The bull case treats Terafab as a customer for Intel. I think the bigger story is that Terafab is buying something else from Intel: its people. To build a leading-edge fab from scratch, Musk needs the small group of engineers and managers who have done it before, and in America more of them work for Intel than anywhere else. If Terafab gets them, Intel keeps a licence fee and loses the capability behind it. The stock is still priced for the partner.
The deal Intel thought it signed
Musk launched Terafab on 21 March as a joint venture between Tesla, SpaceX and xAI. In April Intel joined, promising to design, fabricate and package Terafab’s chips at scale, and Musk later said the fab would run on Intel’s 14A process. That was the partner story, and it helped carry Intel to an all-time high of $142.35 on 30 June.
Then the scope shrank. Musk now says Tesla and SpaceX will build and run the fab themselves. That turns Intel from operator into process licensor. And it raises the obvious question: if you license the recipe, who does the cooking? The answer is whoever you can hire.
The first hire tells you the plan
In June, Tesla hired Gary Jiang as “Director, Terafab” in Austin, the project’s first named leadership hire. Jiang spent nearly 18 years at Intel. Since December 2024 he had been the factory manager responsible for moving 18A from Intel’s development fab in Oregon to high-volume production in Arizona: construction, tool installation, startup. Before that he ran ramps for Intel’s 22nm, 14nm and 10nm-class nodes.
Electrek summed it up neatly: Tesla is “acquiring fab competence one Intel veteran at a time.”
One name isn’t an exodus, and some analysts read the hire as good news for Intel. A licensee staffed by people who know Intel’s process is a more competent licensee. Fair enough. But look at what surrounds the hire.
The machine behind it
Tesla’s job postings are running roughly 50% above their 2025 average, and Musk has made a public call for chip engineers. Listings include lithography process engineers and programme managers for capital projects worth over $100m. Tesla is also recruiting in Taiwan for sub-7nm process experience, with reports of offers up to triple local pay. A Korean engineer has made similar, unconfirmed claims about recruiting at Samsung.
So Terafab is fishing in every leading-edge pond. But for a 14A fab in Texas, the most valuable catch is obvious. An Intel engineer in Oregon or Arizona already knows the process, needs no visa and doesn’t have to leave the country. A TSMC engineer in Hsinchu has to learn a new process and move continents. Intel’s workforce is the cheapest, fastest hire Terafab can make.
Why licensing makes it worse
A process licence is only partly a document. The rest is tacit knowledge: tool recipes, yield-learning habits, the playbook for what to do when a ramp goes sideways. That knowledge lives in people. Each time one of them moves to Terafab, the know-how moves too, and Intel’s leverage in the next negotiation falls.
The timing is ugly. Intel is trying to get 14A into risk production next year, and 14A is the node it needs outside customers for. It is also the node Terafab plans to license. The engineers who would ramp 14A for Intel are exactly the ones Terafab needs to ramp 14A for itself. In effect, Intel risks training its customer’s workforce on its own payroll.
Pay makes it harder still. Intel is competing for senior staff against Tesla and SpaceX equity, with a far higher ceiling and a founder who doesn’t lose bidding wars gracefully.
The national question
Intel is one of only three companies in the world producing 2nm-class logic, and the only American one. Washington has put CHIPS Act money into it and taken an equity stake, on the theory that the US needs a domestic leading-edge champion.
The generous reading of Terafab is that competition spreads know-how. Taiwan’s cluster works partly because engineers move between TSMC, its suppliers and its customers. The less generous reading is that the pool of people who can bring up a sub-2nm node is tiny, and splitting it between two American projects leaves neither fully staffed. Over the next three years, I lean towards the second.
The numbers
Q3 was decent on the top line: revenue of $16.5bn, up 21%, with data centre and AI up 78%. But Intel Foundry lost $1.9bn, and only $411m of its $6.7bn in revenue came from outside customers. In mid-October the stock still traded near 60 times forward earnings.
That multiple assumes the foundry becomes a real external business. A foundry is worth its process and its people. If the process is licensed out and the people follow it, what’s left is a very good CPU company priced like an AI platform.
Disclaimer: All views expressed are my own and are provided solely for informational and educational purposes. This is not investment, legal, tax, or accounting advice, nor a recommendation to buy or sell any security. While I aim for accuracy, I cannot guarantee completeness or timeliness of information. The strategies and securities discussed may not suit every investor; past performance does not predict future results, and all investments carry risk, including loss of principal.
I may hold, or have held, positions in any mentioned securities. Opinions herein are subject to change without notice. This material reflects my personal views and does not represent those of any employer or affiliated organization. Please conduct your own research and consult a licensed professional before making any investment decisions.
The Gary Jiang hire is the detail that stuck with me, you don't poach the guy who took 18A into volume if you plan on staying a polite customer.