The Lens · Deep Dive

Elon Solved the Rocket. Here Are the 9 Problems He Hasn't.

Transport is the part Elon is closest to solving — not the missing piece. The real walls between SpaceX and a Mars city (or data centers in space) are biology we've never measured, industry five-to-nine orders of magnitude beyond today's demos, and an economy that doesn't close. Here's the honest map, with who's actually working on each. Hover (or tap) any underlined term.

Dragonfly Lens · June 15, 2026 · ~18 min read · Built to be a reference, not a hot take. Every claim flagged science (a genuine unknown) or engineering (a scale problem).

The 60-second version

What's inside 1. The premise flip 2. The Life walls 3. The Industry walls 4. The Compute & Economics walls 5. Four timeline scenarios 6. The problems nobody's pricing 7. The opportunity lens (fixes + moonshots) Our read

1. The premise flip: transport is the easy part

The common intuition — "Elon basically has what he needs, he just needs a better way to get there" — is backwards. Transport is the part SpaceX has pushed furthest. Mass drivers and rail guns (a popular "better way") are real, but they solve a different problem: flinging cargo off the Moon's low gravity for lunar export — they can't cross interplanetary space, and the g-forces would liquefy a human. The genuinely unsolved work is what happens after the ship lands, and it splits cleanly into three buckets.

The honest meta-finding from ~200,000 words of primary-source research: the binding constraints are either scientific unknowns that resist rapid iteration (you can't blow up a closed ecosystem 50 times to learn, or speed-run a pregnancy in low gravity), or engineering gaps of 5–9 orders of magnitude from what's been demonstrated. SpaceX is brilliant at transport and energy-access. The open layers are Life, Industry, and Economics.

2. The Life walls mostly science

Problem 1 — Closed-loop life support science

The wall: we can recycle ~98% of water and most air with machines, but no one has ever closed the food + full-waste loop with living organisms and kept it stable for years with no resupply. The ISS recycles ~98% of its water and 0% of its food.

Who's working on it: ESA's MELiSSA consortium (running since 1989), NASA (Veggie / Advanced Plant Habitat), China's Lunar Palace 1 (Beihang Univ.), startup Interstellar Lab.

The cautionary data point: Biosphere 2 saw oxygen fall from 20.9% to 14.5% (uncured concrete absorbed the CO2, a one-way trap), crops failed, and the crew split into factions — in Earth gravity, with sunlight, for only two years. MELiSSA has run 35+ years and still uses ~40 rats with partial food. China's "98.2% closed" headline had 1.8% resupply by design. Food is the unsolved part, and it's biology, not hardware. verified

Problem 2 — Partial-gravity biology (especially reproduction) science

The wall: nearly all our health data is from 0g (ISS) or 1g (Earth) — there's essentially zero long-duration data at the in-between gravities of Mars (0.38g) or the Moon (0.16g), and no mammal has ever been conceived, gestated, and born off-Earth.

Who's working on it: JAXA (the ISS centrifuge — the best partial-g mammal data), Teruhiko Wakayama (Univ. of Yamanashi, space reproduction), NASA's Human Research Program + TRISH, and startup SpaceBorn United.

The newest result cuts against Mars: a March 2026 JAXA study found 0.67g preserved mouse muscle but 0.33g did not — implying Mars's 0.38g may sit below the protective threshold (24 mice, 28 days; read cautiously). Reproduction is at TRL 1–2: embryos have reached the blastocyst stage in 0g, but fertilization happened on Earth, and nothing tests implantation, gestation, or birth. SpaceBorn's 2025 flight carried yeast, no reproductive material. The field's own experts: "decades — and we don't yet know if it's possible." verified

Problem 3 — Radiation engineering (storms) science (chronic GCR)

The wall: galactic cosmic rays deliver ~1 sievert over a Mars round-trip and can't be fully shielded — and the biological effect on humans is known only to within hundreds of percent (mostly from rodents).

Who's working on it: NASA's Human Research Program + the Space Radiation Lab at Brookhaven, StemRad (the AstroRad vest, flown on Artemis I), ESA (which calls it "the radiation showstopper").

