🥛 The most dangerous bet in AI ⚠️

GM. This is Milk Road AI, where we track who's funding, building, and owning the AI stack.

Here’s what we’ve got for you today:

✍️ The most insane factory ever built.
🎙️ The Milk Road AI Show: AI Changed More in 3 Months Than the Last 3 Years... ARK Invest Explains w/ Frank Downing.
🍪 OpenAI kills Sora just months in.

Shareland enables users to trade real estate like stocks with direct access to US housing markets. Trade real estate just like stocks with Shareland.

Prices as of 10:00 a.m. ET.

MUSK IS REBUILDING THE CHIP INDUSTRY

In 1894, Andrew Carnegie's own partner had to drag him into the most important deal of his life.

He thought the iron ore in Minnesota’s Mesabi Range was worthless, but his partner pushed him to buy in anyway.

It turned out to be the richest iron deposit in North America, and from there, Carnegie didn’t just build a business, he built a trap.

He owned the mines, the Pennsylvania coalfields, and even built his own railroad and locked in a 999-year lease to move ore from the Great Lakes straight into his mills.

By 1900, Carnegie Steel was producing more steel than the entire nation of Great Britain.

His first ton of steel cost $56 to make, but he drove it down to $11.50 by owning every step from the ground to the finished beam.

J.P. Morgan watched Carnegie start to squeeze the entire industrial system and realized something fast, it was cheaper to buy him than fight him.

That deal created U.S. Steel, the first billion-dollar corporation in American history.

Carnegie didn’t win by being cheaper, he won by owning the entire system.

Source: Giphy

And now, Elon Musk is trying to replicate that exact model, just on a completely different scale.

But instead of steel, it’s semiconductors, and Terafab is the factory he’s betting everything on.

Welcome to Terafab.

A joint Tesla-SpaceX-xAI chip mega-facility targeting the most advanced chip technology on the planet, producing 1 terawatt of AI compute annually, with logic chips, memory chips, and lithography masks all under one roof.

The entire semiconductor industry heard that and collectively spat out their coffee because here's the thing, that's not how any of this works.

The chip industry is one of the most hyper-specialized ecosystems ever built.

TSMC dominates advanced logic, Samsung has perfected the logic–memory split, and Intel spent decades trying to keep up and is still paying the price.

Nobody does all of it, and no newcomer has ever successfully built a competitive leading-edge fab in modern semiconductor history.

Not Amazon, Google, or Microsoft.

All three of them are worth trillions of dollars, all three of them desperately need custom chips, and all three of them looked at this exact problem and said:

"Yeah, we'll just keep paying TSMC."

So either Musk has lost his mind, or he found the same glitch in the matrix that Carnegie did.

And to be fair, if anyone is built for this, it's him.

This is the guy who was told EVs were a dead end and built the most valuable car company on earth.

Who was told reusable rockets were impossible and now launches 90% of the world's mass to orbit.

Source: @XFreeze

Who looked at the U.S. energy grid and said too slow and built a gigawatt-scale battery business from scratch.

The Carnegie play (owning the supply chain)

Here's the problem Musk is actually solving.

His companies have a compute demand problem that is, frankly, delusional in size.

Global AI compute production is roughly 20 gigawatts per year, while Tesla, SpaceX, and xAI's projected demand combined? 1 terawatt.

Source: @SawyerMerritt

That means the entire planet's chip-making capacity can deliver roughly about 2% of what Musk says he needs.

He can’t just call TSMC or write a bigger check to Nvidia, the supply chain simply doesn’t exist at the scale he needs, so he's building it himself.

And unlike Meta, Microsoft, or Google, which are all building data centers and renting from the same foundries, Musk is going full Carnegie.

Design, manufacture, package, and test the chips, all in one loop, all inside a single facility.

He calls it "rapid, recursive improvement".

Traditional chip development isn’t slowed by design, it’s slowed by handoffs, tape-outs, packaging partners, and foundry slots booked months in advance.

Musk's pitch is to eliminate the handoffs, eliminate the lag, and iterate ten times faster than anyone else.

If that works even a little, it's a competitive moat nobody else can buy but then he goes a step further.

AI-POWERED REAL ESTATE AGENT

Wouldn’t it be cool if you could buy a slice of a New York home in the morning and then sell it in the evening?

That’s exactly what Shareland is building.

Tycoon is Shareland’s AI-powered real estate agent that trades real estate just like stocks.

Here are their key USPs:

Trade real estate without owning the property
Use their AI-powered agent to inform your investment strategy
Tracks housing markets at both the city and neighbourhood level

And you don’t need large capital for this, you can start with as little as $1.

Trade real estate just like stocks with Shareland.

MUSK IS REBUILDING THE CHIP INDUSTRY (P2)

Musk is building three chips, two make sense, one sounds like science fiction.

The AI5 is the boring one (in a good way).

It runs at 150 watts, does the same computing as Nvidia's H100 at 700 watts, a 4.7x efficiency gain, and powers Full Self-Driving and the Optimus robots.

TSMC and Samsung are already making it.

