptional by specialty trade standards. But they are not the audience for this post.
Transformer lead times have ballooned from 24-30 weeks pre-2020 to 80-120 weeks today. Equipment is the first binding constraint on data center buildout pace.
The widening gap that is keeping mid-market MEP awake at night
Construction Owners reported in May 2026 that the ABC Construction Backlog Indicator hit 8.8 months in April, the highest reading in nearly a year. Inside that average, the distribution is bifurcated. Contractors with active data center projects carried average backlogs above 12 months. Firms without data center exposure averaged 7.8 months. The 13% of contractors with data center work were running close to 11 months, per ABC's December 2025 data.
The size-based split is starker. Per ABC chief economist Anirban Basu, firms above $100 million in annual revenue are at their highest backlog levels since 2021. Firms below $30 million are at their lowest. The gap is not closing. It is widening, structurally, as hyperscalers concentrate their work among general contractors and specialty MEP firms with the balance sheet, the credentials, the labor depth, and the prefabrication capacity to bid hyperscale.
If you are running a $40-80 million electrical or mechanical contractor, you are sitting in the middle of that split. You are too small to be EMCOR's bid sheet, too large to be irrelevant. The decision in front of you for 2026-2027 is which version of the data center opportunity to pursue, because doing nothing means watching backlog erode while your larger competitors take share.
Three real paths for mid-market MEP
The mid-market MEP contractor has three plausible paths into this market. They are not mutually exclusive, but they require different capability bets and capital allocation choices.
Path 1: Become a credible hyperscale subcontractor. EMCOR and FIX do not self-perform every trade on every project. They sub out specialty work to firms that can deliver on hyperscale schedules with hyperscale safety and quality standards. The credentials required to be on these bid lists are real: bonding capacity well above your current ceiling, dedicated mission-critical project management resources, prefabrication facilities, BIM/VDC depth, and labor counts you may not have today. The reward is being on the bid sheet for the largest specialty contracts in the country.
Path 2: Frontier-market spillover. As Bricks & Bytes reported in April 2026, hyperscalers are pushing into frontier markets where capacity is available: West Texas, Tennessee, Wisconsin, Ohio. Most of the work in these markets sits outside the data center footprint itself: substations, civil works, water and wastewater infrastructure, behind-the-meter power, hotel and housing buildout for the labor force. This is the work that funds the AGC's 2026 growth percentages after data centers themselves. For a regional mid-market MEP, this is the most accessible path because it leverages existing capabilities rather than requiring entirely new ones.
Path 3: Adjacent-systems specialization. Electrical work accounts for 45-70% of data center build cost (per IBEW data). Transformer lead times have ballooned from 24-30 weeks pre-2020 to 80-120 weeks today. Critical-class equipment runs multi-year lead times because the US makes just 20% of what it needs while China holds 60% of global transformer capacity. There are real opportunities for mid-market firms with deep electrical engineering, switchgear, controls, or commissioning specialization to position as specialists rather than generalists. The bet is depth, not breadth: 14,700 IBEW members in the Northern Virginia local, doubled from 7,000 in seven years, signals where the wages and the work are heading.
The constraints that will actually decide who wins
The data center boom has two binding constraints, and both favor specific kinds of mid-market firms over others.
The first is electrical equipment. The transformer bottleneck is not going away in 2026. Firms that have invested in supplier relationships, advance procurement strategies, or alternative sourcing arrangements have a competitive advantage that is hard to replicate quickly. If you have not had the supplier conversation about 2027-2028 transformer allocations, your competitors have.
The second is craft labor. ABC's December 2025 figures put the construction industry's labor gap at 439,000 workers, led by electricians and pipe layers. Brookings' Darrell West told Fortune the electrician shortage is "quite dire." Wages reflect this: data center pros are pulling $120-180K, and the IBEW Northern Virginia local doubled to 14,700 members in seven years. For mid-market MEP, the workforce question 18 months out matters as much as which contracts you sign this quarter. Headcount inflation without a backlog to absorb it will be the thing that kills firms when the cycle turns.
