Terafab is not one Texas tax agreement. It is eight — four with Anderson-Shiro CISD and four with Iola ISD, both in Grimes County, one per phase per district. Texas has published a summary for every one of them, and the summaries carry three numbers the coverage has not: what the state expects to give up, what the applicant is legally required to deliver, and when the money is supposed to land.
Added up, the eight agreements forgo an estimated $1,660,434,060 of school maintenance-and-operations tax across incentive periods running to 2046. Against that, the state's own summary of each agreement records the same line: minimum required jobs, ten. Eight agreements, eighty jobs.
Two things have to be said about that $1.66 billion before it is used for anything. It is the state's estimate, not a settled bill — it is built on taxable values the applicant itself supplied, projected across two decades of a campus that does not exist yet, and every year past the first few is a forecast. And it is a reduction in tax on property that is not there today, not a transfer of money the districts currently collect. Both caveats cut the number's precision. Neither changes its order of magnitude, and the order of magnitude is the point.
The points
- $1.66 billion of school M&O tax forgone across all eight agreements. Derived: the sum of the eight gross tax benefits the state published, one per agreement.
- $166.0 million of performance bonds, and in every single agreement the bond is exactly 10.0% of that agreement's tax benefit. It is a formula, not a negotiation — which is why quoting one bond as a measure of commitment reads the deal backwards.
- Ten required jobs per agreement, recorded identically in all eight summaries. Grimes County's population of 29,268 puts the project in the statute's lowest job tier.
- A minimum wage requirement of $66,668, against 110% of the county's average annual wage of $66,667.
- Phase 1's eligible property is $16,801,000,000 — Anderson-Shiro's $10,366,217,000 plus Iola's $6,434,783,000. Disclosed in the two applications, and the sum is exact.
- Construction 2026 to 2028; incentive periods 2029 to 2038 for phase 1, then rolling forward a phase at a time to a last incentive year of 2046.
The famous $16.8 billion is two tax filings added together
Every account of this project, including the Governor's own August announcement, puts phase one at "$16.8 billion." That figure has been read as a capital budget. It is not one.
The two phase-one applications each carry a placement-in-service schedule of eligible property. Anderson-Shiro's totals $10,366,217,000. Iola's totals $6,434,783,000. They sum to $16,801,000,000 — the $16.8 billion, to the last decimal place the number is ever quoted at.
That matters for two reasons. Eligible property is a tax category, not a cost. It excludes land and land acquisition, inventory, and anything already standing on the site — all of it explicitly carved out in the applications. So the real phase-one outlay is the $16.8 billion plus whatever the land and the excluded items cost, and nobody outside the consortium has that figure.
And it is split across two school districts because the campus straddles a district line. Neither half is a company, a phase boundary or a funding share. Anyone reading the 62/38 split between the two filings as a split between two balance sheets is reading a map of Grimes County.
The schedule the filings put on the money
The same two applications date the property, year by year. Combined:
| Year | Eligible property placed in service | Cumulative | Share of phase 1 |
|---|---|---|---|
| 2026 | $200,000,000 | $200,000,000 | 1.2% |
| 2027 | $6,000,400,000 | $6,200,400,000 | 36.9% |
| 2028 | $10,600,600,000 | $16,801,000,000 | 100% |
Note the round totals. $200.0 million, then $6.0004 billion, then $10.6006 billion — the applicant back-solved to clean numbers and split them across the district line. Treat the annual figures as a planning shape rather than a forecast anyone will be held to.
The shape is still the useful part: more than a third of phase one is scheduled to be in service before 2028 begins. And placement in service lags cash — property is paid for before it is placed. Whatever the spending curve is, it is earlier than this table.
What the entity name proves, and what it does not
All eight applications are filed by one applicant: TeraFab AI, LLC. Not Tesla, not SpaceX.
In August we measured Terafab against every building both companies have put up and ended by asking whether it lands on Tesla's balance sheet, SpaceX's, or a vehicle outside both. A separately named applicant looks like an answer to that question. It is not one, and it is worth being exact about why, because a named party is a named party and nothing more. It proves a distinct Texas contracting entity exists and is the counterparty on all eight agreements. That is the entire load it can bear.
