Musk bets texas on a homemade-terawatt future that big fab won’t touch
Elon Musk just told the semiconductor industry to get out of his way. On a downtown Austin stage flanked by Governor Greg Abbott, he sketched Terafab—a 2-nm wafer plant that will start inside city limits, scale to a terawatt of off-planet compute, and feed everything from Tesla robotaxis to orbital AI clusters Musk wants to fire off the Moon.
The numbers he tossed around are comically large
100–200 GW of Earth-side silicon each year. Another terawatt for space. One million data-center satellites. A summer IPO aiming to haul $50 billion into SpaceX at a $1.75 trillion valuation. The audience clapped; veteran fab hands reached for antacid. A green-field foundry costs $20 billion and needs three years before a single salable die comes out. Musk wants the first Terafab ovens humming in Austin—a city that still can’t keep the grid lit during February cold snaps.
He has done this before: promised a revolution, borrowed tomorrow’s headlines, then quietly rewritten the schedule. Battery-swapping, solar roof tiles, robotaxis by 2020—each slid into the fog. But chips are not software patches you push overnight. They are physics, chemistry, and the most delicate capital equipment on Earth. ASML won’t sell you an EUV scanner on Elon time; each machine shows up in 747s and needs six months to stop shaking.
Yet the timing is brutal for incumbents
TSMC, Samsung, and Intel are already booked solid on AI accelerators. Tesla’s current generation of Full-Drive hardware ships with Samsung 7-nm silicon made an hour east of where Musk stood. Even that arrangement leaves Tesla scavenging for wafers. Add Optimus robots, the xAI Grok fleet, and the perpetual Starship backlog and you get why Musk is done begging. “Either we build the Terafab or we run out of chips,” he shrugged, sounding like a man who once decided to build rockets because the Russians overcharged him.
The state handed him tax abatements before drawings exist. Abbott called the plant “the next Texas oil boom,” apparently forgetting that oil bubbled out of the ground; etch-a-sketching transistors at 2 nm demands water, stable power, and PhD armies Austin does not yet house. Samsung’s nearby fab sucked so much groundwater last summer that rice farmers downstream saw their fields crack.
Space is the escape hatch
If earthly regulators balk at energy or emissions, Musk will loft the fabs to orbit. Literally. He showed a rendered lunar launch railgun flinging satellite data centers into cislunar space, each node radiating waste heat into the vacuum. One million satellites sounds like a Bond villain’s to-do list, but the FCC filing is already in. The first 100 kW prototype, he claims, will ride a Falcon 9 late next year—hardware, schedules, and physics permitting.
Investors barely flinched. SpaceX’s last tender valued the company at $210 billion; a $1.75 trillion IPO would 8× that in months. Musk’s pitch: owning the stack—launch, silicon, model weights, and the orbiting cloud where the model runs—turns vertical integration into orbital moats. Competitors stuck on Earth pay rent to gravity; Musk would collect the rent.
Still, the house wins when physics meets spreadsheets. Each node needs radiation-hardened memory, laser interlinks that don’t fry in Van Allen belts, and a power budget that makes Texas blush. A single megawatt-class satellite dissipating heat in vacuum needs radiators the size of soccer fields. Ask any space-solar startup how easy that is.
The subtext is control
By owning the fabs, Musk decides which models run, who gets priority during shortages, and how fast competitors can iterate. Tesla already feeds xAI unlabeled driving data; xAI ships Grok dashboards back to the fleet. Fold in homegrown silicon and the feedback loop disappears inside one corporate event horizon. Regulators watching Tesla’s every crash might soon subpoena orbital logs beamed from a server no one can physically raid.
Musk ended the night with his trademark coda: “Abundance is inevitable—if we survive the bureaucracy.” The crowd roared. Outside, the line for the parking garage snarled around the block, every Tesla crawling home on chips bought from the very suppliers Musk now vows to replace. The fab has no dirt turned, no ASML order logged, no public timeline. But the headline is already etched: Texas, 2 nm, one terawatt—believe it when the first wafer ships, not when the slideshow ends.