The AI6 is the next generation, roughly 2x the AI5's performance.

It's targeting the humanoid robot market, which is important because robots will need 10 to 100 times more chips than cars, and that's the volume multiplier that makes Terafab's scale ambitions actually coherent.

Then there's the D3.

The D3 is a space-hardened, radiation-tolerant, high-power inference chip designed to run hot specifically to minimize the size of the cooling radiators it needs in orbit.

Source: @SawyerMerritt

Because 80% of Terafab’s output isn’t for cars, phones, or data centers, it’s for satellites.

SpaceX has filed with the FCC for a constellation of up to 1M satellites in low Earth orbit, each one functioning as an AI compute node.

Solar panels in space generate 5x more energy than on Earth (no atmosphere, no weather, no night cycle), and space is naturally cooled to near absolute zero.

Musk's claim is that orbital data centers will be cheaper than terrestrial ones within two to three years.

Deutsche Bank thinks more like the 2030s.

But still, no foundry on Earth is building radiation-hardened inference chips at scale, making it a completely unserved market with SpaceX’s satellite constellation as the captive demand.

This is where all the Carnegie pieces lock together.

Tesla handles solar energy (targeting 100 GW/year manufacturing capacity).

SpaceX handles the launch, Terafab handles the chips.

And xAI, now a SpaceX subsidiary, with Tesla having invested $2B is the compute demand engine running the whole thing.

He's building the vertical integration stack for a space-based AI civilization.

The ASML problem

There is one company that decides whether Terafab really happens or not.

ASML.

ASML is the sole provider of EUV lithography systems on the planet.

Source: MachinePorn/Reddit

Without EUV, there’s no 2nm chip production, and without that, Terafab is just a very expensive empty building.

ASML's order book is already full through 2028.

TSMC, Samsung, and Intel are ahead of Tesla in line, and Tesla has no existing relationship with ASML as of now.

And each machine costs roughly $350M, with an 18-month lead time from order to delivery.

If Musk places orders tomorrow, he's not getting machines until late 2027 at the earliest.

This is the chokepoint, everything else, from the vision to the satellites to the moon cannons, depends on whether ASML ships Tesla the machines.

So how do you actually play this?

If you want to see exactly how we’re positioning around plays like this, including the stocks we’re actually buying, you can try Milk Road PRO for $1 for 14 days.

Today’s the last day to grab this offer.

Here’s what you’re missing out on, I initiated a buy on AAOI in our Discord on Monday, and it’s already up over 20% in just two days.

Source: Discord & Google

Come join us, this will be the best dollar you spend.

Back to the story, this is where it gets insane.

After Terafab enables terawatt-scale compute, the next step is petawatt-scale, 1,000 times larger.

To get there, he wants to build electromagnetic mass drivers on the surface of the Moon.

Source: Phys

Basically, giant magnetic cannons that accelerate AI satellites to escape velocity, launching them directly into orbit without chemical rockets.

The Moon has no atmosphere (no air resistance), one-sixth of Earth's gravity, uninterrupted solar power, and launch opportunities every ten days.

The vision is Optimus robots manufacturing AI satellites on the Moon, loading them onto mass drivers, and launching them into orbit on a continuous loop.

SpaceX delivers the initial hardware via Starship, then robots handle everything after that.

Critics note that reaching petawatt power would require launching over one million tons of material, roughly 135 Starship launches per day.

The government just handed Musk a co-signer

Here's the part of the Terafab story nobody is talking about.

While Musk was unveiling moon cannons in Austin, NASA was quietly reshaping the Artemis lunar program, and the overlap with Musk's vision is almost too convenient to ignore.

Artemis II is on the launchpad at Kennedy Space Center, with four astronauts set to fly around the Moon on April 1, the first crewed lunar mission in over 50 years.

Artemis IV targets the first lunar surface landing in early 2028, and from 2028 onward, NASA begins building a permanent Moon base and one surface mission per year, every year after that.

The new NASA Administrator is Jared Isaacman, a former SpaceX Polaris commander who is actively accelerating the cadence and pushing Apollo-style step-by-step buildup.

So let's be clear about what's actually happening here.

NASA is building the infrastructure, SpaceX is the primary lunar lander via Starship HLS, and Musk is planning to layer a private industrial empire on top of it.

The U.S. government is essentially co-signing Musk's vision, just without realizing it yet or was this their plan all along?

Either way, the trap is already being built.

Carnegie didn't announce U.S. Steel before he owned the mines. Musk announced Terafab before he owned a single machine.

The difference is that the government is already laying the track.

Alright, that's it for this edition of Milk Road AI. We want to hear from you.

Is Terafab real or the world's most expensive PowerPoint?

Hackers are using prompt injections to break into crypto users accounts. Don’t be a victim! Use private, safe and encrypted chat Okara. Use the code MILKROAD to get a 20% discount.

Summ (formerly Crypto Tax Calculator) is a tax software built specifically for crypto. Get started for free with Summ.

Nexo is back in the U.S. - and new clients get 30 days of Wealth Club Premier perks! Higher yields, lower borrowing rates, and crypto cashback - start here.