SpaceX is integrating global communications networks, computing infrastructure, and energy supply under a single company at an unprecedented pace. A quarterly earnings report that stunned Wall Street shows the company led by Musk is betting on the core lifelines of the AI era more aggressively than any tech giant.
SpaceX reported third-quarter revenue of $14.6 billion, up 161% year over year, beating Wall Street estimates by more than $2 billion and recording a positive operating profit for the first time. CFO Bret Johnsen guided that first-quarter 2027 revenue will reach $33 billion to $37 billion, and that annualized recurring revenue will exceed $100 billion before December this year. At the same time, Musk confirmed that special committees of independent directors at SpaceX and Tesla Motors (TSLA) are discussing a merger of the two companies.
If the merger happens, SpaceX would bring together satellite launch, global communications, AI computing power, and energy supply, along with Tesla's factories, autonomous vehicles, and Optimus humanoid robots. At that point, every layer of the global computing system would be owned by one company.
Starlink: Approaching the Tipping Point That Could Upend the Entire Communications Industry
Starlink's satellite internet service added 1.6 million users in the third quarter, bringing cumulative subscriptions to 13.6 million. Although that number is far below Verizon's 147 million wireless connections and AT&T's 110 million, its growth curve cannot be ignored: from 2.3 million at the end of 2023, to 4.4 million at the end of 2024, to 8.9 million at the end of 2025, with another 4.7 million added in the first three quarters of this year.
By comparison, AT&T added only 379,000 postpaid users in the same quarter, while Verizon added 184,000. In addition, about 12,000 commercial aircraft have connected to Starlink service, and this coverage capability is becoming a key variable in airline competition: United Airlines has connected, while Delta Air Lines has not yet done so.
Starlink Mobile is another growth curve. The service does not require a rooftop antenna and connects ordinary phones directly to satellites through partner carriers such as T-Mobile, and has already reached 22 million monthly active users, growing 37% in the third quarter alone. SpaceX also recently reached an agreement to acquire wireless spectrum from EchoStar for about $17 billion and plans to launch a full mobile carrier service in early 2028.
On the technology side, SpaceX's new-generation V3 satellites each have a download capacity of 1 terabit per second, enough to support about 200,000 simultaneous video streams, while a single Starship rocket can carry 60 of these satellites. In September this year, Starship's 14th flight deployed bandwidth of 26 terabits per second in a single mission, more than eight times the capacity of Viasat's entire ViaSat-3 satellite fleet. According to research firm TeleGeography, global international internet capacity in mid-2026, mainly relying on undersea cables, will be about 2,259 terabits per second. As Starship aims to fly once a week next year and eventually once a day, Musk's network scale may exceed total global internet capacity.
Computing Race: SpaceX Is Becoming the World's Largest Computing Supplier
Computing power is becoming a form of infrastructure, and Musk intends to dominate that infrastructure. SpaceX is currently bringing about 220,000 Nvidia GB300 chips online each month, and Musk said this pace will accelerate further next year.
According to Foxconn, it has received orders from SpaceX for about 936,000 GB300 chips, worth about $52 billion, with deliveries through the first quarter of 2027. Industry estimates put Nvidia's total Blackwell chip output this year at about 5.5 million to 6 million units, with SpaceX alone accounting for about 40%, making it Nvidia's largest single buyer.
SpaceX has already built Colossus 2, the world's largest AI data center, in Memphis, Tennessee, but that is only the starting point. According to the timetable disclosed by Musk and CFO Bret Johnsen on the earnings call, SpaceX's computing power had reached 1.7 gigawatts by the end of September; it is expected to approach 2 gigawatts by the end of October and reach 2.5 to 3 gigawatts by the end of December; by the end of the second quarter of 2027, it will bring online the "world's strongest training cluster" with 440,000 Nvidia next-generation Vera Rubin chips; and its full-year target is to add 10 gigawatts of computing power. Musk said he is "highly confident" about this and that the actual figure is "unlikely to be below 13 gigawatts."
Competitors are actively seeking computing power from Musk. Anthropic has committed to renting about $84.5 billion worth of computing power from SpaceX by 2029; Alphabet agreed to pay SpaceX $920 million per month for computing power from October this year through June 2029. Even Google, which develops its own chips and owns one of the world's largest computing clusters, has chosen to buy computing power from Musk, with the root cause being electricity.