These two constraints are why AI-enabled estimating, scheduling, and project management have become a real conversation among mid-market MEP leadership in 2026. When the bottleneck is people and equipment, the firms that get more output from the people and equipment they already have will take share.
Prefab capacity, BIM/VDC depth, and the ability to staff a mission-critical project to hyperscale schedule and quality standards are the credentials that decide which mid-market MEP firms get on EMCOR's and Comfort Systems' sub lists.
What this means for your 2026-2027 planning
Three concrete moves worth considering if you are running a mid-market MEP business right now.
First, audit your current backlog composition by end market. If your data center exposure is zero, that is a vulnerability worth quantifying. If it is more than 40% of forward bookings, that is a different kind of risk and worth understanding the concentration exposure. The current narrative is that the buildout runs through 2027 at least. The harder boardroom conversation, as Bricks & Bytes put it, is what your revenue mix should look like in the quarter after the cycle turns.
Second, run the capability gap analysis honestly. What would it take to be on EMCOR's, FIX's, or Holder's mission-critical sub list in 2027? The honest answer for most $30-80M MEP firms involves bonding capacity, dedicated MC PM resources, prefabrication investment, and a labor strategy that does not yet exist. Decide whether you want to make that bet, partner your way into it, or focus on spillover work instead.
Third, the AI question. Bid-to-award cycles will be shorter and estimating throughput will be higher because the labor pool will not stretch any other way. The firms that figure out this operating model first will take share. The firms that wait will be the ones with headcount they cannot absorb when the cycle turns.
FAQ
Is the data center buildout going to continue through 2027? According to Moody's and the hyperscalers' own commentary, yes, with capex approaching $1 trillion by 2027. Hyperscalers have added approximately $700 billion in remaining performance obligations over the last two quarters, signaling the contracted revenue backlog supporting continued capex. The constraints are power availability, permitting, transformers, and labor, not demand.
Can a $40M electrical contractor realistically bid hyperscale work? Directly, almost never. As a credentialed sub to EMCOR, FIX, Rosendin, or Holder, yes, but the credential requirements (bonding, mission-critical PM resources, safety record, prefab capacity) are substantial. Most $40M firms are better positioned for spillover work or adjacent-systems specialization.
Where are the spillover opportunities? Substations, behind-the-meter power, civil works, water and wastewater infrastructure, controls and commissioning, and the housing and hospitality buildout in the frontier markets where hyperscalers are siting new campuses. West Texas, Tennessee, Wisconsin, and Ohio are the geographic concentration points.
How does AI fit into this for an MEP contractor? The binding constraint is people and equipment, not capital. Firms that can get more bid throughput out of senior estimators, more job oversight per project manager, and faster turnaround on submittals and RFIs will take share. This is operational AI applied to estimating, scheduling, knowledge capture, and project management workflows, not science-fiction productivity claims.
What is the biggest risk? Headcount inflation without a backlog to absorb it. The $120-180K labor cost structure of data center pros is hard to unwind quickly if the cycle slows.
If your mid-market MEP business is sitting in the gap between hyperscale-direct work and traditional commercial backlog and trying to figure out the right operational bet for 2026-2027, the Granular team works with specialty contractors on exactly this problem. Fixed-price, four-week implementations focused on the estimating, scheduling, and knowledge-capture workflows that actually move bid throughput and project capacity. Book 30 minutes and we can walk through what we have learned from operational conversations across the sector.
Keep Reading
- How to Capture Tribal Knowledge Before Key People Leave. The senior estimator question is not theoretical. Here is the playbook for documenting what your veterans know before the labor crunch makes their departure expensive.
- How Long AI Actually Takes to Deploy at a $50M Company. If the hyperscale work is two years out and the labor constraint is now, the AI deployment timeline matters. Real numbers from real implementations.