Musk Says Terafab Could Make Chips in Volume by Late 2028. He Called It an Aspiration, and the Site Is Still Foundations.

Honest fence (read first): On SpaceX's Oct 20 earnings call, Morgan Stanley's Adam Jonas asked whether SpaceX has ordered EUV chipmaking machines and when they arrive. Musk didn't confirm an order or a delivery date. Instead he gave Terafab its first dated timeline. A simple SRAM test chip should be working at the research fab in Austin "in approximately May or June of next year," and volume production at the main Terafab site at Gibbons Creek in Grimes County, Texas is an "aspiration" for Q3 or Q4 2028. He said he puts "big error bars" on both dates and called them "very aggressive." He also said ASML and other suppliers are "accelerating equipment deliveries." ASML hasn't commented. The outside estimates we cite (Tom's Hardware, using SemiAnalysis) are reports, not company numbers. Embed: Joe Tegtmeyer zkaaB1sAOiw (oEmbed 200), drone footage from Oct 16.

Editorial Thesis

Terafab is the chip half of SpaceXSI's compute story. If SpaceX can make its own "superintelligence chips," it stops waiting in line for everyone else's. Until Oct 20, Terafab had a site, a price tag and a slogan, but no dates. Now it has dates, said out loud with a hedge attached. Dates are useful because they can be checked. The first checkpoint is a working test chip in mid-2027. That's a small, boring part, and it's the right first receipt.

Honest Verdict

Proven: Terafab is a real project with a real site. SpaceX's own Aug 6 post names Grimes County, a roughly $16.8 billion first phase and at least 3,000 jobs, and drone video shows foundations, slabs and trenching as of Oct 16. Promised: a working test chip around May to June 2027 and volume production in late 2028. Musk himself framed the second one as an aspiration. Unproven: the EUV order and delivery dates, the "significant potential breakthroughs in chip fabrication" he wouldn't describe, and whether a brand-new chipmaker can go from foundations to volume in about two years. Outside estimates say it can't.

What Musk said, in order

1. "We're just figuring out chip fabrication... I put big error bars on this because we don't know if we will actually achieve this or not."
2. Test chip: "to make our first sort of test chip functional in approximately May or June of next year. That's just a kind of a SRAM chip something that basically test the line. That's at the research fab in Austin."
3. Volume: "our aspiration is to reach volume production around Q3 of '28 at Terafab at Gibbons Creek." Later: volume production "is at least in a set of possible outcomes for Q3 or Q4 of '28."
4. Suppliers: "very aggressive dates, and we're very appreciative of ASML and all of our other suppliers in accelerating equipment deliveries."
5. The tease: "some significant potential breakthroughs in chip fabrication... Let's not give away the shop here."
6. The goal: Terafab will "make a terawatt of super intelligence chips per year."

A quick translation. An SRAM test chip is the memory-cell test part fabs use to check that a new production line works at all. It isn't an AI processor. "Volume production" is when chips come off the line in useful quantities. A lot of hard work sits between those two milestones.

What SpaceX has put in writing

SpaceX's Aug 6 update ("Breaking Ground on Terafab in Texas") says:

• Terafab combines "logic, memory and advanced packaging all under one roof," with more than 100 million square feet of manufacturing space planned.
• Tesla broke ground on a research fab on the North Campus of Giga Texas in April. That's "the precursor to Terafab," and it's where Musk says the test chip will run.
• The initial phase is estimated at about $16.8 billion from SpaceX and Tesla, with at least 3,000 jobs.
• Chips would go to Optimus robots and Cybercabs (edge and inference chips) and to "SpaceX's space-based data centers" (high-power chips).
• It uses Gibbons Creek Reservoir water rather than groundwater.

The post has no production dates. The Oct 20 call is the first time those were put on the record, and they came as spoken targets, not a filing.

Who runs it: settled two weeks ago

On Oct 7, after talk of TSMC taking a role, Musk posted: "No, we will build and run the fab. Let there be ZERO doubt about that," adding that TSMC could maybe sublease part of it, "but nothing more than that" (Reuters). The same day, Intel CEO Lip-Bu Tan told Bloomberg that Intel, a partner since April, "will remain involved." Musk has said Tesla planned to use Intel's 14A process.

What outsiders expect

• First EUV tool, mid-2027: Tom's Hardware (Oct 14), citing a report by Tim Culpan, says Terafab's first ASML EUV scanner is due in the middle of next year. Its SemiAnalysis-based read is that the development fab's shell won't be finished until late 2027, so the tool could sit idle for a while.
• Operations in late 2029, volume by mid-2030: the same Tom's piece says analysts expect the Grimes fab "will begin operation in late 2029 and then ramped to volume by mid-2030." It cites a SemiAnalysis schedule of 20 months to build, 17 to equip and six to ramp.
• Intel's own 14A timing: Intel plans risk production in the second half of 2027 and high-volume manufacturing in the second half of 2028. Musk's volume "aspiration" lands at the same time Intel itself expects to be in volume on that node.

So Musk's late-2028 volume target runs about a year ahead of the outside estimate for first operation, and roughly a year and a half ahead of its estimate for volume. That gap is the story. If the "breakthroughs" are real, it could close. If not, expect the date to move.

What's actually on the ground

Joe Tegtmeyer's Oct 16 drone flight shows heavy earthwork, debris burn pits, foundations and footings, "large sections of concrete," and trenching with pipe going in on the west end of the site, working east. He says land bought through an LLC he links to SpaceX and Tesla now tops 8,100 buildable acres, or close to 12,000 acres counting the reservoir. Those are his figures from land records; we didn't check them. He also says it isn't yet clear what the first foundation is for. Tom's noted the same about his September footage.

That's real progress for a site announced in August. It's also what a fab looks like two-plus years before it makes chips at scale.

Claim vs proof

What to watch next

• Any ASML or supplier confirmation of Terafab deliveries, or SpaceX filings that mention fab equipment.
• Research-fab progress at Giga Texas, since the test chip depends on that building, not the Grimes site.
• New drone footage showing building shells, not just slabs, in Grimes County.
• Whether the "breakthroughs" ever get described, and whether the 2028 date holds on the next call.

Video note

Joe Tegtmeyer, "Terafab Grows at a Phenomenal Pace of Construction & Expansion! 16 Oct 2026 Terafab Update (9:30 AM)" (zkaaB1sAOiw), uploaded Fri Oct 16, 9:45 AM ET, 26:39, ~110k views. It's an independent drone-photography channel, long known for Giga Texas and Starbase flyovers. The title is enthusiastic and the description carries a Starlink referral link, so use it for what the site looks like, not for dates or claims. It was shot four days before the call. Alt: The Information qoXRKr_p7lw (Oct 7) on the ownership strategy.