Listen to the audio version of this article (generated by AI).
In May 2025, biologists in Canada and Japan found a genetic sequence buried inside an ocean plankton sample that, by most expert metrics, shouldn’t exist.
It belonged to an organism they named Sukunaarchaeum mirabile — a genome less than half the size of the smallest archaeal genome ever recorded. And yet inside that tiny package, it still builds its own ribosomes and its own messenger RNA, the core machinery every living thing needs to make protein.
It just can’t survive alone. It depends entirely on a host cell for energy and nearly every other function of life. Over time, it kept exactly what it absolutely had to build itself, and outsourced everything else.
Scientists called it a biological paradox: part virus, part cell, wholly unique.
It isn’t a stretch to see Elon Musk in the same light… just with rockets and lithium instead of ribosomes.
The Most Compulsive In-Sourcer in Modern Industry
Musk may be the most aggressive vertical integrator in the history of capitalism. Give him a bill of materials, and he is more likely to fire the vendor and build the thing himself than negotiate the price.
Over twenty years, that instinct has swallowed batteries, motors, chips, software, seats, castings, rocket engines, avionics, launch towers, and even entire data centers.
At Tesla Inc. (TSLA), Musk didn’t just assemble cars — he backward-integrated the car itself. Tesla builds its own 4680 battery cells, and by late 2025 the in-house 4680 had become its lowest-cost cell per kWh, undercutting external suppliers. It runs its own dry-electrode process, its own cathode plant, its own LFP line, and a $1 billion lithium refinery near Corpus Christi, Texas. It designed its own FSD inference chip and the Dojo D1 training chip. It replaced roughly 120 stamped-and-welded rear-body parts with a single casting. It writes its own software, sells direct, and services its own cars. Musk has called the factory itself the product: “the machine that builds the machine.”
At Space Exploration Technologies Corp. (SPCX), the in-sourcing runs even deeper. The company manufactures an estimated 80–90% of its rockets in-house — engines, avionics, flight software, satellite hardware — versus legacy competitors that leaned on more than 1,200 outside suppliers. It builds its own Raptor engines, casts them from its own in-house Inconel superalloys, and 3D-prints turbopumps and injectors. Starlink terminals are built in-house at more than 20,000 a day.
At xAI, Musk turned an abandoned Electrolux factory in Memphis into “Colossus” — a 100,000-GPU cluster built in 122 days and since doubled to 200,000, its power swings smoothed by Musk’s own Tesla Megapacks.
The endgame is a roughly $25 billion Tesla-SpaceX-xAI venture, nicknamed Terafab, to fabricate chips in-house.
He in-sources compulsively. Which is exactly why what he doesn’t build is worth studying.
The Elon Test
Think of every component in Musk’s empire as a contestant walking into a coliseum, with Musk’s engineers as the lions. Twenty years of history says most contestants get eaten — Musk brings their function in-house the moment doing so is cheaper, faster, or more controllable.
The suppliers still standing aren’t standing because Musk is loyal. They’re standing because he tried to build what they build, and failed.
That’s the Elon Test. Surviving it is the signal worth listening for.
What survives tends to correspond to a few unforgiving realities: the physics of extreme-ultraviolet lithography, decades of accumulated process knowledge, brutal capital intensity, and ecosystem lock-in. These are moats you cannot dig your way across in a few quarters, no matter how many engineers you throw at them.
Here are the companies that have passed it.
The Unkillable Core: Foundries and Fabrication
Musk designs chips. He does not fabricate them, because he can’t.
Every custom Tesla chip, including the Dojo D1, was manufactured by Taiwan Semiconductor Manufacturing Co. (TSM). The D1 is a 50-billion-transistor die built on TSMC’s 7nm process, delivering 362 TFLOPS per tile.
Even Terafab — the great in-sourcing dream — proves the point. Intel announced on April 7, 2026 that it had joined the roughly $25 billion venture, targeting one terawatt of AI compute per year at Giga Texas. Intel’s own CEO framed it as needing Musk’s help, not the other way around. Per SpaceX’s own IPO filing, radiation-hardened, leading-edge chips require “fabrication expertise that Tesla simply does not have.”
Behind TSMC sits an even purer monopoly: ASML Holding N.V. (ASML), the only company on Earth that makes the extreme-ultraviolet lithography machines required for advanced nodes. Musk cannot build a leading-edge chip without TSMC, and TSMC cannot build one without ASML.
Two choke points. No substitutes.
The Chip Makers Even Musk Buys From
Musk is one of the most aggressive custom-silicon builders in technology — Dojo, FSD chips, AI5, AI6 — and yet when he needed to train Grok, he bought roughly 200,000 Nvidia GPUs for Colossus. When SpaceX unveiled its orbital data-center satellites, Musk named Nvidia Corp. (NVDA) as the supplier for the first wave.
The moat isn’t the silicon. It’s CUDA — the software layer the AI research world has built on for nearly two decades. Switching means rewriting years of code. So Musk, of all people, keeps reaching for his checkbook instead.
The Narrowing Moat: Battery Cells
Tesla makes its own 4680 cells. And yet its best-selling Model 3 and Model Y still run on cells from outside suppliers, and Tesla’s own finance chief has called batteries the company’s “biggest constraint globally.”
In July 2025, one of those suppliers announced a multibillion-dollar deal to supply LFP battery cells starting in 2027 — with U.S. regulatory filings in March 2026 confirming Tesla as the buyer, even as Tesla races to replace exactly that kind of contract with its own cells.
We’d flag this as the weakest moat on the list. These suppliers survive because Tesla needs volume today, not because Tesla can’t eventually build the product itself.
The Specialist Survivors: Power Chips and Precision Casting
The Tesla Model 3 was the first mass-market EV to use a full silicon-carbide traction inverter, and SiC remains brutally hard to manufacture — historically ten times the cost of plain silicon. Even after Tesla said in 2023 it would cut SiC content by 75%, it said it would keep buying die from multiple outside manufacturers.
Then there’s Musk’s most theatrical manufacturing flex: casting a car’s entire rear underbody in one shot. The giant press that does it is built by a single Italian manufacturer, majority-owned by a Hong Kong-listed parent — the only die-caster, by Musk’s own account, willing to take the risk of building a machine to his spec.
Where the Elon Test Can Fool You
We want to be direct about the limits of this framework, because it’s easy to over-extend.
Surviving the Elon Test today doesn’t mean surviving it forever. Tesla’s in-house battery cells were the weakest link on this list for a reason: a moat built on “Musk needs volume right now” erodes as his own production scales. The names that look unkillable today are exactly the names a founder like Musk tends to fixate on next.
And these are still real businesses with real customers outside Musk’s empire — TSMC, ASML, and Nvidia would matter to the world even if Musk vanished tomorrow. Their Musk exposure is a layer on top of already-large franchises, not the whole story. Treat position sizing accordingly.
The Bottom Line
Zoom out, and a portfolio assembles itself — not the flashy Musk-empire tickers, but the companies selling him what he’s tried and failed to build himself: the foundries and lithography behind every leading-edge chip, the GPU maker whose software moat Musk can’t route around, and the specialists holding onto the manufacturing steps that resist in-sourcing.
He is the most demanding customer in the world, and the most motivated to fire any of them. When he keeps paying anyway, that’s a moat no amount of Wall Street trickery can match.
We’re going to walk through this full supply-chain map — every name we believe has passed the Elon Test, and where we think the next one is being written — at a free workshop on , alongside our colleagues Louis Navellier and Eric Fry. We’ll also give away the name and ticker of one company from the research completely free.
.
