E239｜SpaceX Wants to Take Space Compute From Science Fiction to Reality—but Does the Math Work?

Summary

• The episode frames SpaceX’s June 2026 IPO at $1.75T, putting it among the world’s top 10 technology companies, and packaging its rocket business, Starlink and xAI into a single S-1. The real question is not whether space compute can be built, but whether it can be deployed economically—“Putting compute in space can definitely be done; making it economical to put compute in space is the real problem to solve,” as 刘冰雁 puts it.
• SpaceX’s first internal milestone is 1 GW of space compute: roughly 10,000 100 kW, 1-ton “AI minis,” with 100 units per Starship, requiring 100 launches and about 1 million square meters of solar panels. Lewis Hong would bet his own money on reaching it by 2029; 刘冰雁 thinks 100 Starship launches in a year would “not be a problem” by then, but does not believe the data-center product, thermal ecosystem and economics will mature in sync—“I wouldn’t even bet on when.”
• The launch-cost debate spans several orders of magnitude: a ground-based 1 GW data center costs about $50B, with GPUs accounting for 50%-60%; the episode cites SpaceX’s actual February 2026 rideshare price of $7,000/kg. 刘冰雁 uses roughly $200/kg as a conservative assumption, while Lewis sees $200-$500/kg in the near term and a long-term target of $10-$20/kg. Fuel and solar-panel mass alone imply a 2-month electricity payback; realistic launch costs could extend that to 2 years, before 40% redundancy, satellite manufacturing and halved GPU life are added.
• Space is not naturally cool. Vacuum has no convection, and dissipating 1 MW of waste heat requires about 1,200 square meters of radiator area; the International Space Station’s 422-square-meter ammonia radiator system handles only about 70 kW. The good news is that blackbody radiation rises with the fourth power of absolute temperature, so more surface area, active heat transport or a modestly higher operating temperature can solve the problem; the bad news is that solid-state heat pumps, purpose-built chips and the broader supply chain have not cleared the cost threshold for scale.
• Lewis sees orbital data centers as better suited to inference than large-scale training: a 100 kW unit can hold about 100 cards, while inter-satellite links would magnify training’s bandwidth bottleneck. Starlink’s roughly 20-40 millisecond latency in low Earth orbit is slower than the 3-5 milliseconds available over terrestrial fiber, but not enough to impair inference for this generation of LLMs. Radiation is also less of a hard constraint than thermal management: shielding, ECC and redundancy already offer workable paths, and Lewis points to more than 12,000 Starlink satellites with 99.85% still operating normally.
• The Kessler effect is more a governance risk than a near-term capacity constraint: uncontrolled low-orbit satellites gradually reenter, and co-orbital traffic reduces collision risk; Starlink still performed about 300,000 orbital adjustments in 2025. Halving orbital spacing theoretically increases 3D capacity eightfold, but 刘冰雁 is pessimistic that humanity will establish a genuinely effective international coordination mechanism before a collision with serious consequences forces the issue.
• For investors, SpaceX’s launch, satellite, communications and materials R&D capabilities create a clear cost moat. Lewis thinks an independent “space cloud” startup would struggle to generate the 100x returns venture capital requires, with costs potentially at least 3x SpaceX’s. The more investable opportunity lies in platform-agnostic components—semiconductor solid-state cooling, thermal management and the satellite supply chain—that can sell to SpaceX and to future operators.
• The long-term bull case is not saving a few kilowatt-hours today, but the possibility that terrestrial power approvals, heat islands and environmental externalities become much larger constraints after compute grows 10x or 100x. Space can keep power generation and waste heat off Earth. At the same time, 刘冰雁 sees the Moon as a more credible economic asset than Mars: low gravity, mining and deep-space manufacturing could benefit today’s shareholders, while Mars is closer to a civilizational backup and symbolic milestone—“dreams can’t feed you, but dreams matter.”

Deep dive

1. Starbase Turns “Mass-Producing Starships” From a Slogan Into an Industrial Scale

• Lewis returned to the Starbase of 2018—then a mix of tents, Airstream trailers and abundant wildlife—and found headquarters buildings, 2 Mega Bays and rows of Cybertrucks. Eight years of buildout convinced him that SpaceX is not merely producing a giant rocket; it is building a mass-production line.
• Outside SpaceX’s Los Angeles headquarters sits the actual rocket first successfully recovered on December 21, 2015; Starbase gives “1,000 launches in the future” an industrial scale. Lewis says standing beside a system 40 stories tall makes people feel “very small,” while Starship is ultimately meant to become a routinely operated form of transport, like a 747.
• To Lewis, the IPO is merely one checkbox. Inside SpaceX, the focus is a development roadmap for the next 20-30 years, not the listing itself. The base even has a spot reserved for a “Mars embassy,” a half-joking marker that captures the company’s long-term culture.

