AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
Oh sure, SpaceX’s credit risk is pretty high because it’s about to borrow another $40 billion to spend on Nvidia chips, but who cares about that? SpaceX and Tesla have formed special committees to discuss a merger! The degree to which you find this merger compelling likely has to do with the degree to which you believe Elon Musk about wanting to go to Mars, or “spread the light of consciousness,” as he put it in this week’s SpaceX earnings call (and earlier S-1). Now, a reasonably informed person knows that we don’t have the technology — or, frankly, even the raw information to build the technology — in order to support human life in space. So that means robots. But a lag between Earth and the Moon or Mars makes for difficulty in controlling the robots. So having artificially intelligent robots who can make decisions on their own with minimal supervision from Earth is the next sci-fi logic step. If you believe Musk’s story, the technology we are talking about is being developed through self-driving cars (Tesla), LLMs (SpaceX), and humanoid robots (Tesla), all of which require data centers (SpaceX). Conveniently, we have just had earnings for both companies. On the other hand, if you view Mars as a useful fiction Musk uses to explain his antics in public, well, there’s a lot of money and a lot of control Musk gets from the merger. Musk said on the recent SpaceX earnings call that “there’s a lot of merit to combining” SpaceX and Tesla. Why? “There’s a lot to be gained from cooperating on developing superintelligence software.” But artificial intelligence — ah, excuse me, super intelligence software isn’t the business SpaceX is really in right now. In its Q3 earnings on Tuesday, SpaceX reported $7.56 billion in revenue from the artificial intelligence — sorry, sorry, super intelligence division. The subcategory “SI solutions & infrastructure revenues” — which I think is Grok, Cursor, and the neocloud business wrapped together — jumped to $7 billion, or more than three times what it was last quarter. Now some of this was getting Cursor stapled on, which added $608 million in revenue. Most of the rest of that $7 billion was likely renting out compute to Anthropic and Google; Grok is probably contributing less, says Gil Luria, an analyst at DA Davidson. At first blush, this may look comparable to the revenue of a company like Anthropic, whose leaked S-1 shows the company had $16.3 billion in revenue in the first half of this year from its frontier models. But SpaceX’s AI business looks more like CoreWeave than Anthropic. While we are told in the press release there was “increased Grok adoption across enterprise customers and consumers,” that number wasn’t broken out from bare-metal revenue. And if Musk wanted to brag about it, he would. As for Tesla, which released Q3 results on Wednesday, its earnings sucked — a second consecutive quarter of declining profits. That’s because Musk is apparently not especially interested in normal cars anymore. Instead, Tesla is focusing on robots and, I guess, self-driving taxis. (There was also a Roadster hopping car demo recently, but that’s not going into production for a couple years, it’s very expensive, and Tesla cannot guarantee it will be delivered.) Musk did not elaborate on his comments about the merger on the Tesla call. Macroharder SpaceX and Tesla have some tie-ups already; SpaceX, for instance, bought $131 million of the low-selling Cybertrucks from Tesla. It has, in the past, held Tesla bonds. More crucially, the companies have also announced two AI — fuck it, we’re just sticking with AI — projects together: Terafab and Macrohard. Terafab is the chip-making venture that Musk intends to build in Texas and run to supply himself with chips; Intel is partnered, kinda, with Musk to develop the technology. Macrohard is, uh, an AI project that lets people simulate software already made by Microsoft. At least, I think that’s what it’s supposed to be. Typing the name of the latter makes me pray for the asteroid to hit us all right now, so perhaps it’s fortunate that we haven’t heard much about it since the announcement in March. SpaceX’s S-1 from May said that Terafab and Microhard “are in the very early stages, as a result of which we and Tesla have not finalized a variety of details relating to our collaboration, including, but not limited to, financial terms, intellectual property rights, and the ultimate term of our collaboration.” As of earlier this month, Intel’s CEO said the company remained committed to the partnership, but it’s not obligated to remain part of Terafab and the S-1 says there may never be a definitive agreement. Intel’s Thomas Hannaford didn’t immediately respond to my emails asking if anything had been signed in the interim. On this week’s call with SpaceX investors, Musk claimed the Terafab will reach volume production in the second half of 2028. He also said the company will have its first test chip from the facility in May or June 2027. “Musk has a history of setting timelines that his companies have missed,” Reuters notes, with a degree of what I can only describe as wry understatement. When an analyst on the call asked if Musk had placed orders for machines needed to make cutting-edge equipment, Musk did not answer. Companies often are doing more privately than they announce publicly, so I do not view the S-1 filing as fully definitive. But Macrohard isn’t mentioned in the last 10Q. Neither is Terafab. The Matt Levine case for the SpaceX-Tesla merger Musk is going to pound the table about the Moon, Mars, “and beyond,” and a significant chunk of his fan base is very into this. But there are maybe some boring financial reasons shareholders may have an interest. Let’s say you are constitutionally very different from me, and much richer, and for this reason you own a number of shares in Tesla. You’re a big believer in Elon Musk — obviously — but you notice that the man has an attention span of a squirrel