Interview with Elon Musk, uploaded November 6, 2026

Terafab: A Chip Plant Sized to Today's Entire Industry

The most consequential new detail in this wide-ranging interview is the scale and logic of Terafab, the chip manufacturing project Musk has been teasing. He said the name carries three meanings: "terror," "terra," and "tera" as in terawatt. The goal is 1 terawatt a year of AI chip production. By his estimate the world will produce around 20 GW to 25 GW of AI chips this year and perhaps 35 GW to 45 GW next year. A terawatt a year therefore implies roughly 30 times projected 2027 global output, from a single complex. Musk said the project "will cost a trillion dollars, probably more than that over time," but that the capital will not be needed up front because the build will proceed in phases.

Musk was candid about the risk: "If it fails it'll be crazy, and if it succeeds it'll be audacious." He added that he expects it to succeed "at least to some degree." He also corrected the interviewer's framing that nobody else is building fabs in the US. TSMC is building a large cluster in Arizona, and Samsung has operating fabs in Texas, including one that is or will be dedicated to Tesla's AI5 and AI6 designs. Even on optimistic assumptions, he argued, announced US capacity is only about 10% of what exists in Asia. His concern is geopolitical: "If we don't have all the pieces necessary to build chips at scale in America, then we may not have any chips at all or very few."

The technical thesis rests on fab inefficiency. Musk said wafers spend more than half their time in transit or waiting, and that the time during which atoms on a wafer are being changed in a useful way is "maybe 20% of the time." He blames an industry culture of repeating the last recipe with minor changes, because each new fab is so capital-intensive. Tesla is building a research lab at Gigafactory Texas in Austin to try many approaches, "many of which won't work." Proven improvements would then be transferred to the production site in Grimes County, Texas, near Texas A&M. Musk floated the possibility of putting logic, memory, packaging and lithography under one roof, which he said could cut a chip design iteration from a year to a month. The interviewer said Intel CEO Lip-Bu Tan is expected to have an interest in Terafab. Musk did not comment on that. Investors should note that Musk described the lithography and full-stack integration as something he wonders whether Tesla "will have the ability" to do, which is far from a settled capability.

SpaceX Revenue: A Steep Ramp, With Management Hedging the Upside

Musk gave some of the most specific SpaceX financial commentary to date. He said the company is tracking to an annualized revenue rate "in excess of $100 billion in December," and that barring a "natural disaster or something," it will meet that target. He said Q2 to Q3 revenue growth this year was "over 80%," with roughly 70% sequential growth likely in Q4 and something similar in Q1. The interviewer put last year's revenue at $15 billion to $16 billion.

For next year, Musk said it is "quite likely" that annual recurring revenue in December 2027 exceeds $500 billion, though he stressed this is "not a promise" but his best guess. He said a $1 trillion figure, which the interviewer had floated, has moved from impossible to roughly a 2% probability. That is a useful calibration against the headline numbers in circulation.

The driver is compute. The interviewer, whose figures Musk did not confirm, argued that traditional hyperscale data centers take about 2.5 years and cost roughly $60 billion per gigawatt. He said SpaceX builds in 8 to 9 months at around $35 billion and can lease capacity for about $50 billion a year until it needs the compute for Grok. Musk pushed back on the "we got them for free" characterization with a noncommittal "sort of." Neither the lease economics nor the cost per gigawatt was verified by Musk. Musk framed the opportunity as an execution advantage. He used the "Jose meme" of one person digging while eight watch, saying the US has seen a surge in people issuing purchase orders but not in people doing physical work, leaving "a large number of partially completed data center projects." SpaceX and Tesla, he said, are "probably the two best companies at physical hardware on earth outside of China." Musk said the Memphis power plant produces over 1 GW and will reach 1.5 GW to 2 GW, built in about a year versus roughly five normally.

Orbital Compute and Starship: The Timeline Is Now Specific

Musk said satellites carrying an Nvidia GPU and a Google TPU are operating in orbit, "working fine to the best of my knowledge." SpaceX expects to launch its first orbital compute satellites, which he said were co-designed with Nvidia, in the third quarter of 2027. Volumes would start slowly, reach significant levels in 2028, and become "very, very large" in 2029. He caveated the quarter with "I'm just guessing here."