2. The S-1 Is Selling a Space-Compute Platform, Not Just a Rocket Company

• Under the episode’s setup, SpaceX goes public in June 2026 at $1.75T, packaging rocket launches, Starlink and xAI into the same prospectus. Lewis says the direction is unsurprising: low Earth orbit is a new computing platform, and Starlink is merely its first validated application.
• His logic is that any ground-based service tied to large-scale computing could theoretically be redeployed in orbit; once launch costs fall far enough, communications, positioning, Earth observation, real-time databases and AI could all run on the same platform. Platform value comes before any single application, though Elon Musk’s timeline for delivering it remains open to question.
• The host offers a darker interpretation: SpaceX’s capital-markets story is shifting from a remote launch business to the nearer-term space internet, then attaching the hottest AI label to it—“who wouldn’t be envious?” 刘冰雁 adds speculation about cross-shareholdings between Google and SpaceX, saying at minimum it is hard to believe the relevant votes are entirely heartfelt. Lewis acknowledges that space data centers were not a specific plan at the company’s founding, but insists the underlying mission of “opening up the universe” has not changed.

3. The 1 GW Plan Is Essentially 10,000 Distributed AI Satellites

• SpaceX’s current baseline is an “AI mini” rated at about 100 kW and weighing roughly 1 ton; 1 GW therefore requires about 10,000 satellites. With 100 satellites per Starship, that means 100 launches—not a single 5-kilometer-by-5-kilometer orbital megastructure.
• The energy system would require about 1 million square meters of solar panels. Elon Musk has proposed dawn-dusk orbits, keeping the satellites near Earth’s terminator and limiting time in Earth’s shadow to about 35 minutes a day to conserve power.
• 刘冰雁 warns that 10,000 units kept tens of kilometers apart would nearly fill one dawn-dusk orbital band. Space is 3D, so additional orbital layers or battery-backed operation in other orbits remain possible, but prime orbital slots are not unlimited.

4. Lewis Bets on 2029; 刘冰雁 Will Bet Only on Launch Capacity

• Starship currently launches only a single-digit number of times per year, but Lewis stresses that these missions are designed for iteration rather than launch rate. V3 may test more recovery and landing steps and begin sending some payloads up; he expects a mature version could follow the Falcon 9 path to V5, with V4 roughly twice the scale of V3.
• Falcon 9 took roughly 10 years to reach high-frequency reuse, while Starship has reached a similar development stage in 5 years. Lewis therefore expects the ground and orbital operating systems to scale together once “launch once and recover” is reliably unlocked, with annual launches potentially moving from 5 or 10 to 50 and eventually hundreds.
• Lewis’s bet is 1 GW by 2029, with 2030 as the conservative option: “If I were betting my own money, I’d bet on 2029.” 刘冰雁 counters that 100 Starship launches in a year in 2029 or 2030 would “not be a problem,” but sees a huge gap between mass-producing AI satellites, solving thermal management and building a customer ecosystem.
• The deeper disagreement is on demand. 刘冰雁 sees no compelling need to move data centers into space because a 10x improvement in chip efficiency would also benefit ground-based systems. Lewis sees energy, permitting speed and scalability as long-term variables that could favor orbit.

5. The “2-Month Payback” Works Only Under an Extremely Narrow Fuel Assumption

• Building a 1 GW ground data center costs about $50B, with GPUs accounting for 50%-60%; electricity is mainly an operating cost. 刘冰雁 therefore simplifies the comparison to whether the fuel required to launch the solar panels can be offset by the electricity they later generate.
• Assuming orbital solar generation is about 5x as efficient as ground-based generation, the payload fraction is 3%-5%, and only fuel is counted—not the panels themselves or other equipment—he calculates an energy payback of about 2 months. On that narrow energy basis, the economics appear to work as long as the facility lasts much longer than 2 months.
• Lewis agrees that launch capability, lift capacity and launch cost may not be the ultimate hard constraints, but notes that treating fuel as nearly the entire cost requires Starship to be fully reusable. If it can recover only the first stage and reuse each vehicle 10 times, as with early Falcon 9, the conclusion changes materially.

