SpaceX has spoken for the first time about its plans for a terrestrial mobile network to work with its satellite D2D service to rival AT&T, T-Mobile, and Verizon – including plans to roll it out on small cells attached to its ground infrastructure. It is betting on “dead zones” in the near term, plus a future market for outdoor physical AI – cars and robots – that will dwarf the current smartphone market. Analysts are still skeptical.
In sum – what to know:
Terrestrial mobile – SpaceX/Starlink plans to launch next-generation D2D satellites in 2027 and add terrestrial small cells to undercut standard cellular economics, and build a hybrid service to rival AT&T, T-Mobile, Verizon.
Machine comms – There is still some talk about “dead zones”, but the firm framed the opportunity around connecting vehicles and robots, arguing that bandwidth demand will give Starlink a big share of internet traffic.
Quarterly mix – SpaceX almost doubled its revenues in the second quarter, but also raised capital spending by six times – going on launch, satellite, and AI, mostly, compute capacity.
SpaceX has confirmed plans to expand its Starlink satellite service to the mainstream mobile market – on small cells, ostensibly, attached to its existing ground infrastructure. The plan pitches it squarely against AT&T, T-Mobile, and Verizon in the US. The firm will launch new “direct-to-cell” (direct to devices; D2D) satellites next year (2027), and a commercial service by the end of it. It expects to take customers from the big-three US carriers, it said, because its “service will be better”, and to eat into their combined revenues of about $600 billion per year.
But it maintained its primary line about eliminating “dead-zones”, whilst, in customary fashion, also talking up a future world of connected robots and vehicles that will positively dwarf the existing people-led mobile market. Analysts greeted its vaguely sketched infrastructure plan with small cells with a degree of suspicion. “Elon Musk is nuts if he thinks 5G small cells with Starlink backhaul can offer better (and cheaper) service than AT&T, Verizon, and T-Mobile,” responded Dimitris Mavrakis, senior research analyst at ABI Research, writing on social media. (Meanwhile, a spent upper stage for a SpaceX rocket has crashed into the Moon after drifting around space for about 18 months.)
Starlink said on a second-quarter earnings call (August 4) that a 13-fold increase in spectrum capacity (from 5 MHz, via telcos, to 70 MHz, including its own), following the purchase of 65 MHz of AWS 3/4 from EchoStar in late 2025, plus a 10-fold jump in the number of its D2D satellites, will make its mobile service “100 times better” than it is today – by the end of 2027, the implication goes. It also sketched a plan to build a distributed network of small cells alongside existing satellite ground terminals to undercut conventional cellular build-out costs – and effectively upend analyst forecasts that it will take about $100 billion in cap-ex to go head to head with AT&T, T-Mobile, and Verizon.
Except that is not really what Musk et al are planning – to take on legacy telcos as a fourth US operator just for smartphone subscriptions. Instead, chief executive Elon Musk and chief operating officer Gwyn Shotwell talked on the call about connecting vehicles, robots, and other machines in the physical AI ecosystem. Musk said: “It’s not out of the question that, at some point, Starlink will deliver a majority of the world’s internet – at least in countries where we’re allowed to operate…. And it’s not in the infinity future; it’s [in] less than 10 years. With the advent of AI, and humanoid robotics and vehicle robotics, the appetite for bandwidth will be much greater than in the past.”
He went on: “You think of a human consuming or producing perhaps a few hundred bits per second of output. Actually, the average human is outputting less than one bit per second over the course of a day – which is 86,400 seconds. Computers could easily do billions of bits per second continuously over a 24-hour period. This is why I would expect the appetite for bandwidth to grow dramatically with the growth of AI and robotics. Really, I think Starlink is the only thing that can actually service that bandwidth.” Musk suggested a 10-fold increase in V3 satellite volumes, plus a further 10-fold in V3 capability, will deliver a 100-times surge in D2D network capacity.
But given the spiralling demand to connect physical AI, as projected, Starlink reckons it can easily absorb inevitable price erosion. “Even if our monetization per bit dropped by a factor of 10, that would still mean a 10-times increase in the revenue of Starlink. I think people are really underestimating Starlink here. This is a big deal.”
Timeline and architecture
Meanwhile, Shotwell clarified: “We will start providing service [by the] end of next year.” She said the EchoStar deal gives Starlink 65 MHz of its own spectrum, versus a 5 MHz facility enabled via partnerships with local mobile operators, as currently. “It gives a massive increase in capability,” she said. “[It] does have terrestrial components. We definitely intend to build out the terrestrial component.” The firm’s build-out will seek to avoid the standard capital expense incurred by big telcos, she suggested. Instead of deploying conventional macro cell towers, Starlink is looking at “cap-ex efficient” ways to put small cells on its existing broadband terminals.
Shotwell remarked: “Let me give a little hint: you could put a cellular base station on the gear that holds a Starlink broadband dish. You could have these little femtocells around the country, and deploy as you need to. You don’t necessarily have to spend billions of dollars in low-band spectrum all upfront before you deploy your system… We have a lot of great and new ideas for how we’re going to do it. It will be quite cap-ex efficient.” She declined to put an “efficient” cap-ex figure on the project.