Everything around the name points somewhere else. The applications are filed from 1 Rocket Road, Hawthorne — SpaceX's headquarters. The Comptroller's approval letters are addressed to a property-tax manager at Space Exploration Technologies Corp and copied to a second SpaceX employee. The Governor announced the project as SpaceX's and awarded its accompanying $30 million Enterprise Fund grant to SpaceX. And the project description filed with the applications says the project "is being led by a consortium of affiliated advanced technology companies, including Tesla, Inc., Space Exploration Technologies Corp. (SpaceX), and xAI Corp."
So the documents name a separate entity, route the correspondence to SpaceX, and describe a three-company consortium — Tesla, SpaceX and xAI, the third of which has not appeared in this project's public framing before.
None of that is ownership. A tax agreement identifies who signs, not who pays. These filings do not say who owns TeraFab AI, LLC, who capitalises it, whose balance sheet consolidates it, or how its cost is shared among three companies. This is where the reader wants a number and where most analysis quietly supplies one. We are not going to. The honest statement is that the split is undisclosed, that the filings give no basis for inferring one, and that the appearance of xAI means even the number of parties is not what it was assumed to be.
Three job numbers for the same phase
Phase one has a job figure in three documents, and they differ by two orders of magnitude:
- 3,000 — the Governor's announcement of 6 August 2026.
- 1,000 — the applicant's own filed economic studies: 617 direct permanent jobs in Anderson-Shiro, 383 in Iola, all appearing in 2029 and flat for the following two decades.
- 20 — the statutory minimum required jobs, ten in each of the two phase-one agreements.
These are three different things and all three are honest on their own terms: a political headline, a consultant's projection, and a legal floor. The gap between them is not evidence of bad faith. It is what the second chart shows, and it is the reason the floor is worth knowing — the floor is the only one of the three that is enforceable, and it is what the bond is posted against.
The projection is also not a jobs ramp. It is 1,000 jobs arriving in a single year, 2029, and never moving again through 2063. That is a modelling convention, not an operating plan.
What this does to our models
Our published Tesla and SpaceX models each carry a Terafab phase-one programme of $16.8 billion, starting six quarters out from an August 2026 basis — call it early 2028 — and running twelve quarters from there.
The filings say the schedule starts earlier. On the applicant's own placement-in-service dates, $6.20 billion is due in service before 2028 begins, and cash leads placement. Both models are late, by roughly five quarters at the front of the curve. That is a timing correction to our own work, it lands squarely on a driver we carry, and the fix is a model change rather than an article claim — so it is filed as one, not made here.
The split is the softer half. Both models divide the $16.8 billion 40/60 between Tesla and SpaceX, and both notes say plainly that the split is assumed because none is disclosed. The filings do not resolve that. They make it harder: there is one applicant, no disclosed shares, and a third named participant neither model carries. An assumed two-way split of a number that is a tax category rather than a budget is doing less work than it looks like it is doing.
Nothing here touches revenue, and nothing here moves a fair value. This is a capital-timing line and a disclosure question, and pretending it prices out to a share price would be inventing precision the documents do not contain.
What to watch
- Whether the executed agreements require more than the statutory ten jobs. Texas publishes the applications and the recommendations; the signed agreements are the document that would settle it, and the ten is a floor a district is free to negotiate above.
- Whether TeraFab AI, LLC's ownership ever appears in a filing. Tesla's next annual report is the most likely place a consolidation or equity-method treatment would surface.
- The first appraisal roll carrying real Terafab property, which replaces the applicant's projected taxable values with assessed ones and turns the $1.66 billion estimate into arithmetic.
- Whether phase-one spending shows up in 2027 as the filings schedule it, in either company's capital expenditure — the test of the timing correction above.
- What xAI's role is. It is named in the project description as one of three consortium members and appears nowhere else in the project's public accounting.
Every per-agreement figure — tax benefit, bond, required jobs, wage, construction and incentive years — is disclosed in the Texas Comptroller's June 2026 recommendation summaries for the eight Terafab applications, J0035 through J0042. The eligible-property schedules, the consortium description and the Hawthorne filing address are quoted from the applications as filed, and the permanent-job projections from the economic-impact studies filed with them; those projections are the applicant's consultants' work, not a commitment. Every total, ratio, sum and share in this piece is our arithmetic on those published figures. The tax benefit throughout is the state's own estimate, resting on taxable values the applicant supplied and projected to 2046 — not a settled amount. The larger job figure is the Governor's announcement of 6 August 2026. The programme timing and the 40/60 split in our two models are ours, and are assumptions rather than disclosures.
