Construction · Deep dive

Comfort Systems USA

The 1997 Houston HVAC roll-up that survived its own consolidation hangover, quietly compounded for two decades, and then became the purest picks-and-shovels trade of the AI buildout — $3.3B of quarterly revenue, a $14.1B backlog, 58% of sales from hyperscaler data centers, and a stock up 60x in ten years.

well positioned

A $14.1B contracted backlog, a craft workforce of tens of thousands that rivals cannot conjure, and modular capacity sold out into 2027 make the operating position close to unassailable for years — the 42x multiple is the fragile thing, not the company.

Snapshot

Comfort Systems USA is a Houston-based mechanical and electrical contractor — pipes, ducts, chillers, switchgear, controls — operating through roughly 45 autonomous operating companies across 197 US locations. For twenty years it was a solid, boring compounder. Then AI data centers turned its two specialties, industrial-scale cooling and power distribution, into the scarcest inputs in American construction. Q2 2026 revenue was $3.27B, up 50.3% year over year; net income was $441.6M with EPS up 92%; backlog hit a record $14.1B, up 73% in a year; and technology customers — overwhelmingly hyperscale data centers — supplied 58% of revenue, up from 40% a year earlier. The stock has returned roughly 6,000% over the decade to June 2016-2026 (Zacks, June 2026) and the company now carries a ~$60B market cap. The question the market is arguing about, loudly, is whether a construction company deserves 42x earnings.

Founding story

Comfort Systems is a child of the late-1990s roll-up mania. Organized in Houston in 1996, it went public on July 2, 1997, simultaneously acquiring twelve founding HVAC contractors, and then bought 107 more companies by the end of 2000 — the classic consolidate-and-arbitrage playbook applied to a fragmented ~$15B commercial HVAC market. The playbook nearly killed it. Integration strain and a construction downturn forced a humbling 2002 retreat in which the company sold 19 of its operating units to rival EMCOR to pay down debt — the survivor selling limbs to the healthier consolidator. What survived was the useful part: a decentralized federation of local contractors who kept their names, leaders and customer relationships, with Houston providing capital, bonding and discipline.

Brian Lane arrived in October 2003 from a 15-year career at Halliburton/Brown & Root (ending as Regional Director for Europe and Africa, with Capstone Turbine and Kvaerner stops after), rose to COO in 2009 and CEO in December 2011. His two defining moves both looked expensive at the time: buying into electrical contracting with Walker Industries in 2017, and buying off-site modular fabricators Environmental Air Systems (2020) and TAS Energy (2021) just before hyperscalers decided modular MEP was how data centers get built. Succession is now underway — 20-year insider Trent McKenna became President in January 2026 and Craig Sasser COO in July 2026 — with Lane still CEO.

How it works

Mechanically, the company sells installed building systems. On a hyperscale data-center campus, the general contractor or the hyperscaler itself engages Comfort Systems subsidiaries — increasingly on a negotiated or design-assist basis rather than open hard bid, because capacity, not price, is the constraint — to engineer, fabricate and install the mechanical scope (chilled-water plants, cooling distribution, piping, air handling) and, since the electrical expansion, switchgear, power distribution and controls. Contracts are long-duration; revenue is booked on percentage-of-completion, and milestone billings typically run ahead of cost, which is why 2025 operating cash flow ($1.19B) exceeded even record net income — customers are effectively financing the work.

The differentiating machine is modular. At six-plus plants totaling over 3.5 million square feet in Texas and North Carolina, EAS and TAS build complete mechanical rooms, chilled-water plants and power skids on production lines, truck them to site and set them in place — cutting data-center construction time by up to 40% (company figures) and, critically, moving scarce skilled labor from muddy sites into controlled factories. Management said in July 2026 that modular capacity is headed to 4 million square feet by year-end and roughly 5 million by late summer 2027, and that demand still exceeds it. The final input is people: roughly 19,000+ employees, most of them pipefitters, sheet-metal workers and electricians, in an industry the Associated Builders and Contractors estimates was short 439,000 workers in 2025 and needs 349,000 more in 2026. Owning that workforce is the moat.

Product and business overview

Mechanical segment (the core, ~three-quarters of revenue). HVAC, plumbing, piping, process cooling and controls installation for data centers, chip fabs, pharma, hospitals, education and manufacturing. Q2 2026 gross margin: 25.6%.

Electrical segment. Power distribution and electrical construction, built by acquisition (Walker 2017, Eldeco, Feyen Zylstra and Meisner 2025). Q2 2026 gross margin: 26.4% — electrical in a data center is now as valuable as cooling.

Modular/off-site construction. EAS and TAS Energy’s factory-built plant rooms and modules, the fastest-growing and most capacity-constrained product line.