6. Real-World Launch Prices Push the Payback Period From 2 Months Toward 2 Years

• 刘冰雁 uses about $200/kg as a conservative assumption; Lewis sees $200-$500/kg as realistic in the near term and $10-$20/kg as a possible long-term target once the market is large enough and operations are scaled. 刘冰雁 adds that a fuel-only calculation could put the cost below $100/kg.
• As a physics sanity check, Lewis compares the roughly 10,000-kilometer China-US air route, where the cost is about $3-$5 per kilometer, and says claims that put Starship’s cost below $3-$5/kg are not credible.
• The host adds a real-world anchor: SpaceX’s February 2026 rideshare quote was still about $7,000/kg. Moving from $7,000 to $200 and then to $20 depends on V4, V5, full reuse, high-frequency launches and sustained market demand all arriving together.
• If the all-in cost in the 2-month model deteriorates by an order of magnitude, the payback period could stretch to roughly 2 years; whether a space data center can last 2 years is itself uncertain. 刘冰雁’s view is that even the solar-panel mass alone may not pay for itself, let alone the structure, chips and thermal equipment.
• The host also adds about 40% redundancy for the initial fleet, assumes radiation and unknown environmental effects cut GPU life to half that of ground-based systems, and then doubles the cost again.

7. GPUs Are Not the Retrofit Problem; the Unknown Satellite Configuration Is

• Based on the total cost of a 1 GW ground data center and the GPU share, the chips themselves are worth about $25B-$30B and do not disappear simply because they are sent into orbit. SpaceX wants each satellite to cost roughly the same as Starlink V3, or about $1M-$1.2M; GPU costs are separate.
• Lewis considers radiation shielding, ECC and roughly 10% additional checking redundancy to be familiar solutions within existing computing architectures; each added check bit can reduce error probability by multiple orders of magnitude. Adapting chips for space is not the biggest obstacle.
• The unresolved issue is the full configuration: how to combine solar panels, radiator area, radiation shielding, attitude control and inter-satellite communications. The host expects multiple companies to iterate toward similar designs; 刘冰雁 says it is too early to discuss retrofit costs precisely while the form factor remains unknown.

8. Space Is Cold, but It Is Also a Vacuum Thermos With No Convection

• 刘冰雁 first corrects the most common intuition: the movie scene in which everything freezes instantly when an airlock opens is not how it works. With only a few hundred atoms per cubic meter, vacuum has virtually no convection, so waste heat can leave only through electromagnetic radiation.
• The engineering scale cited by the episode is about 1,200 square meters of radiator area per 1 MW dissipated—roughly 4 or 5 tennis courts. The International Space Station’s 422-square-meter ammonia-loop radiators handle only about 70 kW. Space cooling is not free; it may determine the satellite’s entire shape.
• Blackbody radiation rises with the fourth power of absolute temperature, but the improvement across ordinary operating temperatures is less dramatic than intuition suggests: moving from 25°C, or roughly 300 K, to 60°C, or roughly 330 K, increases radiative capacity by only about 50%. The system still has to continuously reject all energy absorbed and generated by the compute load.

9. Surface Area and Temperature Together Set the Thermal Balance of Orbital Facilities

• Low Earth orbit receives about 1.3 kW of solar radiation per square meter, while Earth reflects roughly 20% of that energy. Absorption is concentrated on the sun-facing side, but radiation can use every exposed surface, making the geometry itself a thermal-design decision.
• 刘冰雁 offers an intuitive comparison: one-sided heating on the lunar surface can approach 120°C; a thin sheet absorbing heat on one side and radiating from both sides equilibrates at about 60°C; a triangular configuration with 3 radiating sides reaches about 26°C; and a sphere has 4x the surface area of its projected area, giving Earth a theoretical equilibrium temperature of about 5°C.
• The challenge is not merely having enough radiator area, but moving heat away from the GPUs to distant surfaces. A sphere filled with circulating water is physically feasible but economically useless because of its mass; the cost of transporting heat is what determines whether hard science fiction becomes a product.
• 刘冰雁 says his own view changed: “Then I thought about it and realized, yes, Earth isn’t too hot to tolerate either.” He moved from doubting that space cooling was feasible to concluding that it is “not a myth,” while still seeing no established economic path.