Energy Bottleneck: SpaceX Chooses to Build Its Own Power System and Go Around It
Electricity is the fundamental constraint on data center expansion. GE Vernova is the largest U.S. manufacturer of gas turbines for data centers, and its orders are fully booked through 2030, with a current backlog of 116 gigawatts against annual capacity of only about 20 gigawatts.
SpaceX's solution is to build its own power system. The company has completed its own power generation facility in Southaven, Mississippi, equipped with 41 gas turbines and an installed capacity of 1.2 gigawatts; it is also purchasing five 380-megawatt units from South Korea's Doosan Heavy Industries. Musk said its energy solution is "highly proprietary," with the core being that SpaceX has mastered advanced casting capabilities, derived from rocket engine manufacturing, that can be used to independently produce gas turbine blades, thereby shortening power deployment cycles by 12 to 18 months.
Musk plans to build 20 gigawatts of new power and cooling facilities next year to ensure that computing expansion is not constrained by energy. By one estimate, if SpaceX reaches 20 gigawatts of installed gas turbine capacity by the end of 2027, daily natural gas consumption would reach 3 billion to 4.5 billion cubic feet, equivalent to most of the projected full-year increase in U.S. natural gas production in 2027 expected by the U.S. Energy Information Administration.
Financial Pressure: The Financing Logic Behind Negative Free Cash Flow
Aggressive expansion comes with enormous capital consumption. SpaceX generated about $9.6 billion in cash from operating activities in the third quarter, while capital expenditure reached $19.6 billion, leaving free cash flow at about negative $10 billion. CFO Bret Johnsen guided fourth-quarter capital expenditure at $35 billion to $40 billion, about twice the figure previously expected by investors.
Compared with other hyperscale cloud providers, SpaceX's capital expenditure intensity far exceeds that of peers. Amazon spent about $0.27 on capital projects for every $1 of revenue this year, with Microsoft and Alphabet at similar levels; SpaceX's ratio is about $2.50, roughly nine times Amazon's.
As of the end of September, SpaceX held $89.9 billion in cash and short-term investments, had $38.8 billion in debt, and generated quarterly EBITDA of about $8.8 billion, or about $35 billion annualized. Current liabilities are about one times EBITDA, far below the two-times ceiling for S&P to maintain an investment-grade rating.
However, if SpaceX is to complete 10 gigawatts of computing buildout and maintain an investment-grade rating, the funding gap means it must raise additional capital by issuing new shares, having suppliers such as Nvidia provide chip financing, or collecting prepayments from customers. According to reports, S&P Global already expects SpaceX's borrowing to reach $132 billion by 2028 and expects free cash flow to turn positive only in 2030.
Tesla Merger: The Final Piece in Building a "Full-Stack Empire"
Musk has confirmed that special committees of independent directors at SpaceX and Tesla Motors (TSLA) are discussing a merger. If the deal is completed, SpaceX would gain Tesla's manufacturing plants, autonomous driving technology, and Optimus humanoid robots, and its AI computing power sold by the watt would in turn drive vehicle driving and factory production.
This implies a complete vertically integrated chain: SpaceX launches satellites, Starlink transmits signals, self-built gas turbines supply power, in-house chips run AI models, Grok provides intelligent services, and Tesla robots perform physical labor, with every layer of the global computing system controlled by the same company.
This logic overturns the basic paradigm of specialized division of labor in American industry over the past century. In the AI era, when the cost of intelligence equals electricity plus chips plus transmission, the owner of vertical integration will capture profits at every layer of the chain, while competitors that rely on purchasing various inputs externally face cost disadvantages at every step and will struggle to survive even if they operate well.
Looking at the revenue trajectory, SpaceX is expected to report quarterly revenue of more than $50 billion by the end of 2027, nearly nine times the $5.6 billion recorded in the third quarter of 2025. Whether this bold bet pays off depends on the core variable of continued growth in AI demand for computing power, and on whether Musk can truly get this machine spanning the sky, the ground, and the digital world running at unprecedented speed.
Comments