Service. Maintenance, on-demand repair and building automation — only about 10% of revenue (service projects 4.4% plus service/maintenance 5.7% in Q2 2026), growing ~8%, but the counter-cyclical ballast and the long-term opportunity as the installed data-center base ages.

Business model and pricing

Revenue is booked as contract revenue over time on percentage-of-completion; there is no published price list because every job is engineered. The economics show up in mix and margin: new construction was 75.1% of Q2 2026 revenue, existing-building construction 14.8%, and the two service lines the remainder. Consolidated gross margin reached 25.9% in Q2 2026 versus 23.5% a year earlier and roughly 18-19% in the pre-AI era — a structural re-rating driven by negotiated work, modular manufacturing margins and pricing power born of scarcity. Same-store revenue grew 44% in Q2 2026; management guided full-year 2026 same-store growth to the mid-to-high 30s, an explicit deceleration in the back half. Acquisitions are priced with private-market discipline — Century Contractors cost $84.2M (January 2025) against tens of millions of revenue contribution — and paid for in cash; the balance sheet carries minimal net debt, and the company completed a long-running buyback and raised its dividend to $0.90/quarter in July 2026.

Traction over time

All figures from company releases and filings on the dates shown. The inflection is unmistakable: backlog doubled during 2025 and added another $2.2B in the first half of 2026; technology went from 40% of revenue (Q2 2025) to 56% (Q1 2026) to 58% (Q2 2026). Ten-year revenue CAGR is 21.3% (Zacks, June 2026), most of it organic in recent years.

Market analysis

The US data-center construction market was worth $48.2B in 2024 and is projected to reach $112.3B by 2030, a 15.2% CAGR (Mordor Intelligence). McKinsey estimated (2024) that capital spending on procurement and installation of mechanical and electrical systems for data centers alone is likely to exceed $250B cumulatively by 2030 — this is precisely Comfort Systems’ scope. Structural forces run in both directions. For: AI training and inference capacity, chip-fab onshoring under the CHIPS Act, pharma reshoring, and the physical reality that every megawatt of IT load needs roughly as much cooling and electrical infrastructure as compute. Against: the demand is concentrated in perhaps a half-dozen hyperscaler capital budgets, and the industry’s binding constraint — skilled labor, short by roughly half a million workers heading into 2027 (ABC) — caps how fast anyone can grow while also propping up incumbents who already employ the workforce. Uptime Institute and CBRE surveys in 2024-2025 both flagged labor, ahead of power, as a top constraint on hyperscale delivery.

Competitive intel

The named set is in the sidebar. The shape: EMCOR is the only rival with comparable scale and the same mechanical-plus-electrical spread, and it is buying its way deeper into the theme (Miller Electric, $865M, February 2025) — but it grows at a fraction of Comfort Systems’ rate and lacks an equivalent modular franchise. IES Holdings is the sharpest pure-electrical attacker inside data centers; APi Group wins on recurring, code-mandated service revenue but doesn’t contest the MEP scope; Limbach shows what a service-first MEP model looks like at 1/14th the size; Quanta fights for the same electricians without bidding the same work. The honest description of this market in 2026 is that demand exceeds the combined capacity of every name above — competition is for labor and fabrication capacity, not for jobs. Comfort Systems’ edge is the thing that cannot be quickly replicated: tens of thousands of employed craft workers, 3.5M+ square feet of modular capacity already sold out, and hyperscaler relationships that now negotiate rather than bid.

History and evolution

• 1996-1997 — Houston roll-up organized; IPO July 2, 1997 with 12 founding HVAC companies.
• 1997-2000 — 107 further acquisitions; classic roll-up sprawl.
• 2002 — Reckoning: 19 operating units sold to EMCOR to cut debt; the roll-up nearly unwinds.
• 2003-2011 — Rebuild years; Lane joins (2003), becomes COO (2009), CEO (December 2011).
• 2017 — Walker Industries: the electrical diversification begins.
• 2020-2021 — Environmental Air Systems, then TAS Energy (~$170-190M revenue): the modular bet.
• January 2024 — Summit Industrial acquired (~$360-400M revenue): industrial/EPC scale.
• 2024 — FY revenue $7.03B; backlog $5.99B; joins the S&P 500 as a top-decile 2024 performer.
• 2025 — Century Contractors ($84.2M), Right Way, Feyen Zylstra, Meisner acquired; FY net income crosses $1B; backlog doubles to $11.94B.
• December 2025 - July 2026 — Succession sequence: McKenna President (Jan 1), Sasser COO (Jul 1); Lane remains CEO.
• July 23-24, 2026 — Record Q2: $3.27B revenue, $14.1B backlog, dividend raised to $0.90 — and the stock falls ~2.9% anyway.

