The Secret Plan Behind SpaceX's $100 Billion Project

SpaceX and Louisiana Economic Development announced Starbase, Louisiana on Aug. 25, 2026: a $100 billion Vermilion Parish spaceport. Gov. Landry and Gwynne Shotwell used 125,000 acres on stage for the former Exxon tract; the Times-Picayune puts the spaceport on roughly 40,000 acres inside that tract. Farzad's football-field line (~95,000 fields) is his on-tape conversion of the 125k figure, not a filing. Farzad's cut, posted September 14, 2026 under the title The Secret Plan Behind SpaceX's $100 Billion Project, is not another pad tour. The claim is that the same plot that hosts launch paths also sits on one of the densest natural-gas networks in America, and that dual use (rocket fuel plus on-site power) is how SpaceX stays ahead of the AI power wall while Terafab and orbital compute still need watts. Treat numbers below as what he cites on tape unless a separate filing confirms them. ASR on the YouTube track mangled Terafab as "Paraphab" and Starmind as "Storm Mine" / "Starmine" / "Star Mine." Those are caption errors. Starcloud is a different company.

Pads are the excuse. Power is the product

SpaceX still flies under 300 launches a year on Falcon 9 and Falcon Heavy across US pads. Starship's bet is thousands of flights a year once full reuse works: catch the booster with chopsticks, refuel, stack, and fly again in hours, then minutes, the way airlines turn planes. More flights need more pads. One pad serializes you. Starbase Louisiana exists to give SpaceX land for many paths over the Gulf so mass to orbit can climb without waiting on a single tower.

That is the public story. Farzad's twist is location. Vermilion Parish sits on the Gulf coast for over-water trajectories and away from dense neighborhoods. It also sits near Henry Hub, the Louisiana junction with huge pipeline connectivity, storage, production, and trading. Louisiana officials cited abundant natural gas as a Starbase advantage. LNG is chilled natural gas. After different equipment, the same molecule stream can become electricity in a turbine or become subcooled liquid methane for Starship, which Farzad frames as roughly 85% to 95% of LNG's makeup, plus liquid oxygen taken from air. One supply system. Two outputs. Fuel for the rocket. Watts for the campus. That dual-use read of the LED gas line is Farzad's thesis, not a SpaceX AI-turbine filing for Louisiana.

Chips without power are expensive paperweights

For years the AI scare story was GPUs. Farzad says the binding constraint is shifting to buildings, networking, cooling, and especially power. He cites the International Energy Agency: data centers used about 415 terawatt-hours worldwide in 2024, demand more than doubles by 2030, and roughly one-fifth of planned projects could slip if grid risks are not fixed. Berkeley Lab's Queued Up series says the median time from a generation interconnection request to commercial operation now exceeds five years. That is plants joining the grid, not a data-center load-interconnect clock. Farzad uses it as the wait-for-power analogy. Five years is forever in model cycles.

At a G20 innovation meeting days before this recording, Elon cited analyst consensus that AI chips could face at least a 15-gigawatt power shortfall in 2027. Farzad's translation: racks of valuable silicon with nowhere to plug in. Fifteen gigawatts is continuous power for roughly 12 million average US homes on his framing. SpaceX's May 20, 2026 S-1 disclosed billions in gas-turbine purchases for AI data-center ops (about $2.8 billion across the named agreements) and said Colossus II is expected to run primarily on a dedicated natural-gas plant. The proposed Terafab Texas campus is later/separate: Grimes County + Riley Trettel (Aug 2026) say on-site generation and batteries, not the normal grid, with Trettel citing natural gas. That is not the May S-1 line. Same logic xAI used in Memphis: Colossus fully operational in 122 days, with Nvidia saying first racks hit the floor and training began 19 days later.

