Summary

Starship V3’s first flight last week “was a huge success” — full-system demo of the new V3 Raptors plus a soft second-stage splashdown gives Johnsen conviction “not just in years to come, but really even in the next couple of flights.” Once the second stage returns to the tower for rapid reuse “in the next couple of years,” he expects another 10x drop in cost per kilogram vs. Falcon — chasing “the holy grail of rocketry”: aircraft-like operations at 100 metric tons to LEO.
Starlink is attacking a ~$1.6T telecom market with what Baker calls the first genuinely differentiated product in telecom history — 10M+ customers, 10,000+ satellites, 160+ countries in six years, and Johnsen says 10M “can become hundreds of millions.” Next-gen direct-to-cell will be “5G quality… in the next two years,” and each Starship carries 20x the Starlink V3 capacity of a Falcon launch.
Orbital compute is “racks in space,” not buildings — larger Starlink V3 satellites with more solar, Nvidia GPUs, and a radiative-cooling sheet, virtually networked “literally like another constellation.” First-principles edge: 5x+ solar energy per cell, sun-sync 24-hour power, no land lease, and avoidance of the “data center in my backyard” concern — and while terrestrial data-center costs are rising, the satellite’s cost “is mostly silicon” and benefits from silicon cost reductions, process node to process node.
The launch math today: ~200 Starship launches per gigawatt of orbital compute (first-gen satellite and rocket), so SpaceX is “capacitizing for thousands of launches a year” across four towers — two in South Texas, Cape Canaveral, and LC-37 within the next year — with capabilities demonstrated “as soon as next year.”
SpaceX’s AI business is already a top-5 AI infrastructure operation by Baker’s napkin math — the Anthropic hosting deal puts it at a $3.75B-per-quarter run rate, “50% larger than the company just outside the top four,” while its own models train and do inference on bleeding-edge GB300s and the Cursor deal brings over half the Fortune 500 as enterprise accounts.
Terafab exists because SpaceX fears silicon supply won’t scale — Nvidia, AI5, or TPU all resolve to TSMC one layer down, and the target is “ideally 100 gigawatts a year.” The Intel partnership plus captive customers (“we will take every wafer that you can yield out”) reduces it to “a capital risk only.”
The capital-allocation timing is the quiet thesis: Starship flies Starlink V3, which unlocks cash-generative Starlink growth and presumably high-margin direct-to-cell exactly as orbital compute needs heavy capex. Johnsen: “I wish I could take credit for that, but that’s certainly Elon.” And the vertically integrated stack is open at every layer — competitors can buy launch, Starlink, terrestrial compute, or the model.

Deep dive

1. Starship V3 builds conviction on “the holy grail of rocketry”

Johnsen’s starting point: “It’s hard to be a space company and not have assured access to space.” SpaceX is already lowest cost per kilogram to space ever, and Starship targets another 10x — but the real project is rapid reusability: “It’s one thing to bring back the first stage… we did that 10 years ago with Falcon. What we’re doing now is flying the largest rocket ever built with the goal of it getting to aircraft-like operations.”
On last week’s first V3 flight — which Baker admits felt like “a high-risk event heading into an IPO” — Johnsen counts it “a huge success”: full system capability, new V3 Raptor engines, and the soft second-stage splashdown give conviction “not just in years to come, but really even in the next couple of flights.”
The payoff, hedged as stated: once the second stage returns to the tower for rapid reuse — “we’ll just say in the next couple of years” — expect a 10x in cost per kilogram vs. Falcon today, at 100 metric tons to LEO. That’s the catalyst for “turning the 2030s into something that we had expected when we were kids.”

2. Starlink: telecom’s first differentiated product in a $1.6T market

Baker’s framing as a former telecom analyst: ~$800B internet access plus ~$800B cellular, and “there’s never actually been a differentiated or disruptive product in telecom” because everyone shares the same towers, rights-of-way, and equipment — yet at every location he’s been to, “Starlink is better… faster and lower latency.”
The scale after just six years of production satellites: 10M+ customers, 10,000+ satellites flown, 160+ countries — and Johnsen’s call that 10M “can become hundreds of millions… because it’s so much more efficient to deliver to so many different locations from space than it is terrestrially.” Next-gen direct-to-cell will be “5G quality… in the next two years” — global roaming, no dead zones — and terrestrially, “it will become harder and harder to justify many of the deployments… because Starlink now exists.”
The cost side is as important as the product: Baker’s framing is “You’re not digging ditches” — once the constellation exists, shipping the terminal “is basically most of your customer acquisition cost.”
Starlink was also the operational catalyst for launch — payloads lining up drove Falcon to 165 launches last year, with Starlink stacks staged to fill gaps between third-party missions. Starship repeats the pattern: each launch of V3 broadband satellites carries 20x the capability of a Falcon launch, and next in the queue is AI compute.