What people say

The case for. Sell-side coverage after Q1 and Q2 2026 was close to euphoric on the operations: Zacks and others called out record margins, an “explosive” backlog and the modular franchise as a durable differentiator, and TIKR framed the $14.1B backlog as multi-year revenue visibility a contractor has rarely enjoyed. Zacks noted (June 2026) that $1,000 invested a decade earlier was worth $61,000. Bulls’ recurring theme: this is no longer a cyclical bid-work contractor but a capacity-constrained industrial manufacturer with pricing power. Employee reviews echo the decentralized culture — Glassdoor’s corporate page runs about 3.6/5 across ~190 reviews with culture, benefits and career development the recurring pros, and some subsidiary pages rate meaningfully higher.

The complaints. The bear case is concentration stacked on cyclicality stacked on multiple. Zacks itself published the counter-piece — “Tech and Data Push: Growth or Concentration Risk?” (July 2026) — noting that a handful of hyperscaler buyers now control 58% of revenue, so a single capex pause, project delay or design shift (e.g., liquid-cooling architectures altering mechanical scope) propagates instantly. Seeking Alpha (2026) called long-term growth strong but valuation and technicals “stretched” at ~42x trailing earnings for a business that is still 75% new-construction revenue — the segment that vanishes first in a downturn. Simply Wall St flagged insider selling into the rally. Glassdoor’s recurring negative theme is management/senior leadership at some units, with only 65% recommending — mediocre for a company this celebrated. And the Q2 2026 tape told its own story: a beat on every line, and the stock slipped ~2.9% because guidance conceded decelerating second-half same-store comps — when perfection is priced in, deceleration is a miss.

Outlook: well positioned or at risk?

Well-positioned — because the constraint in this market is capacity, and Comfort Systems owns the capacity. The moat is not technology; it is 19,000+ employed craft workers in an industry short half a million, 3.5 million square feet of modular fabrication sold out into 2027, and a $14.1B contracted backlog with milestone billings that make customers fund the growth. Even the concentration number is less fragile than it looks: the backlog is contractual, hyperscaler credit is impeccable, and the non-tech pipeline — chip fabs, pharma reshoring, healthcare — is itself under-supplied. If AI capex merely slows, Comfort Systems works through two-plus years of booked revenue while margins normalize from 26% toward historical high-teens; that is an earnings hiccup, not a dislocation. The roll-up machine, self-funded and disciplined ($84M for Century, not $8B for anything), keeps compounding underneath.

The honest risk is the multiple, and it is a real one: at ~42x trailing earnings, the stock — not the business — is priced for the boom to persist, and the July 24, 2026 wobble showed how little deceleration it takes to break the tape. A genuine AI capex pause would compress revenue growth, margins and the multiple simultaneously, and 75% new-construction exposure means the P&L is cyclical no matter how good the execution. But “at-risk” in this framework means ripe for disruption or share loss, and there is no plausible attacker: EMCOR grows slower, IES is a fifth the size, no entrant can hire the workforce, and the customers are begging for more capacity, not less. The company’s position compounds; only its shareholders’ entry price is in question.

Sources and further reading

• Comfort Systems USA Reports Second Quarter 2026 Results — StockTitan/Business Wire, July 23, 2026. Q2 revenue, $441.6M net income, $14.1B backlog, dividend raise.
• Earnings call transcript: Comfort Systems USA tops Q2 2026 estimates, stock slips — Investing.com, July 24, 2026. 58% technology mix, modular capacity to ~5M sq ft by late summer 2027, same-store guidance, stock reaction.
• Comfort Systems USA Reports Fourth Quarter and Full Year 2025 Results — Company IR, February 2026. FY2025 $9.10B revenue, $1.02B net income, $11.94B backlog, $1.19B operating cash flow.
• Comfort Systems USA Reports First Quarter 2026 Results — Company IR, April 23, 2026. Q1 figures and $12.45B backlog.
• Comfort Systems’ Tech and Data Push: Growth or Concentration Risk? — Zacks via Nasdaq, July 2026. The hyperscaler-concentration bear case.
• Comfort Systems USA: Strong Long-Term Growth, But Valuation And Technicals Stretched — Seeking Alpha, 2026. The ~42x valuation debate.
• If You Invested $1000 in Comfort Systems a Decade Ago — Zacks/Yahoo Finance, June 2026. The ~6,000% ten-year return and 21.3% revenue CAGR.
• Comfort Systems USA Announces Acquisition (Summit Industrial) — Business Wire, January 2, 2024. Summit’s $360-400M revenue contribution.
• EMCOR Group Completes Acquisition of Miller Electric Company — EMCOR IR, February 3, 2025. The chief rival’s $865M answer to the same demand wave.
• Data Center Construction Labor Shortage 2026 — Build.inc, 2026. Skilled-trades shortage as the binding constraint; ABC workforce-gap estimates.
• Comfort Systems Announces Leadership Transitions Effective July 1, 2026 — Company IR, June 22, 2026. Sasser to COO; 197 locations in 143 cities.