The number that matters: Curiosity's RAD instrument measured a full Mars mission at ~1.01 sieverts — which exceeds NASA's new 600 mSv career limit before the return trip even begins. Solar-storm shelters are near-solved engineering; chronic GCR is a genuine biological unknown, and counter-intuitively, thin shielding makes it worse (heavy ions fragment into secondary neutron showers). verified

3. The Industry walls mostly engineering / scale

Problem 4 — In-situ resource use (ISRU) at scale engineering

The wall: making propellant, water, metals and building material locally — because you can't ship a million people's supplies. ISRU has never been demonstrated above kilogram scale off-Earth.

Who's working on it: NASA (MOXIE oxygen; PRIME-1 ice drill), Sierra Space + Helios (regolith→metal), Redwire / ICON (construction), Interlune / Lunar Outpost (excavation).

The scale gap is staggering: the best Mars demo ever — MOXIE made 122 grams of oxygen total (~10 hours of one person breathing), about 200,000× below the ~25–30 tonnes one Starship return needs. Water extraction has never been demonstrated (the 2025 drill meant to do it tipped over on landing). Sabatier propellant plants are paper. verified

Problem 5 — Surface power at scale engineering

The wall: the honest demonstrated state of the art is ~1 kilowatt. Settlement needs megawatts to gigawatts — a 3-to-6+ order-of-magnitude gap — and solar struggles (Mars gets ~43% of Earth's sunlight with multi-week dust storms; the Moon has 14-day nights).

Who's working on it: NASA's Kilopower/KRUSTY and the Fission Surface Power program (Lockheed, Westinghouse, IX), plus terrestrial micro-reactor firms (Westinghouse eVinci, Oklo, Rolls-Royce) adapting on paper.

Reality check: KRUSTY ground-tested ~1 kWe in 2018; no reactor has ever run on the Moon or Mars (last US space reactor flew in 1965). NASA's lunar target just slipped and doubled (40→100 kWe) in 2025 with no new hardware — it's a political deadline (Exec. Order 14369), not a de-risked plan. Note: a peer-reviewed study found solar+storage actually beats nuclear over ~50% of Mars at crew scale near the equator — nuclear's real edge is poles, dust storms, and industrial loads. verified

Problem 6 — Autonomous robotic labor part science

The wall: no robot anywhere does useful unsupervised construction or mining in an unstructured environment — not even on Earth. Mars adds abrasive dust, extremes, and a 4–24 minute comms delay that forbids real-time remote control as a crutch.

Who's working on it: Tesla Optimus, Figure AI, Boston Dynamics (Atlas) on Earth; GITAI, Astrobotic, ICON for space/construction.

The tell: GITAI's best result is NASA TRL-7 — but that means preplanned tasks in known geometry, not open-ended autonomy. Tesla's Optimus demos were teleoperated, and Musk admitted (Q4 2025) it's "not in our factories in a material way." Useful unsupervised off-world labor is a 2040s–50s problem, and the peer-reviewed roadmap says it'll likely be human-supervised autonomy, not full autonomy. verified

4. The Compute & Economics walls

Problem 7 — Thermal management for space data centers engineering

The wall: in vacuum you can't convect heat to air or water — waste heat can leave only as radiation, which scales with the fourth power of temperature. Dumping gigawatts of GPU heat needs enormous, massive radiators. This radiator mass, not power, is the binding constraint.

Who's working on it: Starcloud (formerly Lumen Orbit), Google's Project Suncatcher (which lists thermal as an open problem in its own paper), Axiom Space (small edge nodes only).

The teardown: an independent audit of Starcloud's own whitepaper found a 40 MW system needs ~63,000 m² of radiators and ~900–1,100 tonnes of thermal hardware — 9 to 16 Starship launches for the cooling alone. The radiators outmass the servers. Today's flown ceiling is a single H100 GPU. Gigawatt-scale is renderings. verified
Lens insight #1 — why the smart money builds in Sweden, not orbit. If cooling is the problem in space, Earth hands it to you free in exactly the cold places: Meta's data center in Luleå, Sweden (near the Arctic Circle), Google's seawater-cooled site in Finland, Iceland's hydro-and-cold combo, and China's underwater data center off Shanghai (paired with offshore wind; Microsoft's Project Natick ran at 1/8th the land failure rate). The space pitch's real edge was never cooling — cooling is worse in orbit — it's power (orbital solar is 5–8× and continuous, with no permitting). So the rational near-term answer to AI's power+cooling+land squeeze is cold-climate Earth + nuclear/hydro/geothermal, with space a 2030s+ bet contingent on Starship. The blockers up north aren't cooling — they're power, fiber, and latency-to-users (great for training, bad for serving).
Lens insight #2 — turn the waste heat into a feature. Instead of just radiating gigawatts away, use it first — the space version of how Stockholm heats homes with data-center exhaust. GPU heat is low-grade (~60–85°C): a great match for greenhouses (crops want 20–30°C), habitat heating, and water distillation; too cool for high-temp industry or efficient power. Honest caveat: plants bank only a sliver as biomass, so ~99% of the heat still must be radiated eventually — it's a synergy (one infrastructure does double duty, and you skip separate heaters), not an escape from the radiator-mass problem. The real prize: co-locate compute + life support + light industry so one system's waste is another's input — an eco-industrial park in orbit.

