Blue Origin's $10 billion, and what external capital signals
The first surprise was Blue Origin reaching for outside money. Vlad's instinct was skepticism: Jeff Bezos has "extremely deep pockets," and finding $10 billion should not tax a man who, by Aaron's half-remembered figure, recently stood up something like a $100 billion fund for his AI ambitions. So why raise externally at all?
Aaron's read was that this is not a capital requirement so much as a stage-setting move. External capital brings governance and urgency, and it signals that Blue Origin is no longer an internal pet project. It also lays the groundwork for the far larger raises, or a public listing, that would follow if the company is nearing a genuine ramp in cadence. He had heard the rumor circulating for a while, so the news itself was no shock, even if the number struck him as "maybe a little light."
Vlad thought the $130 billion valuation was about right, and defended it against the internet's favorite jibe, that SpaceX is worth some $2 trillion while Blue Origin fetches only 130. Launch, he argued, is neither the main revenue nor the main profit engine, and until rockets are rapidly and fully reusable the margins simply are not amazing. The real money printer over the next five years is Amazon Leo, which is a separate business. Blue Origin, lacking SpaceX's pedigree, is priced accordingly.
One hundred thousand satellites
Then came the number Aaron has been quietly waiting on: a Starlink FCC filing for 100,000 satellites. For the better part of a year the hosts have argued that a figure like 100,000 satellites, or 100,000 terabytes of bandwidth in orbit, was coming, and they were "made fun of publicly" for it. The mockery had teeth when Gwynne, in an interview, floated a ceiling of maybe 20,000 satellites and no one pressed her on it; skeptics seized on the remark as proof from inside the company.
Vlad felt the vindication most acutely, because he owns the models. For roughly a year they had suggested that, on the cash Starlink can generate and the demand curve for broadband, 100,000 satellites around 2030 was reasonable, optimistic when first drawn but reasonable. Gwynne's 20,000 gave him "nightmares for days," a genuine fear that the model was badly wrong. The filing was, he said, "very reassuring and relieving." For Aaron, the figure matters less as a headline than as a tell about the scale of the market Starlink is chasing and the scale of investment it implies.
Starlink in the sky, and the net-new markets
The aviation story was smaller but revealing. Starlink Aviation raised hardware pricing from $145,000 to $200,000, with the average plan running around $10,000 a month per aircraft. Vlad did the back-of-the-napkin math: against roughly 30,000 active commercial planes, capturing a third would be about 10,000 aircraft, or roughly $100 million a month, call it a billion dollars a year. Real money, but "a drop in the ocean" beside the rest of Starlink, and better for consumers than for the bottom line. He suspects Amazon Leo will target aviation as a beachhead precisely because it is contained.
Aaron zoomed out to the sprawl of net-new demand, from the Starlink dog collar Vlad now covets for his poorly-recalling dog to humanoid robots, which he reckons could drive one to two orders of magnitude more token usage than a text token. Each market is a drop; the aggregate bends in one direction, back toward the 100,000-satellite logic.
The C rewrite: reading versus typing
The week's analysis carried a disclaimer Aaron insisted on: "We are not AI people." But SpaceX AI is now too large a part of SpaceX to wave at, so the pair have spent the last couple of months teaching themselves how LLMs work, applying their first-principles method and bracing to make mistakes.
The object of study was Elon's "C rewrite," announced two or three months ago: rewriting the AI models to map directly onto the GPUs, with a claimed 10x improvement over JAX, the software that ordinarily bridges an LLM to the hardware. Most observers shrugged; the hosts suspected it mattered more than it was getting credit for.
Vlad reached for an analogy. Reading, you take in whole words at a glance; writing, you go one letter at a time. Models are the same. Fed a prompt, a model can see much of it at once, but replying it emits one token at a time, each picked by probability, and that sequential decode is the slow part. Rewriting in C sharply speeds up every stage that must go letter by letter. Two stages qualify: pre-training, which their research found is a surprisingly small slice, and reinforcement learning, the large slice where the model is fine-tuned by being asked millions of questions and graded. It is there the savings land. In their model, an eight-week process becomes 5.3 weeks.
Aaron drew out the consequence: in theory a release cadence near three and a half weeks, with new models arriving roughly that often over the coming months, faster even than Elon's own timeline, which is a rare place for their model to sit. Better still, the effect grows with model size; pushing from an estimated two trillion parameters toward twenty trillion widens the reinforcement-learning decode savings. He flagged that it helps inference too, a thread they will pull next week when they turn to inference on the AI satellites. The tell, he added, is that SpaceX employees reacting to the fresh Grok 4.5 release are not talking about 4.5 so much as how much faster the iterations behind it are coming.
A guest who zooms out: Peter Hague
To break from what Vlad called a recent fixation on "short-term terrestrial stuff," the hosts welcomed Peter Hague, an astrophysicist by training and one of the most cited independent voices on Starship and human expansion into the solar system. He came to the field in 2015, when SpaceX first landed a Falcon 9 booster and he decided the company was "not just talk." A long-promised book sits on the back burner in favor of his blog, Planetocracy, and a scientific paper in which he coined the term "mass value" to quantify the economic development of the solar system.
Orbital refueling on one side of paper
Vlad opened with the bogeyman: orbital refueling, which skeptics call spooky and unproven, and which Starship needs to reach anywhere interesting. Hague was calm. Propellant transfer in orbit has been routine since the late 1970s, when the Soviet station Salyut 6 first received a Progress capsule, essentially a Soyuz with a fuel tank where the descent module would be. The difference is that those propellants are storable at room temperature in small tanks, coaxed out with an inflatable bladder.