The argument for space is power. US average consumption is just over 500 GW, with peak capacity of roughly 1.2 TW, so adding 200 GW of AI capacity a year would be a 40% annual increase in the US grid, which Musk called "simply not feasible." Solar in orbit yields about 30% more power, with no seasonality and minimal batteries. Musk said Starship, as the first fully reusable rocket, can cut launch costs by a factor of 100 or more, and that within two to three years a ton to orbit could cost less than air freight from California to Australia. The interviewer cited a target under $100 per kilogram, versus $18,500 per kilogram in NASA's 2010 figure.

On reusability, Musk said Falcon 9 is about 80% reusable but discards an upper stage that costs roughly "a medium-size jet." Starship is designed for catches by the tower arms, which in principle allows booster reflight about once an hour. The harder problem is the ship. "No one has ever made a truly reusable orbital heat shield," he said, calling it one of the hardest technical problems humans have attempted. Musk said Ship 42 will attempt a tower catch in about a week, after several precise soft landings at sea. He acknowledged that "success is not always guaranteed, but excitement certainly is."

Starlink: Agentic Traffic Expands the Bull Case

Musk said about 11,000 Starlink satellites are in orbit, mostly V2, with V3 now beginning to launch. Each V3 has over 10 times the capability of a V2, and SpaceX will launch 10 times as many, implying about 100 times today's bandwidth in two to three years. Laser links are moving from about 100 Gbps to 10 Tbps. He said Starlink revenue is "roughly doubling per year at large scale." Musk cited Cloudflare data that agentic traffic already exceeds human traffic and is forecast to be 1,000 times larger in five years, a forecast he thinks is conservative. The interviewer argued the constellation could grow from 10,000-20,000 satellites to 100,000, and that Starlink alone could be a trillion-dollar revenue business. Musk also said SpaceX could acquire a legacy telecom operator if it can use the spectrum "10 times more effectively."

On compute value, Musk said 100 GW a year sent to space at $10 billion per gigawatt of annual revenue would add $1 trillion of revenue per year, and satellites accumulate because "they don't vaporize."

Macro Claims: Superintelligence, Robots and China

Musk estimated that digital superintelligence alone could lift global GDP by 20% to 30%, roughly $25 trillion, mostly by doubling white-collar productivity. For robotics, he reasoned that a humanoid working 168 hours a week is at least five times as productive as a person, so 20 million to 30 million robots could match US goods and services output. He said robot unit cost would initially be similar to a car and later much lower, and that renting rather than selling is the likely model. The interviewer cited $30,000 trending to $20,000.

On China, Musk said it has about three times US power output and is trending toward four, and that industrial capacity roughly tracks electricity. Chips are where China lags, "for now." His conclusion is that the only way for the US to remain competitive is to take intelligence to space: if SpaceX reaches a terawatt a year, "unless China can do a terawatt a year to space, we will have an economic output in excess of China." He said a terawatt a year would be twice US average power usage added annually. He also said the Model Y is China's best-selling EV, and recounted how Tesla won 100% ownership of its Chinese factory after years of negotiation.

Execution Culture, Local Opposition and Musk's Track Record

Musk addressed data center opposition, saying Memphis is "booming," with full restaurants and hotels, wages at times double the prevailing level, and large tax revenues. He argued unpopularity is concentrated where data centers do not yet exist. He also described walking the Memphis site several times a day and personally pulling cables and doing fiber and switch work, "so that I'm not numb to the pain of the people who are actually doing it."

On timelines, Musk conceded he is "sometimes a little optimistic," but said he aims for predictions at the 50th percentile, "half wrong, half right," and that press coverage ignores when he is early. That is a fair self-assessment for investors to apply to the Q3 2027 orbital compute launch, the $500 billion ARR guess and the Terafab ramp. He closed with a long-range view that superintelligence will drive "universal high income," and that today's money will look like "exchanging seashells."