Problem 8 — Space-grade compute reliability engineering

The wall: radiation flips bits, can physically destroy chips, and degrades silicon — but no rad-hardened datacenter GPU exists, so operators must fly commercial H100/Blackwell parts and mitigate. And you can't easily service hardware that goes obsolete in 2–3 years.

The honest status: rad-hard incumbents (BAE, Microchip) are ~10–15 years behind a 4nm H100. Google proton-tested its TPUs; Starcloud flies shielded commercial H100s. But as one chip-radiation expert put it: many have flown Nvidia hardware; none has reported it actually working long-term. On-orbit servicing (Northrop's MEV) extends a satellite's fuel life — none has ever replaced a failed GPU board. verified

Problem 9 — The economics & governance unsolved

The wall: there is no demonstrated closed economic loop for a Mars city. Starship's own published price is ~$100,000/kg to the Martian surface, and Mars has nothing to profitably export back — so absent full self-sufficiency, a large settlement is perpetual subsidy, not a business.

The honest read: asteroid mining is self-limiting (the entire global platinum-group market is only ~$43B/yr — flood it and the price collapses). Orbital data centers are the one genuine near-term exception — but they sell a service to Earth and don't export mass, and they're a LEO business, not a Mars-settlement engine. Space isn't lawless (the Outer Space Treaty binds nations), but there's no property regime, and the framework is fragmenting (US-led Artemis Accords vs. a China/Russia bloc), not converging. SpaceX's "Mars is a free planet" clause has zero binding force. verified

5. Four timeline scenarios

"Solved enough to enable settlement-scale use." Conservative = peer-reviewed skeptic · Balanced = neutral-expert median · Optimistic = everything breaks right · Elon-Speed = his stated target.

ProblemConservativeBalancedOptimisticElon-Speed
Closed-loop life supportNever fullyPartial '30s–40s; closure '60s+~2050City ~2050
Partial-g healthMay be unviableAdult data 2040s~2045"Solved"
Off-world reproductionMay be impossibleAttempt 2050s+~2050"Thrive" (no basis)
Chronic radiationCaps settlementMissions managed '30sMitigated 2040s"Solved"
ISRU propellant at scale2045+Pilot '30s, industrial '40sPilot ~2030With first landings
Surface power (MW-class)2045+100 kWe ~'32, MW '40s100 kWe 2030ASAP
Autonomous labor'50s+ / maybe neverLayered autonomy '40s~2040Optimus ~early '30s
Space DC (GW-scale)Never beats EarthMW '30s, GW doubtfulMW ~2030GW in ~5 yrs
Mars self-funding economySubsidy foreverNever closes w/o millionsNiche ~2050~2050

The pattern: Elon-Speed runs 2–3× faster than Balanced on the engineering problems, and simply assumes away the science-bound ones (reproduction, chronic radiation, closed ecology) — which may have no "faster" gear at all.

6. The problems nobody's pricing

The emergent issues that aren't on anyone's slide but bite during execution:

7. The opportunity lens: every wall is an opening

A bottleneck isn't only a reason something won't work — it's where the next companies and fortunes get made. So for the hardest gaps, here's the prize, the credible path someone's actually on, and a clearly-labeled moonshot — the fun part. We don't have the tech for some of these yet; that's the point. Maybe one sparks the person who builds it.

Closed-loop life support

The prize → whoever closes a stable food + air loop owns life support for every off-world base and a giant Earth market (resilient, remote, and disaster agriculture).

Credible path → sidestep the fragile full-ecosystem with precision fermentation + cultivated food (calories from microbes/cells), modular bioreactors, and AI-tuned hydroponics.