SpaceX and Blue Origin face the harder version. On Earth, gravity settles liquid over the pump; in space the fuel floats in globs and must be forced to the bottom of the tank. Hague walked through the Apollo precedent, where cameras inside the Saturn V's S-IVB stage studied the problem during three-hour parking orbits: the stage vented boil-off through rearward nozzles for gentle settling thrust, then fired ullage rockets for a second before reignition. SpaceX's task is different again, transferring an entire tank's contents into another ship slowly, over hours, which means sustaining some substitute for gravity the whole time, whether by venting cold gas through a nozzle or spinning the spacecraft.
His verdict was reassuring. "You could sort of write a plan to do the orbital refueling on one side of paper." The fundamental problem is simpler than thermal protection, and it is the kind of thing SpaceX iterates on quickly, because two docked ships can loiter in orbit, back off, retry, even upload new software, with multiple shots per mission before the propellant boils away. He expects it solved in the second half of this year or next. The genuine long pole, he argued, is the thermal protection system, because every TPS test costs a full Starship build and reentry, roughly a two-to-three-month cadence, with one destructive shot to learn from. Mars reentry will differ again, faster and into a carbon-dioxide atmosphere that splits molecules and changes the plasma, but he would bet that consistent reentry on Earth translates to Mars.
Cryogenics, methane, and a sunshade
Aaron pushed on cryogenics and boil-off. Hague noted SpaceX's quiet advantage: methane, not the fussy liquid hydrogen that embrittles metals, leaks through every crack, and plagues SLS and Delta IV with valve trouble. Apollo lost about 10 percent of its hydrogen in three hours but only a percent or two of its oxygen; methane sits near oxygen's temperature and behaves. The deeper reason, he said, traces to Elon reading The Case for Mars: methane can be manufactured on Mars from atmospheric CO2 and hydrogen, yielding water to electrolyze for oxygen. Boil-off was a lesser motive.
For a quick refuel, losing a few percent hardly matters. For long-term storage it does, and the answer is a sunshade, a larger cousin of the James Webb telescope's fold-out multilayer shield. Hague let himself wander into a favorite project: making the case for the UK and Europe to build a low-Earth-orbit habitat in a dusk-dawn sun-synchronous orbit, permanently sunlit, its shaded far side a natural home for a propellant depot.
Orbit, Moon, or Mars
Asked to weigh the three destinations, Hague professed himself "agnostic," glad that Bezos aims at free-floating habitats while Elon aims at Mars. Mars is the most Earth-like body we have; its thin atmosphere still buys real radiation and micrometeorite shielding, though bases will want regolith piled on the roof against the cosmic-ray cone from overhead. The open question is gravity. We do not know how much is enough for long-term health, and critics assume the unknown is bad. Hague resists that: ISS centrifuge experiments on mice found Martian gravity preserved muscle mass while shifting fiber type. The real unknowns are pregnancy, raising a child, and whether someone raised on Mars could return to Earth, concerns that bite harder on the Moon at half the gravity. Rotating habitats sidestep it entirely by dialing in whatever gravity you like, a gentle 0.8G for the elderly or a punishing 2G "gym bro" module. Mars offers nearly the whole periodic table; the Moon, rich in metals but starved of volatiles, is the least desirable, redeemed only by being three days from home, which is exactly why NASA's Artemis plan treats it as a place to practice.
Perchlorates, polemics, and living well on Mars
Aaron raised the perchlorates that critics wield as a death sentence. Hague was unimpressed: perchlorate is a soluble, reactive oxidizer that disrupts the thyroid but not permanently, and it washes out of regolith. Settlers will not commute dome to dome in suits; they will live inside large pressurized volumes where the breathing gas is a structural element, and the engineering challenge inverts to holding structures down against pressure rather than up against weight. Perchlorate is kept out of the food by clean-room hygiene, detergent foot baths borrowed from farms, and spot testing. He reserved real scorn for the book A City on Mars, which he called a study in "polemic mode," talking up problems while ignoring or downplaying their known solutions "to try and stop people trying."
The case for going out
Vlad asked Hague to make the case for spending money on any of this. His answer was cosmic in scale: the sun's energy reaching Earth is outweighed by what flies off in every other direction by a factor of about two billion. Expand into that, and you could raise per-capita energy use a thousandfold and the population two-thousandfold, 16 trillion people living as only billionaires do today, still drawing a mere thousandth of the sun's output. The society that results, he argued, will be built entirely by those who choose to participate; the rest, like the Tanzanian valley where humanity first emerged, simply will not matter to the future. His UK pitch is a "Penny for Space," one percent of government spending, less than most people put toward a pension and, in his framing, a civilizational one, enough to double the rest of ESA combined. Aaron folded it into his own theme of abundance, with orbital data centers as the near-term stepping stone that turns the sun's wasted energy into value beamed home, and the frontier itself as a bet on human ingenuity that "points in one direction, and that's out."
What to watch next
The hosts closed on the imminent Starship flight. Hague doubts it slips to next week only if everything goes smoothly, but reads the tightening interval between ship fire, booster fire, and launch as SpaceX finally finding its pace; Aaron expects a week's slip at most, not a month. Vlad thinks Flight 13's goals mirror Flight 12, a booster and ship splashdown, with a ship catch and orbital attempt more likely on 14 or 15, a sequencing Hague notes depends entirely on how 13 goes. He is watching the booster's violent flip, where sloshing propellant can starve an engine and trigger the kind of propagating failure seen last time, a problem that ties straight back to the refueling discussion and raises questions for Raptor V3. Vlad, for his part, just wants to see that flip again and a heat shield that comes home less charred than before. Aaron signed off by pointing listeners to the Martian office renders serving as the hosts' video backgrounds, downloadable on X, because a good Martian future should be one worth living in, trees included.