SpaceX’s $17B Texas Chip Fab Power Plan: Why the Real Trade Is in Turbines, Not Silicon
SpaceX confirmed on August 5, 2026, that it will construct its own natural gas power plants and large-scale battery storage for the Terafab semiconductor megafactory in Grimes County, Texas. Riley Trettel, who leads SpaceX's energy and data center development, told a public meeting the facility "will be self-sufficient," generating electricity rather than drawing from the ERCOT grid. The batteries will almost certainly be Tesla Megapacks—Tesla disclosed $405 million of first-half 2026 Megapack sales to SpaceX against $307 million of cost, a gross margin near 24%. The energy layer is producing revenue before Terafab produces a commercial wafer.
Terafab, the joint Tesla-SpaceX venture announced by Musk in March, is planned at roughly 100 million square feet across 3,000 acres near the former Gibbons Creek coal plant. Phase one carries an announced $16.8 billion price tag with at least 3,000 permanent jobs. Texas has offered a $30 million Enterprise Fund grant and JETI tax abatements. The factory's stated mission: vertical production of advanced AI and memory chips for Tesla's Optimus robots, Cybercabs, and SpaceX's orbital data centers, using Intel's 14A process.
The Grid Cannot Keep Up
Trettel's announcement is pragmatic. ERCOT was tracking approximately 410 GW of large-load interconnection requests in April 2026, roughly 87% linked to data centers. Texas has tightened controls over how fast large loads can connect. For any facility needing power within a commercially relevant timeframe, building behind the meter is the only path around a grid queue that can outlast the AI hardware it serves.
SpaceX has done this before—the xAI Colossus data center used mobile gas turbines paired with battery storage to reach operational capacity ahead of grid connections. GE Vernova, a likely turbine supplier for such projects, reported a $176 billion backlog. Siemens Energy carries €154 billion.
What the Headline Numbers Conceal
At $16.8 billion for 100 million square feet, Terafab's implied investment is $168 per square foot—before cleanrooms, lithography tools, process equipment, or the power plants themselves. TSMC plans $60–64 billion of 2026 capital spending across its established platform and has committed $100 billion more to Arizona. The 100 million square feet almost certainly describes the long-term campus envelope, not phase-one equipped capacity.
Intel's 14A process—the node Terafab intends to use—adds a constraint capital cannot compress. Intel expects internal risk production in second-half 2027, with high-volume ramps targeted for 2028. A delayed ramp delays every downstream step: packaging qualification, defect-density reduction, and commercial yield. The project's own tax filings reinforce the financial thinness: official applications state the Texas tax limitation reduces Terafab's effective burden by approximately 48%, and that the project would likely fail its required return threshold without it.
The Utilization Trap
TSMC produced more than 12,000 products for over 500 customers in 2025. That diversification absorbs capacity fluctuations and generates continuous yield-learning data. Terafab concentrates demand around Tesla, SpaceX, and xAI—related parties with synchronized exposure to AI capital spending. One postponed robot program or delayed satellite constellation can leave billions in equipment sitting idle. Vertical integration that runs at high utilization creates margins; vertical integration at low utilization becomes a permanent internal subsidy.
Own the Bottleneck
The highest-probability institutional position on Terafab is indirect. Every competing AI infrastructure project—Terafab included—must purchase from the same constrained suppliers: gas turbines, high-voltage switchgear, EUV lithography systems, etch and deposition equipment, HBM memory, advanced packaging substrates, ultra-pure-water systems. ASML plans capacity of roughly 65 low-NA EUV systems in 2026. No amount of capital manufactures additional lithography tools faster.
The disciplined trade pairs long exposure to contracted bottleneck suppliers against short or underweight positions in speculative, unpowered AI infrastructure—developers without turbine reservations, battery installers lacking proprietary controls, and semiconductor ventures valued on announced acreage rather than qualified wafer output. Four milestones would convert Terafab from a power-and-infrastructure project with a semiconductor option into a credible megafab: disclosed power MW and battery MWh, binding equipment and gas-transport commitments, a successful 14A tape-out with yield data, and enforceable internal wafer-purchase contracts. None has surfaced in public filings. Until they do, the informed position is to own the picks and shovels that every vision—Musk's included—must buy.
not investment advice