1. Amazon–Globalstar is the second price tag on the outside-in strategy
The $11.6 billion enterprise value, roughly $10 billion of which sits on Globalstar's narrow slice of direct-to-cell (DTC) spectrum, is the second data point we have for what a hyperscaler is willing to pay for DTC. The first was SpaceX's acquisition of AWS spectrum from EchoStar. Price per unit of capacity is roughly comparable across the two deals. That matters, because it calibrates a market that has almost no public comparables.
The skeptical read on the Amazon deal is straightforward: DTC is physics-limited. Spot beams run thousands of kilometers wide, and this spectrum tops out at about 100 Mbps per beam. In dense urban areas (16,000 people per square kilometer in New York City), this is not a broadband solution and never will be. So why pay $10 billion for what, on its own, amounts to texting coverage at the rural edge?
Because the spectrum is not the product. The bundle is the product. Once you pair ubiquitous satellite coverage (home broadband plus DTC) with a wholesale relationship that plugs the urban density gap, you own the customer across every context where they consume connectivity. Layer on Amazon Prime, or on SpaceX's side xAI and Grok subscriptions, and the upsell mechanics become extremely hard to compete with. Our previous TAM work treated DTC as a roughly $40 billion coverage-gap market by 2030. The bundled-ownership frame reaches $100 billion-plus, and the long-term ceiling is the share these operators can take of the $3.5 trillion global telco market.
We've been framing this as a "Grand Canyon to urban canyon" inversion. SpaceX and Amazon have already solved the hardest part of telecom: getting signal where terrestrial economics don't pencil. The remaining gap is density, which is a much more tractable problem. It can be solved with wholesale deals, MVNO-style structures, or targeted owned infrastructure. Amazon just wrote a $10 billion check that only makes sense if they agree.
2. The Starlink Flywheel Holds at 20,000 Satellites Per Year
Gwynne Shotwell recently suggested in a public comment that Starlink would ultimately cap at around 20,000 satellites. For a model that projects well past that number by 2030, this was a meaningful problem, and we were preparing to revise our Starlink revenue projections down significantly. Our bandwidth trajectories all assumed a materially larger constellation.
Elon's April 12 post, committing to roughly 20,000 satellites launched per year at V3 mass (roughly 2 tons each), resolves it. Our current model path runs to roughly 100,000 terabits of delivered bandwidth by 2030, though we flagged on the call that this number will likely be revised downward modestly in the next quarterly report, as more launch capacity gets allocated to orbital compute payloads earlier than we originally assumed.
The related read, which came up in audience questions, is on xAI cash burn. Our financial modeling suggests Starlink free cash flow alone can absorb xAI's CapEx within roughly one to two years. Our $50 billion cash infusion scenario, of which we estimate $20–30 billion would be fresh primary capital with the balance as secondary, compresses the Starlink deployment curve by about two years. It accelerates the flywheel; it doesn't rescue it. The flywheel itself is intact: each V3 satellite, over its five-year operational life, generates enough revenue to fund multiple replacements. The math compounds.
3. New Glenn's reuse is bullish; its grounding reprices everyone else
Two things happened on NG-3. Blue Origin's first-stage booster landed successfully on its second flight, which Aaron framed as a genuinely bullish signal for reusability outside of SpaceX. And the upper stage underperformed, leaving the ASTS payload in the wrong orbit. The FAA has grounded New Glenn pending a mishap investigation. The duration is open.
The cascading implication is what matters. SpaceX's launch manifest is increasingly absorbed by internal payloads, Starlink today and orbital compute tomorrow, which means New Glenn has effectively become the primary ancillary launch provider for everyone else. Amazon Leo's deployment schedule and the ASTS manifest are the two most exposed. Every New Glenn setback is a delay for the non-SpaceX side of the industry.
Aaron's framing on execution risk deserves emphasis. Public-market inflows into space-exposed ETFs ran roughly 10x during the recent SpaceX IPO speculation window. A lot of that capital is being priced as if execution risk is evenly distributed across the sector. It isn't. SpaceX is, functionally, at 99% on execution. The next-best executor is Rocket Lab in small launch, and the step down from there to the rest of the field is large. Blue Origin sits somewhere in that middle tier; the reuse milestone is real progress, but the orbital miss is a reminder of where they are. There should be a meaningful and persistent valuation premium between SpaceX and the field. We don't expect it to narrow on the timelines public markets seem to be pricing.
Disclosure: Mach33 holds a position in Stoke Space. Aaron's view on the call was that Stoke has the right architecture for the long run, full-flow staged combustion engines among other elements, and that their capital efficiency gives them more shots on goal than most competitors. But execution-wise, like every non-SpaceX launcher, they are functionally at zero until orbit.
4. The SpaceX–Cursor option is a call on the orchestration layer
The mechanics: SpaceX holds a call option to acquire Cursor for $60 billion by year-end, with a $10 billion walk-away partnership fee as the alternative. In the interim, the two companies are working together. Cursor's ARR has doubled to roughly $2 billion in a few months, with 1 million daily developers and two-thirds of the Fortune 500 as customers.
The thesis signal matters more than the transaction. Elon's historical public posture has been that compute and LLM scale are the variables that matter, with orchestration and surface layers treated as marginal. Our view, informed by our co-founder Brant's roughly two years of deep engagement with the AI stack, is that the stack is four layers: compute, probabilistic models, orchestration (which turns probabilistic outputs into more deterministic software), and the user-facing surface. Aaron put it directly on the call: orchestration may capture as much as 50% of the total value in the stack, or more. The Cursor deal is the most explicit departure we've seen from the "just scale the LLM" framing on the xAI/SpaceX side.
Our lean on the option itself: the $10 billion walk-away is the more probable outcome. SpaceX has historically preferred to observe, absorb the learning, and replicate rather than acquire. The conditions that would flip it toward exercise are founder retention and a judgment that customer-distribution inheritance compresses time-to-revenue meaningfully. For public-market watchers pricing the SpaceX IPO: we don't expect the xAI productivity narrative to be front-loaded into the initial book. Expect it to land as a "surprise" beat in the first or second quarterly print after listing.
5. What to watch next
Two audience questions flagged items worth tracking.
On the FCC's recent approval of a roughly 7x power increase for satellite broadband: the headline number is misleading. Spectrum remains the binding constraint. As Vlad put it on the call, the spectrum is the straw and the satellites are the juice. More power primarily buys throughput for existing users inside a given beam, not proportional gains in population density, because raising power also raises inter-beam interference and constrains spectrum reuse. Expect better service for existing Starlink customers, not a structural capacity unlock.
On competitive M&A, Vlad flagged AST SpaceMobile as the most logical Amazon acquisition target. This is speculation, not a forecast, and not financial advice. Amazon paying $10 billion for narrow DTC spectrum implies either an in-house DTC payload program already in development or an intent to acquire one. AST is the only operator with high-bandwidth DTC hardware today, they've been riddled with execution issues, and they just lost their payload on NG-3. A third serious outside-in operator, Amazon with ASTS inside, would change the competitive shape of the DTC market materially. It's a live option, not a base case.
Our next Private Podcast is Thursday. Subscribers can access the full recording on the platform. Questions, feedback, and topic requests are welcome.