The Entire Tech Industry Is About To Be “Musked”

SpaceX Is Going To Dominate Unlike Any Company In History

The world’s tech industry just shit itself.

SpaceX’s (SPCX) quarterly report last night was the most extraordinary earnings call I’ve ever heard. And it signals the beginning of a death march for many of the industry’s existing business models.

Let’s start with communications bandwidth.

SpaceX’s satellite internet service Starlink added 1.6 million subscribers in Q3 – reaching 13.6 million users. That’s still small compared with the big American phone companies – Verizon (VZ) has 147 million wireless connections and AT&T (T) has 110 million. T-Mobile (TMUS) no longer reports its line count, but at the end of June it had 34.7 million postpaid accounts, many of them with several lines each.

As an investor though, you must look at where the “ball” is going.

Starlink had 2.3 million subscribers at the end of 2023. It had 4.4 million at the end of 2024. It had 8.9 million at the end of 2025. It added another 4.7 million in the first nine months of this year.

Starlink’s 1.6 million new customers in Q3 compares with 379,000 new postpaid phone customers at AT&T in the same quarter, and 184,000 at Verizon in Q2. Oh, also, about 12,000 airplanes have signed up for the service too, with Starlink availability becoming a major competitive factor between Delta Air Lines (DAL) (which doesn’t have it) and United Airlines (UAL) (which does).

In a few years… it seems likely that SpaceX will put every other communications network out of business.

Starlink Mobile, which connects ordinary phones directly to satellites, already reaches 22 million users a month. Those users are separate from Starlink’s 13.6 million rooftop-antenna subscribers. Starlink Mobile doesn’t need a rooftop dishy (as it’s called): it works through partner carriers like T-Mobile, which sells it as T-Satellite, and it connects an ordinary phone straight to a satellite wherever there’s no cell tower. Its monthly users grew 37% in Q3 alone.

SpaceX recently agreed to buy about $17 billion of wireless spectrum from EchoStar (ECHO). Starting in early 2028, it plans to offer full mobile carrier service on that spectrum. And it will, undoubtedly, offer a service that’s 10x better than any other carrier for a similar price.

SpaceX’s new V3 satellite has one terabit per second of download capacity. That’s enough for roughly 200,000 simultaneous video streams, per satellite. Starship is designed to carry 60 of these satellites per flight.

In September, Starship Flight 14 deployed 26 terabits per second of bandwidth, globally. That’s about 10 times what a single Falcon 9 launch of SpaceX’s older V2 Mini satellites delivered. It’s more than 8x the capacity of Viasat’s (VSAT) entire new ViaSat-3 fleet (three satellites rated at a little over one terabit each).

At the end of 2025, Starlink’s whole constellation, built since 2019, had about 450 terabits per second of capacity. A fully loaded Starship carrying 60 V3 satellites would add 60 terabits, more than 13% of that total, in one flight.

Next year, Starship will be flying weekly. SpaceX founder Elon Musk says eventually Starship will be flying daily. It will not be long before Musk’s network is larger than the rest of the internet’s total global capacity, combined. For scale, all of the world’s international internet capacity, carried mostly on undersea cables, totaled 2,259 terabits per second in mid-2026, according to the research firm TeleGeography.

But virtually limitless global connectivity is only one part of the enormous computer system that Musk is building.

Intelligence – driven by large-scale compute facilities – is becoming a utility. And Musk is going to own most of it. SpaceX is bringing roughly 220,000 Nvidia (NVDA) GB300 chips online every month, and Musk said that pace will accelerate next year.

Foxconn (2317.TW), the tech contract manufacturer, has disclosed a SpaceX order for about 936,000 of these chips, worth $52 billion, for delivery between now and Q1 2027. Nvidia doesn’t publish its sales by customer. But industry estimates put this year’s total output of its Blackwell chips at 5.5 million to 6 million units. SpaceX is taking something like 40% of production. It is Nvidia’s largest buyer by a wide margin.

SpaceX already owns the largest AI data center on Earth, Colossus 2, in Memphis, Tennessee. But that’s only the beginning.