on cocaine. He’s on X basically constantly and often racistly. He keeps showing up at the White House to do stuff that pisses off would-be Tesla customers. Then he publicly feuds with the president. Sometimes he’s very excited about rockets, and sometimes he’s very excited about Dogecoin and sometimes he’s really into Grok and sometimes he thinks about cars. But you notice that one way or the other he’s been thinking about cars a lot less lately. Since he seems to be focused — insofar as he can be said to focus — on AI, you notice some pretty big conflicts of interest. Musk’s companies “all need the same scarce Nvidia chips to power their AI, for instance,” notes Matt Levine, a longtime Musk observer. But if you combine the companies, well, maybe you build the best general-purpose AI. And then? “If you build the best general-purpose AI model, maybe it can drive the cars and operate the robots and upload people’s brains and serve right-wing takes on Twitter.” Also, there’s a massive AI infrastructure buildout happening, and so basically all you have to do is say “Nvidia” and someone will give you a fuckton of money. The merger puts all the AI projects in one pile, plus also some cars and rockets and satellites I guess. “‘Give us $100 billion to build data centers in space, but if for some reason we don’t build data centers in space you can repossess our car factories and data centers on earth’ is a good pitch to credit investors,” Levine writes. Musk also has a history — not always successful — of building physical infrastructure. This is, at least from my point of view, a better bet when it comes to building data centers than some Bitcoin miners who pivoted. “If Elon builds at the rate he talked about, he will have lots of extra capacity and be able to sell much of it,” says Luria. From this point of view, obviously SpaceX and Tesla should merge, Levine writes: > It’s crazy that they are separate companies! One guy runs two gigantic physical AI companies with huge overlaps! What a nightmare of conflicts of interest and red tape! Just merge them, so that the one guy can come into the office each day with a simple mandate: Make the choices that maximize overall value for the giant AI complex, and for the shareholders of that complex. If some chips or engineers or drills or whatever should be working on robots, work on the robots; if they should be working on data centers, work on the data centers. The shareholders have come this far; surely they trust Musk to make those decisions. It’s just that right now, working on the robots is good for Tesla and bad for SpaceX; working on the data centers is bad for Tesla and good for SpaceX. Musk can’t put both of their interests first. There’s an easy fix. It’s not exactly that simple, of course. The details of the deal matter; a lot of people are assuming a stock-for-stock merger. (“I assume the acquirer would be SpaceX, in part because it has a bigger market capitalization but mostly because of Musk’s greater control there,” Levine writes.) So does the regulatory environment. The case that SpaceX and Tesla should combine due to AI is also kind of an antitrust reason they shouldn’t? The current regulatory environment is, in large part due to Musk’s meddling via his DOGE cronies, let’s say lax. Donald Trump, the current president, has two more years left in his term. Also, the markets have started getting waves of nervousness around AI. Earlier this summer, there was the Situational Awareness thing, and then there was the more recent CoreWeave thing, and, well, investor sentiment may be changing. That means “there’s a push to announce a deal by the first half of 2027 so that it closes during this administration,” Bloomberg writes, attributing this to “people who are familiar with the matter.” Shares and shares alike Musk has more control at SpaceX than he does at Tesla. His voting power is above 50 percent, because he owns special supervoting shares. He is the CEO, CTO, and chairman of the board. Shareholders are very limited in how they can legally challenge that power. The filing puts it this way: “This will limit or preclude your ability to influence corporate matters and the election of our directors.” At Tesla, however, Musk only has about 20 percent of the vote. (“Only!”) So in a stock-for-stock merger, where SpaceX is the acquirer, Musk suddenly has voting control of the whole enchilada. “He gets voting control of Tesla in the only way he can,” says Ann Lipton, a professor at the University of Colorado Boulder’s law school. “He can’t change it to a dual-class entity now that it’s publicly traded. This is the only way to do it.” Both companies are based in Texas, which makes it very difficult for shareholders to sue, says Lipton. “The chances of a shareholder being able to sue on the Tesla or SpaceX side are vanishingly small because Texas has erected so many barriers to shareholder litigation.” So there’s that risk gone. Leave aside the more complicated issues of national security (for SpaceX) and antitrust. At a federal level, securities laws are largely “Did you disclose it?” notes Lipton. So the SEC, or what remains of it, will want to see that everything is disclosed to shareholders before the vote. As long as the disclosures are accurate, that’s that. At the state level, there may be some issues — it’s a breach of fiduciary duty to create an unfair transaction as a result of conflicts of interest — but Shari Redstone did merge Paramount and CBS, so it’s not impossible. What does Musk get from this merger? Well, control, a thing he historically loves most of all. Potentially also, the combination adds $110 billion to his net worth, according to Bloomberg. More interesting, however, is how the merger interacts with Musk’s massive Tesla compensation package. If there’s a legal change of control — for instance, because SpaceX buys Tesla — the operational milestones in Musk’s enormous pay package fall away. What matters then is Tesla’ [truncated for AI cost control]