Tesla Terafab Suppliers: Power & Cooling Names That Book the $25B Buildout First
The publicly traded infrastructure names that book Terafab orders before a single 2nm wafer ships.
The publicly traded infrastructure names that book Terafab orders before a single 2nm wafer ships.
In March 1953, Convair had ten people assigned to a missile project in San Diego. Seven years later it had twelve thousand workers on a two-million-square-foot floor, and behind them stood 30 major subcontractors, 500 lesser contractors, and 5,000 suppliers scattered across 32 states (Atlas: Summary of the Atlas ICBM - The Military Standard), all booking revenue against an Air Force program called Atlas before a single warhead-capable missile ever went on alert. The chips had not flown. The contracts had.
A version of that procurement event is now happening in Austin. Elon Musk took the stage at the defunct Seaholm Power Plant in Austin to officially unveil TERAFAB, a $20-25 billion joint venture between Tesla, SpaceX, and xAI, and within four weeks dirt was moving on the North Campus of Giga Texas. The investment case for Tesla shareholders is contested. The case for the suppliers is not. Power equipment makers and cooling specialists will book the first wave of Terafab orders, and those orders will land before the first AI5 chip is qualified.
What Terafab Actually Is, and Why Suppliers Care
Terafab is a fab, but the way Musk has framed it, it is also the largest single industrial power load any private company has ever proposed. Terafab is designed to produce chips for Tesla vehicles, humanoid robots, and AI data centers, with a target of one terawatt of compute annually, double current U.S. capacity (Tesla to use Intel 14A chips in $3B Terafab push). One terawatt is not a typo (it is also not, to be fair, a confirmed nameplate; Musk says "ultimately"). Terafab is designed for an initial output of 100,000 wafer starts per month, with ambitions to scale to 1 million wafer starts per month at full capacity (electrek.co), which at the high end is roughly 70% of TSMC's total global output from one Texas site.
Two-nanometer fabs are the most power-hungry industrial facilities ever built. Every wafer-start needs deionized water, ultra-stable 480V power, redundant UPS, and chilled water loops sized for racks of EUV scanners. That is the supplier opportunity. The Atlas program created decades of recurring procurement for General Electric, Rocketdyne, and AC Spark Plug long before the missile became a space launcher. Terafab's chosen suppliers are about to embed themselves into Tesla's manufacturing stack the same way.
Power: The First Bottleneck
Switchgear, transformers, UPS, and on-site generation are the long-lead items. A 2nm fab cannot break ground without committing to its medium-voltage gear two years out, which is why Eaton (ETN), GE Vernova (GEV), Schneider Electric, and Bloom Energy (BE) have already had their stocks repriced on AI infrastructure exposure (GEV up roughly 14.6% on the most recent tape; BE up 11.2%). GEV brings the gas turbines and grid equipment a campus this size needs to island itself off ERCOT during a Texas heat dome. Bloom brings on-site fuel cells (the bridge power option for sites where the utility interconnection is the bottleneck, which in Texas it always is).
The Atlas analog here is the engine contract. Rocketdyne won the propulsion contract and held the franchise across Atlas, Thor, and Jupiter for the next 30 years. Power is the propulsion of a fab.
Won't Tesla just power it themselves? The obvious pushback — that Tesla self-islands Terafab on solar and Megapacks and cuts the supplier book out — does not hold up. 
Terafab is in the ERCOT interconnection queue, Taylor approved a 2.4-million-square-foot Oncor substation to feed the campus, and the "solar" thread in Musk's keynote is about space-based generation for AI satellites, not Austin rooftops. 
On the Q1 2026 call, he was still being asked how Tesla would scale terrestrial solar to meet Terafab demand — not the language of a committed captive solution. A multi-gigawatt 2nm fab would also require tens of square miles of PV plus tens of GWh of storage to island, cannibalizing the merchant Megapack book Tesla sells to utilities at a higher margin. On-site Tesla generation may show up at the margins; the first-wave power spend still flows through the third-party vendors above.
Cooling: Liquid Wins at 2nm Density
EUV lithography tools dissipate heat at densities that air handlers cannot touch. Direct-to-chip liquid cooling (cold plates bolted directly onto the silicon, with facility water removing the heat) is the only path forward at 2nm and the GPU densities Tesla will want adjacent to the fab for AI5 validation. Vertiv (VRT) is the consolidated leader in critical digital infrastructure for data centers, and Modine Manufacturing (MOD) has built a fast-growing liquid cooling franchise out of an old HVAC business. Both are pure-play exposures to the metric that matters: rack density per square foot.
(The market already knows this. The chart below is the receipts.)
The Chart: Who's Already Growing Into the Order Book
The shape of those bars is the argument. EMCOR's mechanical and network construction segment, Vertiv's critical digital infrastructure line, and Bloom's data-center-driven top line are all growing materially faster than the diversified power-equipment averages. GEV's 4% is the conglomerate signal (its Power segment is dragged by legacy gas-turbine cycles); the orders that matter for Terafab are inside the Electrification line, which compounds faster. Translation: the names with the most concentrated DC exposure are also the ones most levered to a Texas fab order.
Construction & EPC: Who Builds the Building
Before any of the above arrives on a truck, somebody pours the slab and runs the conduit. That is EMCOR (EME), Quanta Services (PWR), Fluor (FLR), Jacobs Solutions (J), and Primoris (PRIM). EMCOR's U.S. mechanical and network/communications business is the cleanest read on data-center build-out exposure, and its 26.2% segment growth in FY2024 (the tallest bar above) is not a coincidence. PWR owns the electrical EPC franchise that ties high-voltage substations into hyperscaler campuses, which is exactly the work a Terafab campus needs first.
The interesting historical rhyme: there were almost 4,000 subcontracting firms associated with the Atlas program over the course of its operational life, 73 percent of which were small businesses (Key Industry Partners in the Atlas Success Story). The publicly traded EPCs are the prime contractors. Most of the Terafab dollar flows through them on its way to the long tail.
Risks: Single-Customer, Permitting, and the Musk Factor
Three things to underwrite before sizing positions. One, Terafab is one customer. Tesla's CFO acknowledged that the full Terafab cost, estimated at $20-25 billion, is not yet incorporated into Tesla's record capital expenditure plan for 2026, which means the financing path is still being assembled. Two, ERCOT interconnection and water permitting in Texas have killed faster projects than this one. Three, Musk timelines (the 4680 cell promised 3 TWh by 2030 and has not gotten close) are an input, not a fact. The supplier names with diversified hyperscaler exposure ride those risks better than the ones whose growth narrative requires Terafab specifically.
The Verdict
Tier-1 picks: VRT and EME. Both are pure plays on the work that has to happen first, both already show segment growth that proves the order book is forming, and both survive a Terafab slip because hyperscaler demand absorbs the slack. Tier-2: GEV and ETN, where the data-center exposure is real but blended into larger industrial books that move on slower cycles. Speculative: BE, where a single Tesla on-site power deal would re-rate the stock, and the downside is an unprofitable fuel-cell business if it does not.
Convair's Kearny Mesa floor went from ten engineers to twelve thousand workers because the contracts hit before the missiles flew. Terafab's switchgear and chiller orders will hit before the first 2nm wafer ships. The suppliers know which side of that ledger they want to be on.