Fifty Pentagons

On 6 August, Tesla and SpaceX announced an initial joint investment of $16.8 billion to start building Terafab, an advanced chip factory in Grimes County, Texas. The plans are not modest. Regulatory filings put the first phase at $55 billion and the full build-out at up to $119 billion. SpaceX describes more than 100 million square feet of planned manufacturing space, and Elon Musk claims the finished complex will be fifty times the size of the Pentagon: the largest building ever constructed, by his telling the most valuable. The chips it makes are earmarked for Tesla's Optimus robots and Cybercab robotaxis, for xAI's models and for the data centres SpaceX wants to put in orbit.

The claim on the table

The frontier-ledger claim (chapters 2, 6 and 9) says the AI build-out is history's largest private capital-formation event, and that the gains are being banked as assets owned by the frontier with no line reserving a claim for anyone else. Terafab is that claim with a county name attached. One family of companies, one shared principal shareholder, is building sovereign-scale industrial capacity to make the processors for its own robots, its own vehicles and its own models. The vertical stack the book describes in chapter 6, from silicon to model to product, is here compressed into a single site plan.

The scale, in public money

The comparison that matters is not the Pentagon. The US CHIPS and Science Act, the largest public semiconductor programme in American history, allocated $52.7 billion in federal funding. One private fab project now carries filings at up to $119 billion, more than twice the entire public programme, financed by two companies and answerable to their cap tables. When the book argues that the productive substrate of the AI economy is being assembled privately at a pace no public instrument matches, this is what a data point looks like.

What a fab for robots means

Terafab's stated products are worth reading slowly. Chips for humanoid robots. Chips for driverless taxis. Chips for orbital data centres. This is not capacity for phones or consoles; it is capacity for machines that perform work. The factory that will one day be the world's largest building exists to lower the cost of automating labour, and every dollar of value it creates accrues to the shareholders of Tesla, SpaceX and xAI. Nothing in the public announcements or filings reserves anything for anyone else, and there is no reason it would: that is how private capital formation works, which is the book's point rather than an accusation. Notably, the announcements do not discuss labour at all. In the book's reading that silence is the substance: the capital that automates work enters the ledger in nine-figure lines, while the work being automated appears nowhere, because the people who do it are not a party to the deal and no instrument exists through which they could be. That instrument is what predistribution proposes, and Grimes County is a measure of how far ahead the concrete is of the idea.

The honest hedge

Announced is not built. Musk projects have a documented habit of arriving late and smaller than the launch event, and $119 billion is a filing ceiling, not a purchase order. The committed number today is $16.8 billion, which would still rank among the largest single industrial investments in US history. If the ground stays unbroken or the phases stall, this entry gets a follow-up saying so.

Why this confirms

The book's window argument says the ownership of the AI capital stock is being settled now, during formation, while the public debate looks elsewhere. On 6 August a piece of that stock the size of a small nation's annual output moved from prospectus to committed capital, its ownership fully allocated before a single wall exists. That is the window argument in miniature: the decisive moment is not when the factory opens but when the money is assigned, and that moment has now passed for Terafab. The window did not close this week. But it narrowed by one very large building, and the standing question has never had a more literal shape: not whether the machine will take your job, but who owns the machine.