10. Heat Pumps and High-Temperature Chips Are Two Routes, but Maturity Remains in Dispute

• The current approach uses liquid-ammonia loops and mechanical thermal pumps to move energy from the hot side to radiators; a more aggressive design would actively move heat from a GPU at about 50°C to a radiator surface at about 90°C. Raising radiator temperature takes advantage of the fourth-power relationship and can materially reduce required area.
• 刘冰雁 is particularly interested in semiconductor solid-state cooling: an electrically driven “solid-state air conditioner” with no compressor. Products can already serve phone-scale loads, but the cost, weight and power density required for data centers remain at an experimental stage and may require an entirely new supply chain.
• Lewis adds that Starlink has already accumulated data on small-scale cooling pumps and thermal management, so SpaceX is not starting from zero. NVIDIA, AMD and the Terafab envisioned by Elon Musk could also optimize chip temperatures. Raising the operating temperature by only a few degrees could produce meaningful radiator-area savings in orbit.
• The host had initially interpreted “raising chip temperature” as pushing toward 120°C and assumed that might force the semiconductor industry to revisit its materials stack. Hearing that Lewis meant only a few degrees softened some of his skepticism, but he still doubts suppliers would absorb major adaptation costs if the orbital market reaches only 1% of the overall compute market.

11. The Sweet Spot for Orbital Compute Is Inference, Not Large-Scale Training

• Lewis’s clear view is that space is “absolutely unsuitable” for large-scale training: 100 kW corresponds to roughly 100 cards, enough perhaps for small fine-tuning, but large training requires high-bandwidth, tightly coupled interconnects, and inter-satellite bandwidth would multiply the cost.
• Lewis places the product between a hyperscaler and the edge: each unit is large enough to be meaningful but close enough to users to support inference or online learning. Large training stays on the ground while massive inference workloads move into orbit.
• He also points out that a successful model should generate hundreds or thousands of times more inference than training; otherwise the model itself is “too much of a failure.” Space can therefore be unsuitable for training without losing the largest demand market.
• Latency is not the key weakness: terrestrial fiber to a nearby data center takes about 3-5 milliseconds, versus roughly 20-40 milliseconds over Starlink, while people are generally insensitive below 200 milliseconds. The host adds that for current models, compute time may itself be longer than link latency.

12. Low Orbit and Error Correction Turn Radiation Into a Manageable Failure Mode

• High-energy particles can strike GPUs and cause computation errors, but Lewis thinks users may not notice small effects in inference workloads. More important is choosing an orbit below 1,000 kilometers, where Earth’s magnetic field provides stronger protection.
• Training demands strict consistency, making radiation errors more troublesome; inference can use shielding, ECC, software checks and satellite redundancy. Higher orbits are not only more radioactive, but also more difficult to make economical in launch and maintenance costs over decades.
• Lewis points to Starlink’s operating record: more than 12,000 satellites launched since 2019, with about 99.85% still operating normally. Roughly 0.15% failed and reentered as planned; only 1 early satellite was actually struck and lost control. That does not prove GPU longevity, but it suggests radiation is not the primary risk to the satellite business model.

13. Debris Risk Can Be Mitigated, but Governance May Take an Accident to Emerge

• Satellites in dawn-dusk orbits generally travel in the same direction, and low-orbit equipment that loses station-keeping gradually decays and burns up, making it more controllable than high-orbit debris that can remain for thousands of years. 刘冰雁 accepts that the Kessler effect could occur, but thinks its near-term probability is lower than intuition suggests.
• The host says Starlink completed about 300,000 orbital adjustments in 2025. 刘冰雁 had also noted that Starlink can autonomously detect obstacles and change orbit, showing that SpaceX has built a continuous-operations system. Lewis therefore asks orbital startups: “Why won’t SpaceX do this? Why will you do it better than SpaceX?”
• Earth and orbital space are vastly larger than the current population of satellites; today’s 10-kilometer-scale spacing could eventually shrink to 1 kilometer or a few hundred meters. Halving the spacing theoretically increases 3D capacity 8x, and technological progress can partly offset the scarcity of prime orbits.
• 刘冰雁 is more pessimistic about governance: every operator will want to leave a little less safety distance and occupy a little more capacity until a serious collision forces cooperation. Lewis stresses that the physical environment will compel political rivals to keep communicating because “if one side screws up, everyone screws up.”