When turbines become the bottleneck, make blades

Everyone else racing for grid interconnects hit the next shortage: the turbines themselves. Farzad cites GE Vernova selling delivery slots years out, with customers already buying many 2030 slots because construction teams could not promise sites ready for turbines in 2029. SpaceX moved one step further back. It is hiring for an industrial gas-turbine blade and vane manufacturing line in Bastrop, Texas: five-axis milling, laser drilling, EDM, nickel-based superalloys, casting tools for internal cooling passages. Elon's line on tape: casting blades and vanes is the limiting factor in natural-gas turbine production, and bringing that work inside can pull turbine deployment forward by as much as 18 months. Follow the constraint until you own the constraint.

Starbase Louisiana then becomes an option almost no pure AI company can copy. An industrial campus next to deep gas infrastructure already needs power for manufacturing, propellant, launch ops, and the rest of Starship. Once that energy system exists, adding compute is a smaller step than begging a utility from an empty field.

What a shortfall does to Nvidia's loop

Nvidia does not win when a customer merely owns GPUs. Nvidia wins when customers install the next generation, run high utilization, and order again. A 15-gigawatt shortfall threatens that loop: capacity slips, deliveries back up, revenue pushes out, and a chip glut becomes possible even while long-term demand for intelligence stays huge. SpaceX is trying to keep the loop moving. Nvidia (Aug 24, 2026) said Vera CPUs and Vera Rubin will power SpaceXAI's ground AI factories and that the first-generation Starmind orbital satellite is an optimized Vera Rubin NVL72. Starcloud is a separate YC orbital-compute firm, not SpaceX's constellation. Nvidia supplies the compute architecture. SpaceX supplies ground power, launch to orbit, Starlink to tie sites together, and eventually far more solar in continuous sunlight. Jensen's incentive, in Farzad's framing, is that SpaceX expands the physical market for Nvidia systems instead of acting like one more stranded customer.

Two timelines share one campus. Near term: natural gas, batteries, and an internal turbine supply chain beat the terrestrial shortage. Longer term: orbital solar compute escapes Earth utilities after Starship is rapidly reusable and satellite manufacturing hits volume. Louisiana is the bridge.

Launch rate is the hidden math under orbital watts

Farzad rates SpaceX's AI1 / Starmind design at about 175 kW average. At that rating, 1 gigawatt of orbital compute needs roughly 5,700 satellites, which he maps to on the order of a couple thousand launches per year before replacements, nearly 10x today's Falcon cadence. Treat 175 kW as on tape. SpaceX's June 2026 video was 120 kW average / 150 kW peak; Elon on July 16 said average ~160 kW / peak ~250 kW. The launch count is his conversion, not a manifest. Starbase Texas and Florida keep launching. Louisiana adds many more paths and is framed as the biggest of the set. SpaceX still talks about up to a million satellites over time. That number is impossible without factories and pads that run continuously.

Starlink rides the same capacity. Larger birds carry more bandwidth and more mass to orbit. More pads mean more network expansion, replacements, direct-to-cell cash, and funding while Starmind scales. The loop he closes: Starlink cash funds launches; more launches improve Starship turnaround; cheaper Starship deploys Starmind; Starmind creates more launch demand and AI revenue; that revenue funds rockets, sats, power, chips, and solar, with SpaceX and Tesla each aiming at 100 gigawatts per year of solar production capacity in his closing stack.

What to watch

The public Starbase Louisiana announcement already priced the $100 billion and the acreage. This Exclusive is the power thesis on top of that. Watch whether Louisiana gas-to-power and propellant actually stand up on SpaceX's clock. Watch Bastrop blade-and-vane hiring turn into real turbine throughput. Watch whether terrestrial hosting keeps Nvidia's install-and-reorder loop alive through a 2027 shortfall scare. Watch whether Starship flight rate and Starmind economics ever make the 175 kW-per-sat math real. Pads without watts are a launch story. Watts without pads leave orbital compute on a slide. Farzad's point is that Starbase Louisiana is built to own both.