Musk plants a flag in austin, betting $50b he can outrun the chip famine he predicts
Elon Musk just told the crowd at an Austin warehouse what the semiconductor giants refuse to admit aloud: the planet’s appetite for AI-grade silicon will soon exceed the planet’s ability to cook it. His answer is Terafab, a 2-nanometre foundry that doesn’t exist yet, financed largely by a SpaceX IPO he wants to price this summer at a $1.75 trillion valuation.
A factory that orbits, eventually
The roadmap is pure Musk. Start in Texas with a pilot line able to tape-out “any chip on Earth,” then loft the same process into vacuum-hardened satellites that together cough out one terawatt of compute per year—enough to power a million Optimus robots and the Starship fleet he still insists will colonise Mars. He showed renderings of refrigerator-sized orbital data centres, each radiating 100 kW of heat into the void, later generations scaling to full megawatt nodes. The FCC filing for one million such satellites is already stamped and waiting.
Ground zero is a scrubby patch east of Tesla’s Gigafactory, a site chosen because the governor, Greg Abbott, promised expedited permits and because Samsung’s neighbouring plant can babysit the first wafers while Musk’s own EUV lithography ships arrive. Tesla will design the edge-inference silicon for Cybercab and the humanoid line; SpaceX and its captive AI shop xAI will swallow the high-power SKUs aimed at orbital super-clusters. Musk swears the split is 80 % xAI, 20 % everything else.
Billions in the air, zero fabs on the ground
Conventional wisdom says a leading-edge fab costs $20 billion and six years of yield hell. Musk waved that away: “We’ll run lean, iterate in-house, and copy Tesla’s vertical-integration playbook.” He neglected to mention that Tesla still buys FSD chips from TSMC and memory from Micron because its own designs weren’t ready. Analysts at SemiWiki note that even Intel, with fifty years of process heritage, bled cash trying to leap from 14 nm to 7 nm. Musk’s timeline—pilot output in 2026, full terawatt orbitals before 2030—is, at best, aspirational.
Yet the capital is already moving. SpaceX’s planned $50 billion raise would eclipse the largest tech IPO in history, eclipsing Alibaba’s 2014 float. Bankers whisper that the prospectus will earmark 35 % of proceeds for “off-planet compute infrastructure,” a phrase that would have sounded like sci-fi boilerplate five years ago. Musk needs the money: each terawatt constellation demands ten thousand satellite launches, even at Starship’s dramatically lowered kilo-to-LEO price.
The chip shortage he refuses to outgrow
Behind the spectacle lies a simpler truth. Tesla’s current fleet burns through roughly 1.5 million custom SoCs per quarter. With full self-driving still unfinished and Optimus prototypes multiplying on the Fremont test floor, that number will triple within two years. TSMC’s Arizona plant won’t come online until 2028; Samsung’s Taylor expansion is already sold out to Apple and Qualcomm. Musk’s choices are queue forever or build his own roulette table.
He chose the table. Whether the ball lands on red or the house cleans him out won’t be clear until the first wafer exits Austin under Tesla’s own laser stamp. For now, the wager is registered: $50 billion of other people’s money, a million satellites, and one man’s promise that physics will bow to ambition. The desert outside Austin is quiet tonight, but the shovels arrive at dawn.