14. The Conclusion Is Not “Impossible,” but That No Self-Reinforcing Economic Path Exists Yet

• The host’s interim conclusion is that launch, cooling, radiation and collision risk contain no clear physical prohibition; but every added condition—manufacturing, lifespan, redundancy and maintenance—pushes the original 2-month payback further out. The broadest point of agreement is that space compute is physically feasible but not yet economically viable.
• 刘冰雁’s precise formulation is worth preserving: “Putting compute in space can definitely be done, no question. Making it economical to put compute in space is the real problem to solve” (把算力能够经济地放在太空). He does not know whether an investment-cost reduction-reinvestment feedback loop will emerge.
• The bullish variable is that the environmental externalities of ground-based cooling are not fully priced in. If compute grows 10x or 100x, data-center heat islands could become a climate problem; Lewis says terrestrial heat rejection has a meaningful upper limit, while space-based radiative cooling theoretically does not face the same limit.
• Electricity and permitting could also reverse the comparison. Over the long term, power, manufacturing costs and approval timelines for ground-based data centers will become increasingly difficult. China is “temporarily” not short of electricity, but nobody knows whether the next wave of demand growth will be 20% or 200x.

15. SpaceX’s Real Moat Is Speed and Vertical Integration

• SpaceX is advancing Starship, satellite production, Starlink communications and parts of its solar and materials R&D in parallel, turning its expansion rate into an internal variable. Lewis’s view in the AI race is that compute built today is worth more than compute built tomorrow.
• Once Starship reaches stable reuse, launch, ground operations and satellite manufacturing can all enter exponential scaling together; that is the system condition other space-data-center companies lack. Even superior chip performance cannot easily offset the cost gap created by having to buy launch and communications capacity.
• Orbit is not an unlimited resource, but neither is it a case of first come, permanent ownership. Safety distances, inter-satellite control and orbital layers can continue to improve, and every operator must coordinate with other countries; first-mover advantage is not absolute property rights.

16. Google and NVIDIA Validate the Technology, Not Necessarily the Commercial Demand

• The host mentions the Google-SpaceX partnership; 刘冰雁 says he has seen companies put NVIDIA H100s into orbit. He considers such experiments relatively straightforward: about $1.5M can put 150 kg into orbit, and even a phone weighing a few hundred grams could potentially operate there for several hours.
• For 刘冰雁, sending a chip into space proves only that it can be done, not that it is worth doing. He also speculates about cross-shareholdings between Google and SpaceX and does not want to treat the spending automatically as an independent market vote. A technology test is not demand validation.
• Lewis adds that communications, GPS, observation and computing were always extensions of the same orbital platform; AI is simply a later application, not evidence that SpaceX has abandoned its original mission. The S-1 may downplay the distant space economy because investors are unwilling to pay for a 100-year narrative.
• Lewis does not expect SpaceX to remain the only launch supplier forever. 刘冰雁 says China has at least 3 different companies working on reusable rockets, including Zhuque and state-backed projects. Space data centers could then use the best available rocket rather than relying exclusively on SpaceX.

17. An Independent Space Cloud Is Unlikely to Be a 100x Company; Components Look More Investable

• Lewis is not optimistic about startups trying to replicate SpaceX head-on. Starcloud, a YC company, initially envisioned a 5-kilometer-by-5-kilometer GW-scale space station; even after raising about $170M recently and shifting to a distributed architecture, its funding remains insufficient for the system-level challenge.
• Without its own rockets, satellite factory and communications network, an independent operator could cost at least 3x as much as SpaceX. Such companies might serve niche markets, partner with terrestrial firms or get acquired, but Lewis sees no 100x or 1,000x venture path.
• He considers Blue Origin the only potential US number 2 with both long-term capital and accumulated technical capability, though it remains nowhere near SpaceX’s scale. Founders building orbital collision-avoidance services would also have to overcome SpaceX’s advantage in operating data.
• The more sensible investment targets are platform-agnostic technologies such as solid-state semiconductor cooling, thermal management and satellite components. Unlike rockets, satellite supply chains do not have to be localized and can source globally; selling shovels is more realistic than building a space AWS.