Here’s the schedule Musk and SpaceX CFO Bret Johnsen laid out on the call yesterday.

SpaceX ended September with 1.7 gigawatts of compute. It expects to reach nearly 2 gigawatts by the end of October and 2.5 to 3 gigawatts by the end of December. It’s turning on about 220,000 GB300s a month now, and Musk said that pace “will obviously accelerate next year,” when most of the new capacity will be Nvidia’s next-generation Vera Rubin chips.

By the end of Q2 2027, SpaceX plans to have what Musk called “the most powerful training cluster” online: 440,000 Vera Rubin computers.

For all of 2027, Musk said he’s “highly confident” of adding 10 gigawatts in a single year.

I would be surprised if we brought on less than 13 gigawatts of compute by the end of next year.

That’s more than seven times what SpaceX had in September (!), added in five quarters. SpaceX is also building 20 gigawatts of new power and cooling next year (!), so power never limits the build.

How are they going to afford that?

SpaceX spent $19.6 billion on equipment and construction in Q3, and CFO Johnsen guided Q4 spending to $35 billion to $40 billion (!). This represents twice as much capital spending as the company told investors to expect earlier this year.

How does that compare with the other hyperscalers? In absolute dollars, only three companies are spending more than SpaceX. Amazon (AMZN) spent $54.2 billion in its June quarter, Microsoft (MSFT) guided to more than $50 billion for September, and Alphabet (GOOGL) spent $44.9 billion. But these are the largest companies in technology.

If you measure this spend against revenue, no other large business is spending more than SpaceX on compute. Amazon is spending about 27 cents of every revenue dollar on capital projects this year. SpaceX will spend 9x that, about $2.50. Among the companies building AI at scale, only CoreWeave (CRWV) spends at a similar ratio. But CoreWeave has no rocket business, no satellite network, and no investment-grade credit rating. That leaves it at a major competitive disadvantage in the race for scale.

Whoever produces the most compute can sell it at the lowest price and still earn a margin. The margin pays for more capacity, and more capacity lowers the cost again. In a business built this way, the largest producer wins. That’s why Musk is ramping up the build-out of SpaceX’s compute systems.

Today, SpaceX doesn’t own the most computing power in total. OpenAI and Anthropic will each end this year with about five gigawatts of compute, up from about two in January. But much of that capacity is rented from someone else, including from SpaceX. Meanwhile, Musk is promising to add 10 gigawatts of compute next year alone.

And that means… the entire industry will soon depend on Musk.

Anthropic, maker of the Claude AI models and a direct competitor of SpaceX’s Grok, has committed $84.5 billion to rent SpaceX computing capacity through 2029. Alphabet, which designs its own chips and runs one of the largest computer fleets on Earth, agreed to pay SpaceX $920 million a month from this October through June 2029 for compute.

Why would Google rent compute from Elon Musk?

The answer is power. SpaceX ended September with computers consuming 1.7 gigawatts of electricity. A gigawatt is a billion watts, roughly the output of one nuclear reactor. Electricity is the binding constraint for data centers, so the industry now uses electrical consumption to measure the size of a data center. Musk said SpaceX will reach 2.5 gigawatts to 3 gigawatts by year-end, and, as we explained earlier, he said he is “highly confident” the company will add 10 gigawatts of compute in 2027.

How’s that possible?

GE Vernova (GEV) is the largest American maker of the gas turbines that power data centers. And it is sold out through 2030. Its current backlog is 116 gigawatts. But it can build only about 20 gigawatts a year. If you want to power your computers with their turbines, you’re going to wait years.

So how can SpaceX add 10 gigawatts of compute next year?

Officially, Musk says his power solution is “highly proprietary.” But we know SpaceX has built its own power center in Southaven, Mississippi, with 41 turbines making 1.2 gigawatts. It’s using a range of suppliers to build these new facilities, including buying five 380-megawatt machines from Doosan Enerbility (034020.KS), the Korean manufacturer.