Moonshot speculative a self-tuning "digital-twin" biosphere where AI continuously rebalances the ecology in real time — the closed-loop control Biosphere 2 fatally lacked. The sensing + compute is only now arriving.

Partial-gravity biology

The prize → the answer unlocks (or redirects) the entire settlement thesis — it's the highest-information experiment in the whole program.

Credible path → stop guessing: fly a dedicated partial-gravity research module (a small spun centrifuge) to actually get the 0.38g data, plus pharmacological countermeasures.

Moonshot speculative skip the unknown entirely — build spin-gravity habitats that deliver a full 1g, so humans never live at 0.38g at all. No physics blocker — if 0.38g proves unsafe, this becomes the answer.

Radiation

The prize → a real galactic-cosmic-ray countermeasure unlocks long-duration deep space for everyone, not just Mars.

Credible path → live underground (regolith or lava tubes), hydrogen-rich / water-wall shielding, solar-maximum mission timing, and radioprotectant drugs.

Moonshot speculative an active mini-magnetosphere — a superconducting field that deflects particles like the planet does. Stuck at the lab stage today; cheap superconductors + power could change that.

Thermal / data centers in space

The prize → crack heat rejection and orbital compute opens up — or you realize the heat is an asset.

Credible path → cascade the waste heat instead of dumping it — run the coolant from the chips through habitat and greenhouse floors (radiant-floor heating in space) before it's radiated, so one loop cools the GPUs and heats everything that needs warming. Run the hardware hot to exploit the fourth-power radiative law; liquid-droplet radiators.

Moonshot / honest answer → …or don't go to space at all — put the compute in cold-climate Earth (Sweden, Quebec, the ocean) on nuclear/hydro power. Sometimes the best "solution" is noticing the premise is wrong.

Surface power at scale

The prize → cheap megawatt-class off-world power is the master key — ISRU, life support, and industry all wait on it.

Credible path → scale NASA's fission surface reactors (the 100 kWe program) and pair equatorial solar with hydrogen storage.

Moonshot speculative beamed power — collect solar in orbit and beam it to the surface by microwave/laser, or a lunar mass-driver feeding orbital solar farms. Physically allowed, decades off, but it sidesteps dust storms and the night entirely.

The pattern: notice that several of the best "fixes" don't beat the wall head-on — they route around it. Spin-gravity skips the 0.38g question; precision fermentation skips the closed-ecosystem; cold-Earth compute skips the space-cooling problem. The most valuable opportunities are often the ones that quietly change the premise.

Our read, in one paragraph

Elon has the hard part of getting there further along than anyone in history — and that is genuinely civilization-scale. But "colonize Mars" and "data centers in space" are gated by biology we haven't measured, industry five-to-nine orders of magnitude beyond today's demos, and an economy that doesn't close. The engineering walls will fall on some timeline — the question is just how optimistic. The science-bound walls (closed ecology, partial-gravity biology, chronic radiation) may not have an "Elon-Speed" gear at all, because you cannot iterate your way through an unknown you've never measured. The most credible single sentence across all of it: we can build the box; we don't yet know if the biology survives inside it. For an investor, the useful translation is that the missing pieces are the bottlenecks — closed-loop life support, ISRU, space-grade power and chips, thermal, and off-world robotics — and bottlenecks are where the next companies, and the next returns, get made.

If this was useful

This is what we do every week.

Dragonfly Lens maps a frontier — the real bottlenecks, who's solving them, and where the value flows — in plain English, every claim sourced and flagged, no hype.

Join the Lens →
More from the Lens: The SpaceX IPO, Explained · All explainers · Dragonfly Lens home

Sources & method: synthesized from primary and peer-reviewed sources current to June 2026 — NASA (NTRS, MOXIE/JPL, KRUSTY/Glenn, Human Research Program, RAD/Curiosity), Science / Science Advances (JAXA partial-g mouse study, space reproduction), Nature Astronomy (Mars solar-vs-nuclear), the National Academies (radiation limits), ESA (MELiSSA), Google Research (Project Suncatcher), independent thermal teardowns of Starcloud's whitepaper, Northrop Grumman (MEV), UNOOSA (Outer Space Treaty), and the Weinersmiths' A City on Mars. Every figure is labeled science (a genuine unknown), engineering (a scale problem), or verified. Educational research, not investment advice; Dragonfly Lens is not a registered investment advisor. Forward timelines are scenario estimates, not predictions — the large uncertainty is itself the point.