Deep Dive
Elon Web Services and GPU Monetization Mechanics
Aaron walks through how SpaceX's data center business monetizes GPU capacity. At 10% training and 90% inference allocation, SpaceX targets two to 2.5 gigawatts by year-end—roughly on pace with Elon's public guidance. The real opportunity lies in token density: Vera Rubin chips generate substantially more tokens per watt than older GB300 models, which means the same physical footprint generates higher revenue if utilization stays high. Aaron ran the math and found that internally monetizing Vera Rubin via Grock at reduced pricing and 50% utilization still yields $100-200B per gigawatt annually—higher than the $30-50B SpaceX commands from external customers like Google. This creates a virtuous loop: Nvidia wants SpaceX as a flagship customer to showcase chip performance, SpaceX maximizes utilization through coherent infrastructure, and that profitability feeds back up the stack to all participants. The fourth external client Aaron suspects is Microsoft, given deal sizes north of $1B monthly.
Why Space-Based Compute Makes Physical Sense
Aaron tackles the thermal dissipation problem head-on. Space-based GPUs face one core challenge: radiating heat into vacuum instead of air. The solution involves massive radiators—flat metal panels with cooling pipes—that point away from the sun in sun-synchronous orbit, dumping thermal load into the coldest part of space. The engineering hurdle is material science: radiators must withstand extreme heat differentials and reliability must be proven over time. However, Aaron emphasizes this is an engineering problem, not a physics limit. Solar panels in space also generate five times more energy than ground-based panels for the same footprint, since they bypass atmospheric interference. The AI1 satellite will be roughly twice as heavy as Starlink V3, with larger solar wings and new radiator panels, but still compact enough to fit on a Starship. Aaron expects early iterations to cost money and carry risk, but once prototype radiators are validated, manufacturing scales and costs decline. The host raised Kessler syndrome—orbital debris collisions—but Aaron counters that space is vastly larger than the garbage patch, and active debris management solutions exist if funding materializes. Current collision avoidance already happens three times per satellite per day.
SpaceX Revenue Stack and Competitive Moats
The host breaks down SpaceX's four-legged stool: rocket launches, global communications (Starlink), AI compute, and defense/government contracts. Each leg provides optionality—if one falters, others sustain cash flow. Starlink direct-to-cell poses the largest single opportunity: at 30-40% global penetration of the $5 trillion cellular market, that business alone justifies massive valuations. Aaron notes that Starlink is their fastest path to highest internal rate of return over three years, while compute is a longer-term, higher-upside bet. When asked about competitor risk, Aaron argues SpaceX enjoys a 5-10 year moat because hardware execution is far harder than software replication. Blue Origin setbacks cost billions and set them back years; Rocket Lab can't match Starship's cadence or payload capacity. The differentiator is cumulative: SpaceX has launched more rockets, deployed more satellites, and maintained more working systems than any company in history save the USSR (over 70 years). Aaron frames this as an execution premium that hasn't been fully quantified but is substantially larger in space than in AI or software.
Cost Curve to Orbit and Starship Economics
Aaron lays out the asymptotic cost curve for kilograms to space. Current costs hover around $100-200 per kilogram; the curve asymptotes toward $10-20 per kilogram. The mechanism is mechanical: reusability cuts the $250M vehicle cost in half per reuse; current boosters already log 30-40 flights, proving the concept. Build more vehicles and manufacturing cost falls from $250M per unit toward $50M or less. When Starship achieves full reusability at scale—say 20 reuses—the cost per kilogram plummets further. The host's Starship math assumes 200 launches at 30,000 tons to orbit per flight, generating $999B in lifetime revenue. At 20x sales, that's $20T valuation; at 40x, $40T. Just this revenue stream alone implies $1,500-3,000 per share. Elon has publicly stated 10,000 launches by the early 2030s, meaning the 200-launch scenario is conservative. Aaron stresses the key question isn't whether reusability works—it already does—but how fast SpaceX can scale production and maintain flight rates. The competitive advantage here is durable because newcomers must solve the same hard problems but have zero flight heritage.
Model Consensus and 2028-2030 Targets
The host presents 11 valuation models ranging from crude ($210 stock, early 2026) to detailed ($1,383 by 2030). Morgan Stanley targets $3.5T revenue by 2040, but Elon challenged that and said $3.5T by 2033. Semi-Analysis recently released a $457 compute-only target for Q4 2027. Aaron's own Monte Carlo runs 800+ variables and consistently produces probability distributions with long right tails—meaning extreme upside is plausible. Both agree on a floor valuation: at $1.75-2T market cap, SpaceX looks like a stable Starlink bond with predictable cash flows from launches and bandwidth. Everything above $2T is growth optionality. The host's Monte Carlo median estimates $527 per share by 2028 (3.61x from current levels) and $926 by 2030 (6.18x). Aaron says these align with his bottom-up models and he may actually be slightly more bullish on Starlink than the host's assumptions. The consensus takeaway: there's a floor of predictable value and a ceiling of nearly unlimited upside depending on execution risks around Starship, direct-to-cell, and agent-driven token demand.