18. Returning to the Moon Is a Political Project and a Testbed for the Mars Route

• The host describes NASA administrator Jared Isaacman as Elon Musk’s “private astronaut” and a traditional Mars advocate who nevertheless began pushing a lunar base after taking office. Lewis sees Isaacman as extremely efficient and pragmatic; the shift may initially reflect Donald Trump’s desire to leave a Kennedy-style lunar milestone during his term.
• The episode also mentions SpaceX’s plan to launch a lunar lander in March 2027 and compete with Jeff Bezos’s camp for NASA contracts. Lewis’s interpretation is that if the government is willing to pay for the Moon, SpaceX will naturally use the funding to accelerate longer-term technology.
• In Lewis’s view, the “Moon camp” and the “Mars camp” are not competing routes: the Moon is Mars’s port and Gateway. Its proximity to Earth makes it suitable for iterating life support, engines, landing and long-duration survival technologies, as well as serving as an intermediate node for deeper exploration.
• The difficulty of reaching Mars is not just getting there, but shortening the trip, decelerating and keeping people alive. The Moon therefore serves political display, engineering validation and near-Earth economics at the same time.

19. The Moon Could Generate Cash Flow First; Mars Still Has No Economics in the Chemical-Rocket Era

• 刘冰雁 and Lewis disagree less on whether the Moon is useful than on how much its value should be weighted. 刘冰雁 believes that if Starship pushes launch prices down to $10-$20/kg, lunar development could generate substantial returns for today’s shareholders; Mars remains a destination with little economic value.
• He cites helium-3, deep-space power stations, space data centers and mining as possible lunar industries: the Moon could become a raw-material and manufacturing base. Its gravity well is “almost as if it weren’t there,” making the fuel required to send cargo from the lunar surface into orbit far lower than launching from Earth.
• The clearest comparison is that the Saturn V used for the Apollo landings weighed about 3,000 tons, while the hardware needed to carry astronauts from the lunar surface back to lunar orbit was only “the size of a trash can,” essentially a lunar three-wheeler. No matter how cheap SpaceX becomes, launching outward from the Moon could still be many times cheaper.
• Lewis accepts the Moon’s practical value but insists it ultimately serves farther exploration: “These two things don’t really conflict; they’re on the same line.”

20. Mars Is More a Civilizational Symbol Than a Habitable Asset This Century

• The host summarizes the two camps’ resource philosophies: the Moon lacks carbon, nitrogen and easily extractable water, while its oxygen is locked in rock; Mars at least has a carbon-dioxide atmosphere, water ice and nitrogen, making it closer to the chemical system required by carbon-based life. The Moon is a resource stepping stone; Mars is a civilizational backup.
• 刘冰雁 says the two operate on entirely different timelines. Based on the episode’s discussion, current Mars travel takes at least 1 year; cutting it to 1 month would require nearly ignoring gravity, traveling in a straight line at high speed and decelerating quickly at the destination. Stopping is as difficult as accelerating.
• On terraforming, he offers a scale warning: humanity can engineer roughly 1 cubic kilometer of material per year, while the target is an entire planet—a gap of many orders of magnitude. Elon Musk has discussed using nuclear weapons to melt Martian ice, but 刘冰雁 does not consider that an engineering project worth discussing within a human lifetime.
• 刘冰雁 assigns Mars symbolic value as a first step: if humanity cannot reach even the nearest target planet, “opening up the universe” is empty talk. More than 500,000 people reportedly said in an early SpaceX questionnaire that they would give up everything on Earth for a one-way ticket; even if the first person to land never returned, it would still be a species-level milestone.

21. Space Compute Ultimately Runs Into Sovereignty, Resources and the Question of How to Live

• Lewis sees orbital data centers potentially providing hard-to-intercept data links and a new form of data sovereignty; 刘冰雁 immediately warns that space is not beyond the law, only beyond current enforcement capacity. Who can enter, inspect or destroy a facility will ultimately reshape jurisdiction.
• Open questions include whether national sovereignty extends 20-30 kilometers above the ground; which authority governs a satellite that circles Earth 15 times a day; and whether data belongs to the company’s place of registration, the country of the terrestrial user or the orbital operator. Lewis says they are at least considering these issues from an investment-risk perspective.
• The host raises asteroid mining as a more distant economic opportunity. 刘冰雁 says low Earth orbit is “closest to the money,” but once a vehicle escapes Earth’s gravity well, the incremental energy required to reach higher orbits and near-Earth asteroids is not as large as intuition suggests. The minerals from a single asteroid could potentially support Earth’s demand for years, even though the engineering remains unresolved.
• The 2 guests ultimately bring the grand narrative back to controllable gravity and hot pot in space: artificial gravity that supports eating, using the bathroom and long-term living would unlock truly crewed space stations. The Moon landing was once “going to the Moon for the sake of going to the Moon”; today’s conclusion remains that “dreams can’t feed you, but dreams matter” (梦想不能当饭吃，但是梦想很重要).