The core thesis is that SpaceX/Tesla math is being mispriced because the build-out of power, data centers, chips, robo-taxis, and Optimus is compounding faster than markets are discounting it. The notes argue that mega-scale compute can throw off tens of billions per gigawatt, while SpaceX can turn stranded Nvidia supply into 2026–2028 revenue faster than traditional builders, with Starlink and compute alone supporting a much higher valuation. Tesla upside is framed around 2,000+ Cybercabs by Christmas, 10,000 by January/February next year, and eventually 100,000-plus triggering takeoff, while Optimus and semi-scale manufacturing add another layer of optionality. The bear case is not demand, but power, permits, politics, and customers running out of money before the build-outs finish.
Brier score around 0.25 is treated as roughly coin-flip accuracy, while much lower scores imply ~71% right only when prediction-market outcomes resolve cleanly; the stated trading target is 80% correct, with Brian Wang cited as having ~35,000 articles and nearly 6 years on YouTube.
Falcon 9 is claimed to have launched 600–700 times with two failures, implying 99.8% reliability versus 98%–95% for others; newer rockets are said to be near three nines after only nine launches, while two-nines systems require under 10 launches.
Speed-of-information asymmetry is framed as the edge: Elon Musk’s X breadcrumbs let models update faster than traditional media, and if the market has not yet believed an event that already happened, the edge is waiting for dissemination rather than waiting for the event.
The Shuttle’s $5 million-per-launch target is described as having turned into $1 billion, and nuclear buildout is said to have stalled even though the world could have built about 1,000 reactors instead of 100; cited reasons are 1970s high rates, media, and oil/gas/coal influence over the Nuclear Regulatory Commission.
A 2028 political-loss scenario is flagged as a real execution risk if Elon Musk is forced into defense mode instead of building, and corruption is contrasted: in China 10%–20% can be taken while the bridge still gets built, while elsewhere enough is skimmed that nothing gets built.
Over the prior 24 hours, crypto was hit by a claimed $6.3 billion hack, described as the biggest in crypto history, with rumors pointing to North Korea and uncertainty about how access was obtained.
CZ-related selling pressure is dismissed as non-threatening because the money sits in a fund that needs board approval; the stock is said to have fallen back to August levels after Elon trolling CZ, even though the story is bigger than four months ago.
For Tesla, the three biggest issues are framed as compute power, robo-taxi, and Optimus, with confidence highest in compute power, optimism on Q2/Q3 ramp timing, and possible slippage into Q3/Q4.
A robo-taxi tipping point is defined as 2,000 Cybercabs rolling autonomously and making money, with a Christmas date cited as plausible; the end-of-year range is 3,000–5,000, and 10,000 is expected by January/February next year.
The takeoff threshold is set at 100,000 robo-taxis, with a valuation map of 100,000 vehicles at $50,000 each putting the business near $50 billion; the notes also say 1,000 to 10,000 is easier than 10,000 to 100,000 in ramp terms.
Austin density data is used to extrapolate from 712 of 1,080 Texas-registered vehicles in Austin to 274 per million people, then to 10,000 vehicles across Houston, Dallas, and Miami by January/February next year.
Optimus skepticism is countered with vertical-integration logic: V3 was already in hand by July, scale production was supposed to start in September and missed by only a few weeks, while semi scale production actually hit in September.
Tesla’s $3 billion–$5 billion new factory and unboxed process are cited as process changes; CyberCab wiring was redesigned around the 48V Cybertruck architecture, and Cybertruck HVAC is said to use no refrigerant lines.
The Optimus cost stack is modeled down to a $20,000 or $10,000 brain build, with roughly $5,000 for memory and $5,000 for the chip, allegedly yielding 100x energy/compute versus a phone and 20x–50x versus a laptop.
SpaceX and Tesla are described as not spoon-feeding earnings-call details, so people connecting breadcrumbs from X get extra advantage; the same logic is applied to data-center and robot build-outs being visible in real time through drone footage.
For AI compute, one gigawatt of capacity is claimed to generate $50 billion of revenue and about $10 billion a year at a 20% margin; utilities are said to make only a fraction of 1% on a gigawatt of compute.
Mini hard is priced at $13 billion for 400 MW, or 40% of a gigawatt, and is said to house 220,000 B300s; payback is claimed in six months, versus CFO guidance of 9–12 months, and construction is said to take five or six months instead of two years.
Macro hard, mini hard, and another iteration are said to complete in September, November, and December respectively; Colossus 1 and Colossus 2 are described as finishing in about a year, faster than the two-year norm that only 20% of projects beat.
CoreWeave is used as the benchmark: five-year Meta/Microsoft leases, 20%–30% markup, and roughly 2.5-year payback via SPV financing, contrasted with the faster SpaceX model that can build in five months and pay back in six.
SpaceX is said to have 2.7 GW by year-end as a solid target, with 10 GW next year and 20 GW an extreme long shot; Jensen is quoted as saying 1.1 million chips are coming in for sure, and Elon is said to have secured 40% of Nvidia output for Vera Rubin.
The 2027 framework is 51 GW total, with 35 GW domestic and 6–8 GW from China; at 20 GW, SpaceX would be near 70% of domestic 2027 build, while stranded GPUs could cover about 15 GW of chips plus another 10 GW later.
The power mix for 10 GW is sketched as 4 GW grid, 4 GW natural gas, 2 GW solar, and 2 GW battery, with solar and batteries smoothing peak-hour draw; relaxing 99% uptime and being first to be turned off could unlock 100 GW more in the U.S.
The 2026–2027 valuation model uses only compute and Starlink: compute at $50 billion per GW, 2 GW in 2026 equal $100 billion revenue, and CFO-guided ARR of $100 billion by year-end can include Starlink.
Starlink is said to be just over $5 billion a quarter, or about $21 billion a year, with a 42% profit margin; compute is assumed at 20% margin, and together those two pieces alone justify a $217 stock price today under a conservative PE.
Next year, Starlink is modeled at 14% growth, compute at 10 GW with a 98% chance, and average selling price at $54 billion because nobody can deliver; compute-alone revenue is then put at $530 billion, with the stock target at $856.
An upside variant adds 100–200 billion from renting unused data centers without chips, potentially pushing the price toward $1,000 or higher; a separate risk is that if modular builders catch up, demand could drop from $60 billion to $40 billion.
Reuben is treated as the next leap: each ship uses 1.5x the power of a GP300, but delivers 3x–6x, or as much as 10x, the inference; that lifts implied economics from $26 billion per gigawatt to $150 billion per gigawatt in the full-monetization case.
Ruben Ultra is said to improve economics further, shrinking each data center and enabling 1.5–5 mini-hards per gigawatt instead of 2.5; the notes call a tripling of revenue a conservative Reuben assumption.
Anthropic is said to be paying $3.1 billion per month versus $1.2 billion earlier, with that flow expected through year-end; Microsoft, Google, Amazon, and about 200 AI startups are all described as needing more capacity.
The moat is framed as full-stack synergy: brain, hands, and manufacturing at scale, with Tesla leading all three; competitors need a conglomerate of Chinese firms or a $50 billion Figure-style effort just to close the gap.
The 12-month SpaceX range is modeled at $755, $855, and over $1,100 depending on measurement, based only on Starlink and compute and excluding everything else; a 5x prior call is referenced as now looking modest.
A 7,923% four-year profit-growth estimate from John Ellishman is used to argue that even rough math points to an $11,000-per-share outcome, and Elon is said to have chimed in that Tesla would be very high too.