TechCrunch reports that SpaceX’s planned Terafab chip fab in Texas will run on its own natural gas plants and battery arrays β and the fuel-source detail is less interesting than what it exposes: the AI race has quietly become an energy race, and whoever controls firm power controls the pace.
TERAFAB β KEY FIGURES (ESTIMATES & ANNOUNCED TARGETS)
As TechCrunch reports, SpaceX energy lead Riley Trettel told a public meeting in Grimes County, Texas β roughly 45 miles northwest of Houston β that the company would be “bringing our own power” for Terafab, its planned semiconductor fab. The power mix announced: the company’s own natural gas plants, firmed with large battery arrays. Terrestrial solar is absent from the disclosed plan, though the reporting notes the absence rather than quoting an explicit rejection β and critically, batteries (Tesla’s domain) are very much included. “Musk abandons solar” overstates what was actually said.
Terafab is being built with Tesla as a partner and is intended to produce chips for data centers operated by SpaceX and its xAI subsidiary. In exchange for the project, SpaceX secured a 100% county tax abatement, roughly $30 million in Texas state incentives, a commitment to spend at least $5 billion by 2030, and a target of at least 1,800 jobs by 2035. The initial phase carries a budget estimate of approximately $16.8 billion β a figure that has already moved once, down from an earlier estimate in the ~$55 billion range. No megawatt or gigawatt capacity figures were disclosed. This is an announced, locally approved project, not a built one, with a long execution horizon.
The project arrives shortly after SpaceX’s IPO, which raised approximately $87.5 billion. Musk has made a series of bets on natural gas infrastructure in recent months, and the pattern matters β but it is an industry-wide pattern, not an idiosyncrasy. Across the AI infrastructure landscape, hyperscalers and independents alike are reaching for dispatchable gas generation as the fastest way to put firm power beside compute. The engineering logic is consistent regardless of who is doing it.
TERAFAB β PROJECT TIMELINE
Earlier estimate (prior public disclosures)
Terafab project scoped at approximately $55 billion β figure subsequently revised downward to ~$16.8B initial phase.
2026 β SpaceX IPO + Grimes County meeting
SpaceX raises ~$87.5B in its IPO. Energy lead Riley Trettel announces “bringing our own power” at public meeting; 100% county tax abatement secured.
By 2030 β Spend milestone
Committed threshold: β₯$5 billion in capital expenditure. Project execution is phased; each phase is a place it can break.
Committed threshold: β₯1,800 jobs. Terafab intended to supply chips for SpaceX and xAI data centers β if the project reaches operation.
The key insight: The phrase “bringing our own power” is the strategic signal, not the fuel source. Building behind-the-meter generation bypasses the multi-year grid interconnection queue that is now the real gating factor for AI capacity in the United States. SpaceX is not making a statement about solar; it is solving an infrastructure bottleneck that every serious AI builder faces.
The Structural Read
Peel the AI infrastructure story back a layer at a time and you arrive at an uncomfortable truth. The binding constraint started with chips, then moved to memory, then to assembly and the compute floor. Now it has migrated one layer further down the stack: to electrons. A semiconductor fab running around the clock alongside a data center complex represents exactly the load profile where firm, dispatchable generation wins on engineering, not politics. Intermittent solar cannot supply 24/7 baseload at that scale and reliability β gas plus battery firming can. That is the honest, narrow claim. It is also an industry-wide reality, not a Musk idiosyncrasy or a referendum on renewables.
The deeper play is the interconnection queue bypass. Getting new capacity connected to the U.S. grid now takes years β the queue is measured in gigawatts of stranded projects waiting for transmission studies and regulatory clearance. Building your own generation behind the meter converts power from a utility you wait on into an asset you own and control. Speed and reliability become structural advantages, not operational preferences. This is the logic that explains the choice, irrespective of any ideological reading.
Terafab, read alongside SpaceX’s first public earnings, completes a picture of vertical integration that almost no other organization is attempting at this scope: generation to fab to chips to models to data centers, one owner from the electrons up. That is a staggering execution bet. Owning each layer is also a place each layer can break β the project is early, the budget has already moved once, and the timeline runs nearly a decade. The vertical-integration reading is real; it should not be inflated into inevitability.
Map of AI β Power Layer
“The constraint that defines this era keeps migrating downward. It was compute, then memory, now power. The companies that anticipate the next migration β and own the asset before the queue forms β set the pace for everyone else. ‘Bringing our own power’ is how you win a race where the bottleneck is the grid, not the GPU.”
Where the Constraint Has Landed
Compute / GPUs
MANAGED
Nvidia supply has expanded; the GPU bottleneck has eased for major players, though allocation still matters at the frontier.
Memory / HBM
ACTIVE
HBM supply remains tight; the memory layer beneath the GPU is where the real throughput ceiling currently sits for many workloads.
Power / Grid Access
BINDING NOW
Firm baseload is the new scarce input. Grid interconnection queues run years. Behind-the-meter generation is the emerging structural advantage β Terafab is one data point in an industry-wide pattern.
Three Implications
IMPLICATION 1 β BRING-YOUR-OWN-POWER BECOMES A MOAT
Behind-the-meter generation is no longer an engineering curiosity β it is a competitive differentiator. Organizations that can site their own dispatchable generation next to compute will be able to scale faster than those waiting in the interconnection queue. The scarcest input has shifted from chips to electrons, and the moat now belongs to whoever controls the supply. Read alongside Beyond NVIDIA’s Moat for the full framing.
IMPLICATION 2 β VERTICAL INTEGRATION TO THE ELECTRONS IS AN EXECUTION BET, NOT A GUARANTEE
Generation to fab to chips to models to data centers is the most ambitious vertical stack anyone has announced in this cycle. Each layer represents both a competitive advantage and a new failure point. The project is early β a public-meeting statement, local incentives, a budget that has moved once, and timelines stretching to 2035. The structural logic is sound; the execution is years of hard work away from being proven.
IMPLICATION 3 β GAS-FOR-AI IS AN INDUSTRY PATTERN, NOT A BRAND STORY
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