SpaceX will soon be one of the largest electric utilities in the United States. How? It is routing around GE’s backlog, primarily by casting its own turbine blades, which are the key functional part of the machine. The blades inside a gas turbine spin in gas hotter than the melting point of the metal they’re made from, which is an engineering challenge – but similar to the physical challenges of building rockets. Musk says SpaceX has the most advanced casting capability, which it uses to build rocket engines.

That means that we can bring power online 12 to 18 months faster than anyone else.

It’s hard to imagine what all of this means for our economy. These are private investments of unprecedented scale.

By our estimate, using the fuel rates of today’s gas plants, SpaceX’s 20 gigawatts of gas turbines (estimated capacity by the end of 2027) would burn 3 billion to 4.5 billion cubic feet of natural gas a day. The U.S. Energy Information Administration expects American gas production to grow by 4.2 billion cubic feet a day in 2027.

SpaceX is planning to burn most of a year’s growth in American gas production by the end of 2027.

How Is Any Of This Economic?

Of course, none of this is economic unless AI begins to replace an enormous swath of white-collar workers. And that is what’s going to happen. In fact, despite the huge increase to installed compute capacity, the spot price for compute is going up.

On the August SpaceX quarterly call, Musk guessed that each watt of Nvidia’s next-generation capacity would earn $30 to $50 a year. On yesterday’s quarterly call, SpaceX CFO Johnsen said pricing is now tracking “slightly above the high end” of that range. Musk said short-term contracts are renewing “at higher dollars per watt, not lower dollars per watt.” And Johnsen said that at current prices, new computing capacity pays for itself in less than a year.

Johnsen reported SpaceX revenue was $14.6 billion, up 161% from a year earlier. That was $2 billion above what Wall Street expected. And it posted real earnings for the first time as a public company, with operating earnings before non-cash charges that soared more than 300%. CFO Johnsen told investors to expect $33 billion to $37 billion of revenue in Q1 2027 – more than 100% growth from the last quarter. Johnsen also explained that by December of this year SpaceX will be booking recurring revenue at more than a $100 billion annual pace.

I have been reading quarterly reports for more than 30 years. I’ve never read one in which a single company described absorbing several entire economy-wide functions at the same time, at scale: the global communications network, the national electric power generation, the global compute… all to power software that thinks.

SpaceX’s Total Dominance

SpaceX is going to completely dominate technology unlike virtually any other company in history. Have we ever seen anything like this before? From 1877 until 1984, the Bell System ran America’s phone system. AT&T owned the long-distance lines. Its subsidiary Western Electric manufactured the phones, the cable, and the switches. Bell Telephone Laboratories designed the network and invented the technology inside it, while 22 local operating companies installed the phones and sent the bills. By the time the Justice Department broke it apart, the Bell System employed more than 1 million people, held $150 billion of assets, and was the largest corporation in the world.

That’s where SpaceX is heading.

Do the math to see where this leads. At $50 per watt, a gigawatt of compute currently produces $50 billion of revenue a year. Nvidia CEO Jensen Huang puts the cost of a gigawatt data center at $50 billion to $60 billion, including the chips, the buildings, and the power. At today’s prices, then, a gigawatt earns back roughly its own construction cost in revenue every 12 months.

And no one is going to build more of it than Musk’s SpaceX. How will he pay for it…?

In Q3, SpaceX operations produced about $9.6 billion of cash against $19.6 billion of capital spending, so free cash flow was roughly negative $10 billion. In Q4, we estimate operations will produce about $12 billion in cash, based on our assumption of roughly $22 billion of revenue, against $35 billion to $40 billion of spending.

That is negative free cash flow of $23 billion to $28 billion in a single quarter.

And that’s what you’ve seen the bears warning about. We’ve warned about the same issue with less well-funded companies, like CoreWeave. But Musk has cell phones and rockets.

For 2027, we modeled three cases, using our Q4 2026 estimate as the starting point and holding cash margins near this year’s level of about 55% of revenue.

In the first case, SpaceX keeps spending $40 billion a quarter, its Q4 pace, for all of 2027. That’s $160 billion of spending. After counting the cash its businesses generate, SpaceX would burn about $82 billion. In the second case, we use Morgan Stanley’s estimate of $278 billion of 2027 spending, which would burn about $195 billion. In the third case, Musk builds his full 10 gigawatts. We put that bill near $425 billion and the burn near $326 billion.

