Musk accelerates “superintelligence” ambitions: SpaceX (SPCX.US) and Tesla (TSLA.US) discuss a $3.5 trillion combination, Asmack accelerates delivery for Terafab

The Zhitong Finance App learned that after the US stock market on Tuesday, during the SpaceX (SPCX.US) third-quarter earnings conference call, Musk unraveled the two clues that had originally been circulating. The two messages point to the same thing: in the AI computing power chain, Musk is simultaneously solving the two problems of “can't create” and “can't fit.”

One is a chip: He said ASML.US (ASML.US) and other key equipment suppliers are already speeding up equipment delivery for Tesla (TSLA.US) and SpaceX's Terafab projects. “These dates are very aggressive, and we are very grateful to Asmack and all of our other suppliers for speeding up equipment delivery to achieve this result.” The other is capital operations: SpaceX and Tesla have each formed a special committee of independent directors to explore the possibility of a merger — the most complete public statement on this potential deal so far.

Terafab's timeline: Test release in May 2027, mass production in the 3rd quarter of 2028

Musk gave Terafab's first batch of specific internal nodes and took the initiative to add a qualifier — “The margin of error is huge.” According to him, the goal of the research fab in Austin is to run the first test chip in May or June next year. The chip “resembles an SRAM chip” and is used to verify the production line; after that, the goal is to achieve mass production in the third quarter of 2028 at Gibbons Creek in the main factory area.

Only by putting these two nodes back into the project itself can we see their weight. Terafab was officially announced in Austin on March 21, 2026, and is being jointly promoted by SpaceX, Tesla, and xAI. The goal is to produce 1 terawatt (1 TW) of AI computing power per year — the reason given by SpaceX in the project announcement is that “the combined chip demand of SpaceX and Tesla is expected to exceed 1 terawatt of computing power, which is significantly greater than the current global supply.” The plant is located at the site of the former Gibbons Creek coal-fired power plant in Grimes County, Texas, covering an area of more than 6,000 acres, about 90 miles northwest of Houston; a groundbreaking ceremony was held on October 6. Land clean-up began in August, and official construction is scheduled for December 1, and all will be completed by the end of 2028.

The investment scale is also hierarchical: the first phase of investment of $16.8 billion was confirmed by SpaceX's announcement on August 6 and Texas Governor Abbott; according to the incentive application documents submitted to Texas, the total value of the project after completion could reach up to US$119 billion. The Texas Enterprise Fund (Texas Enterprise Fund) provided $30 million in grants for the first instalment, and the County Board of Commissioners passed 100% property tax relief 4-1 in June.

The real manufacturing capacity is not in Musk's hands. Intel (INTC.US) joined the project on April 7, promising to cover the full vertical chain of design, manufacturing, and packaging. Terafab mass production will use the Intel 14A process; on October 2, it was reported in the market that TSM.US (TSM.US) may participate, and on October 7, Musk clearly delineated on X — “We will build and operate this factory”, and TSMC can only sublease at most a portion. Adrien Sánchez, an analyst at research firm Yole Group, sums it up more bluntly: Terafab “is more like an expansion of Intel's fab; Tesla, SpaceX, and xAI are anchor customers.”

Asmai's role: the only bottleneck and the most direct beneficiary

Thanks to Musk, the landing point is a company whose production capacity has reached 2030.

Asmack's third-quarter results announced on October 14 were a record: net sales of 11.662 billion euros, up 25.0% month-on-month and 55.2% year-on-year, but slightly lower than Wall Street's average forecast of 11.735 billion euros; full-year net sales guidance was raised to 44-45 billion euros, and gross margin guidance was raised to about 56%. CEO Christophe Fouquet (Christophe Fouquet) confirmed at the time that he had direct communication with Musk about Terafab, describing the other party as “very serious (very serious)”, and at the same time gave the premise that the project could only be called an opportunity if Asmack itself did not become a supply bottleneck, and the company has yet to do this with existing customers.

However, the equipment may be in place earlier than in the factory. According to chip industry watcher Tim Culpan, citing people familiar with the matter, Terafab's first ASML EUV lithography machine may be delivered in the middle of next year; while SemiAnalysis estimates show that the main structure of the Austin research fab will not be completed until the end of 2027, and the equipment will be moved even later, which means this expensive stand-alone device will be idle for quite some time. This is a heavy financial burden for traditional chip makers, but not for Terafab.

Why don't you have to make your own? A financial report incorporating AI into the main business

Terafab's sense of urgency can be quantified in the same night's earnings report.

SpaceX's third-quarter revenue was US$14.575 billion, up 161% year on year (US$5.576 billion in the same period last year); operating profit of US$2,791 billion and net profit of US$4.404 billion, all of which reversed losses; adjusted EBITDA of US$8.809 billion, up 306% year on year; GAAP diluted earnings per share of $0.35. All three were higher than analysts' expectations aggregated by S&P Global Market Intelligence (revenue of approximately $12.8 billion, net profit of approximately $2 billion, and adjusted EBITDA of approximately $6.8 billion).

Structural changes are more important than total: the AI business, known by the company as “Super Intelligence (Super Intelligence),” had third-quarter revenue of US$7.56 billion, surging 830% year over year. For the first time, it surpassed Starlink as the largest source of revenue, accounting for about 52% of the company's total revenue and achieving US$431 million in operating profit. The driving force is cloud computing power leases — Anthropic agreed in May to pay nearly $45 billion over three years to SpaceX's Colossus data center, and Google signed a monthly payment of $920 million (AFP) in June 2029 in June 2029.

