TSLA376.37+0.0%
SPCX185.00+0.0%
Tape1

Gold-lane node · merge relevance 98

Terafab — the shared chip moonshot

A cited ~$55B-class captive advanced-packaging / process node that would feed Tesla inference and SpaceX AI satellites. Framework agreement exists. Projects still “subject to separate negotiations.” Score it as a letter of intent, not a factory.

Completion

8%

5–12%

Merge relevance

98

Relevance

Cited bill

~$55B

$40.0B–$70.0B

Tesla inference Cars · Optimus · Cortex

TERAFAB shared node

SpaceX silicon AI sats · Starlink V3

Today both companies are tenants. Terafab is the argument they should not be.

Daily desk · live + forensic

Today’s update

Sep 15 2026

Completion

8%

range 5–12%

est. from public artifacts

Since last quarter +2 pts · last official Jul 1 2026

Completion tape

  • Q1'24 0%
  • Q3'24 1%
  • Q1'25 3%
  • Q3'25 5%
  • Q1'26 6%
  • Q3'26 8%
IN TRENCHES

X1

Terafab (shared chip moonshot)

Merge relevance 98

Child trenches · % complete

Today’s still

Reconstruction — empty packaging hall. Not a Tesla/SpaceX cleanroom. No wafer starts on this card.
Reconstruction — empty packaging hall. Not a Tesla/SpaceX cleanroom. No wafer starts on this card.

Today’s video

Reconstruction · Empty hall reconReconstruction clip. A dark bay is not a fab. Label it as such.

Stills

  • Reconstruction — empty packaging hall. Not a Tesla/SpaceX cleanroom. No wafer starts on this card.
    Reconstruction — empty packaging hall. Not a Tesla/SpaceX cleanroom. No wafer starts on this card.
  • Reconstruction — a fenced Texas plot. Dirt is not a process node. Site file (TF1) is still 12%.
    Reconstruction — a fenced Texas plot. Dirt is not a process node. Site file (TF1) is still 12%.
  • A wafer. Inference for cars, bots, and sats is the cited bill. Yield is not on the page.
    A wafer. Inference for cars, bots, and sats is the cited bill. Yield is not on the page.

Videos

  • Reconstruction · Empty hall reconReconstruction clip. A dark bay is not a fab. Label it as such.
  • Reconstruction · TERAFAB explainerUnofficial explainer. Not a wafer start. Not Tesla footage of a plant that does not exist.
  • Reconstruction · Terafab launch eventEvent talk. Applause is not a tool order. Framework remains subject to separate negotiations.

Forensic pulses

Eight information trenches

The parent card is a letter of intent. These eight files are what a real fab would have to clear. Bills stay on the parent so the rings do not double-count ~$55B.

What a fab actually is

  1. 1. Dirt, power, water, a state incentive — then a process letter, not a slogan.
  2. 2. Packaging is the scarce step. CoWoS-class, not a 2 nm fairy tale.
  3. 3. A yield org and an ASML slot. Neither has shipped.
  4. 4. Offtake split and a transfer-pricing collar a Tesla minority can take to court.

Pulse

  • 2026-09-11

    Terafab still an LOI. Eight child files opened on the board.

Open the wire

Promised

A captive advanced-packaging / process node so neither company sits in a TSMC queue forever — inference for cars and bots, silicon for AI satellites.

Shipped

A cited number, a framework agreement, and a name in a bio. Related-party talk. Not a clean room.

Missing

The fab. The process. The $55B. The talent. Customers other than themselves. CHIPS / CFIUS path. Transfer pricing a Tesla minority holder can audit.

Why this is the merge trench

Highest merge-relevance node in the set (98). Not because a factory exists — because the bill is large enough that combining the P&Ls is the only politically honest way to eat it. Tesla holders funding a fab for SpaceX sats is the conflict. A merger collapses the related-party file. That is the trench.

  • Shared necessity40/40
  • Bill neither wants alone30/30
  • Agency conflict20/20
  • Execution (LOI is real, fab is not)8/10
  • Total98/100
IN TRENCHES

This is not a fab

No public yield. No wafer starts as a Tesla/SpaceX plant. Elon’s bio pointing at Terafab is not a process node. 5–12% complete, estimated from public artifacts.

The bill neither wants alone

Cost to date is unknown design/pre-fab spend ($0.2–2B range). Cost to finish is the cited ~$55B-class ($40–70B). Value if hit: inference independence for cars, bots, sats (~$300B option). Value if miss: both remain Nvidia/TSMC tenants; the merge case weakens (~−$80B).

Cost to date

$200M–$2.0B

Unknown — design/pre-fab spend; estimate range

Cost to finish

$40.0B–$70.0B

Cited ~$55B-class. Shared bill.

Value if hit

$300.0B

Inference independence for cars, bots, sats

Value if miss

−$80.0B

Both remain Nvidia/TSMC tenants; merge case weakens

$ / remaining point

$598M

Remaining cost divided by remaining %. A 90%-done cheap item does not look like a 15%-done civilization-scale item.

Hit $300.0B · Miss −$80.0B

Engineering gates

  • Process node choice
  • Advanced packaging
  • Talent
  • Yield

Regulatory gates

  • CHIPS / CFIUS
  • Export controls
  • State incentives