Its Three Levels Of Borrowing

Can SpaceX borrow that much and keep its investment-grade credit rating?

Lenders and rating agencies answer that question with one ratio: debt divided by EBITDA (earnings before interest, taxes, depreciation, and amortization). It’s a rough measure of the cash a business produces each year before it pays its lenders and the taxman. Each year of EBITDA is called a “turn.” A company with $100 billion of debt and $50 billion of annual EBITDA is levered two turns: it would need two full years of cash earnings to pay off its debt.

SpaceX ended September with $89.9 billion of cash and short-term investments and $38.8 billion of debt. In Q3, it produced $8.8 billion of EBITDA, a pace of about $35 billion a year. So today SpaceX owes about one turn of EBITDA, and it holds more than twice as much cash as it owes.

S&P’s current rating assumes SpaceX keeps its debt below two turns of EBITDA. In our first case, SpaceX spends most of its cash and borrows about $40 billion more. Its debt rises to about $77 billion against roughly $87 billion of 2027 EBITDA – less than one turn and well inside investment grade. In our second case, SpaceX has to borrow about $150 billion. Its debt reaches about $190 billion against $95 billion of EBITDA. That’s two turns, right at S&P Global’s limit for investment-grade status. In our third case, SpaceX needs about $280 billion of new money. If it borrowed all of it, its debt would reach about $320 billion against $116 billion of EBITDA. That’s nearly three turns, and it would cost SpaceX its investment-grade rating.

So to build all 10 gigawatts and stay investment grade, SpaceX would have to raise some of the money another way: by selling new stock, by having suppliers like Nvidia finance the chips, or by getting customers to pay in advance. S&P already expects SpaceX’s borrowings to reach $132 billion by 2028, and it doesn’t expect free cash flow to turn positive until 2030.

How certain is Musk that this will work? Musk said SpaceX rents out its computing “at effectively the spot market rate as opposed to signing these long-term deals.” So what happens to SpaceX if the price of a watt of computing falls by a third?

That’s the risk investors are taking with SpaceX. But that risk has two sides. What if demand for compute continues to grow? In that case, SpaceX’s revenue and earnings will be larger than our forecasts. And I doubt we’ve begun to see what Musk can do with all of this compute.

Yesterday, Musk confirmed that special committees of independent directors at SpaceX and Tesla (TSLA) are discussing a merger of the two companies. If it happens, SpaceX gains Tesla’s factories, its self-driving cars, and its Optimus humanoid robots, and the intelligence SpaceX sells by the watt would start driving cars and working factory floors.

Merging SpaceX And Tesla

Think about what that means: SpaceX launches the satellites, Starlink carries the signal, its turbines make the power, its chips run the AI models. Grok answers the questions. Tesla’s robots do all the work. Every layer of the global computer would be owned by one company, SpaceX.

For most of the 20th century, American industry grew through specialization. Steel came from one firm, electricity from the local utility, and telephone service from Bell – and each of those companies earned its own margin on its own piece of the economy for decades.

Computing at this scale reverses that logic. When the cost of intelligence is power plus chips plus distribution, the integrated owner of all three earns every margin in the chain. Its smaller competitors will fail even if they’re well run, because they pay a markup on every input the leader produces at cost.

Why not buy the stock?

On Thursday, about 1.3 billion shares come out of their post-IPO lockup, the largest single release on the schedule. Roughly 22% of SpaceX’s 13.2 billion shares can trade today, rising to about 32% after Thursday, and 40% after December 8, while Musk’s own 6.4 billion shares stay locked until June 2027.

I expect SpaceX to report a quarter with more than $50 billion of revenue before the end of 2027, nearly 9x the $5.6 billion it reported in Q3 2025.

I’m almost certainly going to be a buyer of the stock. I just want to wait until it’s mostly freely trading first.

Good investing,

F. Porter Stansberry
Stevenson, Maryland