3. Orbital compute is “racks in space” — a logical extension, not a moonshot

Johnsen had the same confusion the public does: “I thought, ‘Oh wait, how are we going to connect all these pieces together?’ And they’re like, ‘Johnson, this is literally like another constellation.’” The design: virtually networked satellites that look like larger Starlink V3s — “a lot more solar and now compute over the top,” starting with Nvidia GPUs and “a large sheet of metal” for radiative cooling. “There’s so much of what we’re already doing with Starlink today that we immediately get to benefit from.”
The first-principles case, and he starts with regulatory: “people are already having concerns about ‘I don’t want this data center in my backyard.’” In space: solar cells get roughly five times if not more energy per cell, sun-sync keeps them lit 24 hours a day, no glass protection needed in vacuum, radiative cooling replaces liquid-cooling plumbing, and no land lease — “your cost is your satellite and your launch.”
The ex-semiconductor CFO’s cost-curve argument: “most of the cost is silicon… we’re benefiting from silicon cost reductions process node to process node, so our costs are going to go down. If you look at the terrestrial solutions, the curve is going the other direction — power bills are not going down, and land/regulatory is getting more and more challenging.”

4. The scale math: 200 launches per gigawatt, thousands of launches a year

Johnsen relays his engineers’ pushback verbatim: “You got to make sure people understand there’s a lot of work still to do” — putting gigawatts a year into space is “a very hard challenge.” His answer: SpaceX has already demonstrated scale in launch and in building thousands of satellites a year, and will “demonstrate capabilities as soon as next year.”
Today’s ratio: roughly 200 launches for every gigawatt — “first gen of the satellite and the rocket.” SpaceX is “capacitizing for thousands of launches a year right now”: two towers in South Texas, Cape Canaveral almost done, LC-37 within the next year, with more locations in discussion. Johnsen says the people supporting orbital compute think it’s a lot further away because you can’t do it without launch — the rapid-reusable launch Starship is going to provide.
Baker’s vision of the closed loop: ask Grok a question, inference runs on an orbital compute satellite, the answer comes down via direct-to-cell to his phone — “whenever it happens… I will personally find that very exciting.”

5. The AI business is already top-5 infrastructure: Anthropic, Cursor, Grok

The Anthropic hosting deal proves the orbital business model on the ground now: “there’s nothing better to prove out that business model than demonstrating it right now… with our terrestrial data centers.” Dense training clusters make it “premium compute,” internal models stay training on bleeding-edge GB300s, and he expects “more and more folks want to lean to us.” Baker’s napkin math: a $3.75B run rate per quarter — “50% larger than the company just outside the top four,” i.e., a top-five AI infrastructure business.
On Cursor — a candid admission of a change of course: “you saw us not be happy with where we were at on the enterprise solution on the coding specifically, and so we went out and did the deal with Cursor.” The team (likely Michael Truell — captions garble the name) brings over half the Fortune 500 and thousands of enterprise accounts; Baker notes Composer’s “very significant leap” after just weeks of mid-training RL on Colossus 2. With Grok as the LLM and Grok Build as the harness, “it’s pretty magical.”
The differentiator Johnsen keeps returning to: real-time X data feeding “a truth-seeking model… in the end a huge differentiator.” Integration is moving fast — “it’s now SpaceX AI. If you go up there at Palo Alto, it is SpaceX and you feel it already.”

6. Terafab: 100 GW/year requires assured silicon, and TSMC is the choke point

The motivation is defensive, not opportunistic: “when you start talking about Nvidia or the AI5 chip or a TPU, all of a sudden you start talking about TSMC… our concern is that the supply chain won’t be there for us to ramp to many, many gigawatts. Ideally 100 gigawatts a year is our target.”
The structure de-risks a normally impossible business: a SpaceX–Tesla partnership now bringing in Intel’s decades of know-how, with captive customers saying “we will take every wafer that you can yield out” — “then it takes that risk away and really the risk is a capital risk only.” Plus the Elon factor: “the entrepreneur innovator of our lifetime… sitting with process engineers that have 50 steps to their requirements list and pushing on each one.”

7. Elon’s step-by-step IP machine, and vertical integration that stays open

Johnsen’s 15-year view of how the mission became a business model — his Broadcom analogy: where a chipmaker acquires missing IP to win a socket, SpaceX builds it organically, step by step: orbit → reusable rockets → heavy lift → man carriage (Dragon) → comms in space → rapid reusable launch. When a Mars window opens, “you have a fleet of vehicles ready… and you didn’t have to have a crazy investment anymore.” In 2011, Mars talk drew eye-rolls; “nowadays when we say that, literally the response is ‘what year?’”
The culture behind Jensen’s verdict that Colossus 1 online in 122 days was “superhuman” and “only Elon could have done it”: Baker’s understanding is that Elon works whatever’s in the critical path directly with engineers; Baker believes Raptor was once the critical gating factor and describes a standing meeting late Sunday or Monday night for its engineers — “in the trenches with you and working probably harder than any employee that I know.”
On funding the capex ramp: capital allocation runs like just-in-time manufacturing — towers, air separation units, satellite factories, terrestrial data centers “all happening at the same time… you really have to map it out quarter by quarter and just not get ahead of yourselves.” Baker spots the sequencing: Starship flies Starlink V3 → unlocks cash-generative Starlink growth and presumably high-margin direct-to-cell → right as orbital compute needs racks in space. Johnsen: “I wish I could take credit for that, but that’s certainly Elon.”
The closing structure: vertically integrated, yet open at every layer — competitors can buy launch, use Starlink, rent terrestrial compute, and access the model. Each element standing alone as a business “just gives you more scale, further lowers costs, plays into the vertical integration.”

Market Overview: June 9, 2026 Palo Alto Networks CEO: "AI Found 5 Years of Bugs in 6 Weeks"