Dear Income traders,

Elon Musk wants to put a chip plant on 100 million square feet of Grimes County, Texas.

Filings with the Texas Comptroller list an initial investment of $55 billion, with later phases carrying the total to $119 billion.

SpaceX and Tesla committed the first $16.8 billion on August 6.

At full build the structure would run about two and a half miles end to end, roughly 10 times the footprint of Giga Texas. Musk calls it the largest and most valuable building on Earth. The project is called Terafab, a joint venture between SpaceX, Tesla, xAI and Intel (INTC), which is contributing its 14A process.

The stated target is one million wafer starts per month at 2nm (a wafer start is one silicon disc entering the line, and 2nm is the current leading edge). TSMC runs about 1.6 million wafers a month across every node it operates, and its entire 2nm line should reach roughly 100,000 a month by December. Musk is describing 10 times TSMC's leading-edge output from a single site.

Now look at the middle of the site plan.

There's a faint circular structure in the rendering SpaceX released. Observers read it as a particle accelerator, which would point to a free electron laser: a different method of producing the extreme ultraviolet light that etches circuits onto silicon.

Today that light comes from ASML Holding (ASML) machines that vaporize tin droplets with a laser, and ASML is the only company on the planet that sells them.

When someone posted that theory on X, Musk replied "FEL FTW."

ASML CEO Christophe Fouquet told Reuters in May that he'd spoken with Musk directly and called him "very serious." A fab at 2nm can't run without ASML tools, so Musk could become the largest single order in the company's history. He's also pointing at the one component that would let him route around it.

He's running versions of that play everywhere. SpaceX collects about $26 billion a year renting compute to Anthropic and Google. It filed with the FCC in January for authority to launch up to one million orbital data center satellites.

It also has permits in for a 10 gigawatt solar cell factory in Bastrop, and Tesla filed this month for a $10.1 billion cell plant near Houston. He holds a position in all seven layers of the stack: compute, chips, models, distribution, power, orbital compute and robotics.

So the question for the rest of us is what sits inside the blast radius and what sits outside it.

Memory looks like the clean answer, and the reason is timing rather than capability. Terafab does plan to make memory. Tesla is already hiring DRAM process integration engineers, but first chips aren't expected until late 2027, with volume production targeted for 2028.

Until then he's a buyer like everybody else. On Tesla's July earnings call, Musk said Micron Technology (MU) had been "making room for Tesla in the years to come."

The supply math is why that line matters. IDC puts 2026 DRAM supply growth at 16 percent and NAND at 17 percent, both below the 20 to 30 percent that used to be normal, while wafer starts grow 6 to 8 percent. Hyperscaler capital spending runs past $600 billion this year, up about 40 percent.

HBM (the stacked memory that feeds AI accelerators) eats roughly three times the silicon per gigabyte of standard DRAM, so every wafer moved to HBM pulls three wafers' worth of DDR5 out of the market. New fabs from Micron and SK Hynix don't reach volume until 2027 at the earliest. Terafab doesn't fix that, it adds one more very large buyer to the queue.

Grid hardware sits outside the radius too. Transformers, switchgear and transmission equipment carry delivery times measured in years, and a solar cell plant in Bastrop does nothing to shorten them.

Don't compete with Elon. Own what he has to buy.

Here for a good time AND a long time,

Hans