Brian Wang models SpaceX reaching over $1,100 a share on Starlink and compute alone

Brian Wang's track record frames the conversation

The host introduces Brian Wang by highlighting his prediction-market scorecard, which Wang says places him "number eight out of like 34,000" forecasters based on a Brier score he estimates at around 0.09, implying roughly 90-95% accuracy on resolved, money-backed predictions. Wang attributes this to grounding forecasts in "hard physical quantifiable things" rather than vibes, citing Falcon 9's reliability record (two failures across 600-700 launches) as an example of a quantifiable edge.

Wang also discusses past misses, including underestimating how politics and NASA's institutional incentives stalled space shuttle cost targets in the 1970s, and overestimating nuclear reactor buildout due to interest rates and regulatory capture. He flags political risk going forward:

"if you know the House and Senate are lost... things will slow down and then in 2028 if the presidency is lost then... he could be running around trying not to be thrown in jail versus actually doing stuff."

Robotaxi and Optimus ramp claims

Wang says he is "utterly confident" in SpaceX's power buildout but more cautious on robotaxi scale, noting Tesla has 712 of 1,080 registered Austin vehicles and expects 10,000 robotaxis by January or February next year based on extrapolating Austin's registration density to Houston, Dallas and Miami. He puts his own estimate at 3,000 to 5,000 by year-end. The host's threshold for Wall Street confidence is 2,000 cars; Wang's is 100,000, which he associates with roughly $50 billion in revenue.

On Optimus, Wang argues Tesla's edge comes from redesigning its entire factory and vehicle architecture (48-volt wiring, no-refrigerant HVAC) around an "unboxed" process, unlike Mercedes or BMW, comparing it to Costco redesigning stores around forklifts. He also stresses the chip inside Optimus as the real value driver:

"the caviar is the chip."

Mini hard data centers and the compute payback math

Wang describes SpaceX's "mini hard" data center as 400 megawatts housing roughly 220,000 chips, built in about five to six months versus an industry norm nearer two years, with a payback period he estimates at six months against CoreWeave's cited roughly two-and-a-half-year model. The host cites a $13 billion construction cost figure. Wang credits supply-chain mastery developed during COVID and efficiency gains (around 15-20% less energy, smaller footprint) from iterative redesign.

The two discuss where additional chips for a scale-up beyond 13 gigawatts would come from, floating "dark" or stranded GPUs sitting unused at Microsoft, Amazon and others as a likely source, alongside direct allocation from Nvidia. Wang references semianalysis estimates of roughly 51 gigawatts of global 2027 capacity, with 35 gigawatts domestic, implying SpaceX could represent near 70% of U.S. buildout if it reaches 20 gigawatts.

Valuation modeling: from $217 to over $1,100

The host runs a valuation model crediting SpaceX $50 billion in revenue per gigawatt of compute at a 20% margin, plus Starlink revenue, arriving at a current sandbagged price of $217. Extending to 2026 with 10 gigawatts of compute and a $54 billion average sales price per gigawatt, the model produces roughly $530 billion in compute revenue and a modeled stock price of $856. Applying Wang's higher 13-gigawatt, 98%-confidence scenario pushes the modeled price over $1,100.

Wang adds that renting out extra data center capacity without chips could add another "100 to 200 billion," and that margin expansion toward 60-70% (versus the 20-30% assumed) would push figures "towards a thousand or higher." Both note the host also cites a Deutsche Bank-style figure of $50 billion revenue per gigawatt at an "easily" conservative 20% margin, and references hedge-fund analyst John Ellishman's forecast of 7,923% expected profit growth across names including SpaceX, Tesla and Marvell, which Wang calls directionally consistent with his own numbers but not yet proven on margin.

Risks the pair flag

Both identify the main risk as whether other hyperscalers (Google, Amazon, Microsoft) solve modular fast construction themselves, which Wang says would compress SpaceX's pricing power "from 60 billion down to 40 billion." The host raises customer demand risk, noting Amazon's Andy Jassy has said the company is "tapped out" on compute through 2027-2028. Wang also discusses stranded-chip logistics, estimating "two months" to relocate unused Vera Rubin chips plus roughly a month each to rack and install them.

On chip generations, Wang estimates Rubin chips use about 1.5 times the power of GB300s but can deliver three to six times more inference revenue, potentially pushing revenue per gigawatt toward $150 billion in a fully monetized scenario, though this remains his estimate rather than a confirmed figure.

Where the speakers disagree or hedge

The host and Wang diverge on the robotaxi threshold for triggering Wall Street confidence (2,000 cars versus Wang's 100,000), and on whether SpaceX might ever become a compute "landlord" renting to others — Wang calls this "a long shot" since Elon Musk "likes to do everything himself." Wang declines to call Tesla's AI/robotics moat "insurmountable" by rivals, saying China could eventually compete via a conglomerate of firms including Huawei and BYD, though "years behind." Neither speaker gives a single confirmed figure for total 2027 gigawatt capacity, treating 13, 15 and 20 gigawatts as a probability range rather than a settled number.

What's next

The episode ends before covering Tesla's robotaxi economics, Optimus pricing, or space-based data centers in detail; the host says these will be addressed in a planned follow-up episode. Wang's closing prediction is that SpaceX, based solely on Starlink and compute, could trade "somewhere between 755, 855 and over $1,100" within the discussed timeframe, which he frames as his own modeled view rather than a certainty.