The cost is written in capital expenditure: capital expenditure for the third quarter was 19.593 billion US dollars, of which 16.4 billion US dollars were invested in super smart businesses, totaling about 48 billion US dollars in the first nine months. The computing power scale was 1.7 GW as of September 30, 0.6 GW a year ago; Musk says it will reach 2.5-3 GW by the end of the year, and is certain to increase another 10 GW in 2027, or even “a little bit likely” to 20 GW, and expects SpaceX's annualized recurring revenue (ARR) to reach $500 billion to $1 trillion by the end of 2027.

How aggressive this range is, can be calculated: the revenue of $14.575 billion in the third quarter corresponds to an annualized annualized amount of about $58.4 billion. If we were to reach $500 billion in 15 months, that would mean an ARR increase of at least 8.5 times. If you look at it together with Terafab, the logic is self-consistent — when the slope of computing power demand is steep, the supply chain itself is the biggest constraint, and Musk's solution is to build the factory yourself.

Merger Signals: From “You Can't Talk on the Phone” to the Special Committee

What really took the market's breath away was the second thing.

In July of this year, in response to questions from Wells Fargo analysts about the merger of the two companies, Musk answered, “This kind of matter cannot be discussed on an earnings call; proper procedures must be followed.” On October 20, he presented the procedure itself: “SpaceX and Tesla have both set up special committees at the request of shareholders to discuss the idea, and these special committees are composed of independent directors. Since I'm a stakeholder, I can't actually participate in discussions. It's a bit like a weird kabuki dance (kabuki dance) where two groups are talking and I can't participate in it.”

At the same time, he emphasized three boundaries: the Special Committee has yet to agree on the terms of any deal, and “obviously there is no guarantee” that they will reach it; the deal requires shareholder support, and “Tesla requires 50% of all shares to vote, not just shareholders participating in the voting” — he called this “a fairly high threshold”; “Some people in the outside world think the merger is a foregone conclusion; I think that is a bit exaggerated.”

But he also expressed his judgment unreservedly: “Do I think this is a good idea? Yes.” The reason was a collaborative checklist — super-smart software, chip design, Terafab, robotics, StarMind constellation orbiting capacity, and the “connectivity” that President Schottwell added on the spot. If so, the scale is unprecedented: the two companies have market capitalizations of around $2 trillion and $1.5 trillion, respectively.

The tension at the governance level is even more noteworthy. According to SpaceX's IPO documents, Musk holds more than 82% of voting rights in SpaceX, while Tesla's voting rights are about 20%; Barclays analyst Dan Levy estimates that if SpaceX acquires Tesla on an all-stock basis, Musk will hold about 37% of the economic interests and hold about 56% of voting rights after the merger — a “minority share, majority control” transaction.

The market is already setting prices, but the differences are just as clear

The path given by the seller is already quite specific. Barclays's Dan Levy determined that the deal could be announced before Tesla's annual shareholders' meeting (the company has announced that it is scheduled to be held on December 3). Based on the quoted SpaceX $2.1 trillion and Tesla's valuation of $1.5 trillion, Tesla shareholders can exchange about 0.7 shares of SpaceX per share; if this window is missed, it may still be announced within 6 to 12 months. Wedbush's Dan Ives gave “the probability of completing the merger by the end of 2027 is over 80%.” RBC Capital has included a 25% — 30% SpaceX acquisition premium in Tesla's price target calculation.

The skepticism is equally clear. Gary Black of The Future Fund reminded the bulls to “stop advocating the TSLA-SPCX deal,” believing that the share swap will be priced according to Tesla's lower valuation multiples and that arbitrage will erode the premium paid for SpaceX shares before delivery; Morningstar indicated earlier this year that valuation differences may be significant, especially on Tesla's shareholders' side. Sellers were less divided on target prices but were generally conservative: the 12-month average of $397.83 given by 38 analysts, J.P. Morgan $415, Morgan Stanley $400, Goldman Sachs $360, and UBS $391, mostly neutral or holding ratings.

What the market predicts is another set of numbers: the contract price of “officially announced merger by the end of 2026” on Polymarket is about 23.5% (up 6 points within 24 hours), and about 64.5% “by the end of 2027” - this is the trader's price, not the existence of an agreement.

Ambitious price tag: $48 billion in capital expenditure meets tighter financing environment

Looking at the two lines together, Musk proposed a “balance sheet in exchange for time” plan.

The hard constraints on progress are still clear: SemiAnalysis estimates that the research fab will take 15 months to build and about 6 months to install the equipment before starting production; the Intel 14A process transfer itself requires extensive equipment certification; and ASML's high-NA EUV production capacity will still prioritize satisfying existing customers until 2028. Musk's own qualifier for this batch of schedules was “the margin of error is huge.”

The financial constraints are just as real. SpaceX's capital expenditure for the first nine months was about US$48 billion, and capital expenditure for the fourth quarter was US$35 billion to US$40 billion. Meanwhile, the AI infrastructure financing environment was weakening during the same period: CoreWeave and Nebius shares both fell by about 28% in October, IREN fell 35%, and Australian data center operator Firmus withdrew its IPO plan due to market fluctuations.

Procedural uncertainty is easiest to underestimate. Up to now, the existence of the Special Committee has only come from Musk's oral statement at the results meeting — SpaceX's financial reports are not mentioned in any of Tesla's documents, nor have the two companies announced committee members or any deal plans (Drive Tesla Canada). Additionally, the deal is subject to two shareholder votes, an antitrust review, and a national security review that could be triggered by a defense contractor absorbing a car company with deep Chinese exposure.

In other words, the signals Musk gave off this night were clear and consistent — equipment was expedited, committees were in place, and computing power targets were written into guidelines; but from “signal” to “fact,” what is in between is the threshold he himself has emphasized over and over again: shareholder voting.