Musk added: “Instead of having to deploy these very expensive and difficult-to-locate large cellular base stations, we feel reasonably confident we can deploy a large number of sort of small stations essentially. They’re really just Starlink dishes that also provide connectivity in the mobile spectrum bands, and have them be all over the place. The Starlink antennas are located on the roofs of houses and businesses, so they’ve got a sort of clear view for providing connectivity directly to cell phones on the ground. In fact, providing connectivity that is, we think, probably better and higher bandwidth than what is currently available from cellular providers. This is an important point.”
Shotwell stated: “I anticipate us to be able to acquire quite a few of their (the big operators’) customers because I think our service will be better. We will eliminate dead zones leveraging basically the satellites in orbit.”
Caution and watch-outs
Telecoms analysts greeted the rather-vague infrastructure plans with some caution. Writing on social media, Dimitris Mavrakis, senior research director at ABI Research, stated: “Elon Musk is nuts if he thinks 5G small cells with Starlink backhaul can offer better (and cheaper) service than AT&T, Verizon, and T-Mobile. Sure, if [SpaceX/Starlink] deploy millions, then perhaps. But connecting them is not the biggest problem; getting power to them and finding the optimal location are the biggest problems – let alone economics, ROI etc. Unless it is thinking of existing Starlink locations, which is not correlated with cellular coverage requirements.”
He added: “The industry has already concluded femtocells are not a viable primary deployment model, even for MNOs… Sure, [they] can be used for filling in coverage blackspots in the home, but not for nationwide mobile broadband services.”
Meanwhile, Subhadip Roy, a senior analyst at Counterpoint Research, responded: “Starlink is no longer positioning itself solely as a satellite connectivity partner… [It] still has a long way to go [but], while 65 MHz is only a fraction of the licensed spectrum held by incumbent US operators, SpaceX appears to be betting on a hybrid satellite-terrestrial architecture and its existing Starlink infrastructure to deploy mobile coverage faster and more cost-effectively than traditional macro networks.”
He added: “The timing is interesting. AT&T, Verizon and T-Mobile announced their Direct-to-Device (D2D) joint venture nearly three months before SpaceX publicly revealed its terrestrial ambitions. While the JV wasn’t announced as a response to SpaceX, it now looks like a strategically valuable move as satellite connectivity becomes increasingly competitive. AST SpaceMobile’s position may become even more important. With AT&T and Verizon already aligned with AST, operators now have an alternative D2D ecosystem as SpaceX moves closer to competing for retail subscribers. Existing Starlink partnerships are worth watching. Starlink currently works with operators including T-Mobile, Rogers, KDDI, Optus, One NZ, Salt, Entel and many others. If Starlink eventually launches its own retail mobile offering, those partnerships could face channel conflicts or evolve into new partnership models.
Second-quarter scores
All told, SpaceX reported higher second-quarter revenue and improved operating earnings, and beat analyst expectations as its rocket, satellite, and AI units all posted gains – although shares later fell, erasing most of their early gains as investors digested the company’s capital spending. Revenue rose 92 percent, almost double, versus the same period a year ago, finishing the quarter at $7.8 billion – comfortably ahead of forecasts of $6.81 billion. Adjusted EBITDA almost tripled to $3.5 billion, up 191 percent from a year earlier. The company’s net loss narrowed by $467 million to $541 million. But its spending sky-rocketed to $18.3bn – more than six times its year-ago spend.
Shares initially rallied, closing up 9.4 percent before falling 8.6 percent in after-hours trading to $114.60. The company spent $18.4 billion during the quarter, including roughly $15.8 billion on AI compute infrastructure. It ended the quarter with $100 billion in cash and marketable securities and a backlog of $47.5 billion. Management said capital expenditure will remain at current levels through the second half of the year.
But the results underline the company’s diversified portfolio across space transportation, satellite broadband, and AI computing. The Starlink business remains its largest, generating $4.3 billion in revenue, up 66 percent year-on-year, with 1.7 million net-new subscriber additions in the quarter. The launch business also continued to expand, with space segment revenue reaching $962 million. SpaceX said it completed 78 launches during the first half of 2026, delivering more than 1,000 tonnes of payload into orbit. It aims to increase annual payload capacity from around 2,500 tonnes today to more than one million tonnes a year, with a longer-term ambition of 10 million tonnes.
The AI division was the fastest-growing, with revenue up by 247 percent to $2.6 billion. SpaceX will standardize on Nvidia’s Vera Rubin architecture and begin launching ‘Starmind’ AI satellites next year. The company expects AI compute capacity to approach 10 GW by the end of 2027. Grok 5, trained on SpaceX’s engineering data accumulated over the past 25 years, is expected before the end of this year. The company expects to finish 2026 with more than 2 GW of AI compute capacity, up from 1.4 GW at the end of the second quarter, before expanding to around 15 GW of power and cooling capacity by the end of 2027, with a longer-term target of approximately 20 GW.
It expects to achieve a $100 billion annualised revenue run rate by the end of 2026. Management also pulled forward its internal target for $1 trillion in annual revenue to 2030, from a previous projection of 2031 – while suggesting there is also a “non-zero chance” of getting there as early as 2029.
