[Recorded before earnings released] SpaceX reported its first public quarter earnings this week, and the number that matters is not the Q2 2026 print. Guidance is the entire event. The Anthropic, Google, and Reflection AI data center agreements all landed too late in the quarter to move reported revenue materially, so the historical figures will understate the trajectory by a wide margin. What we want from the CFO is specificity on how the AI and compute business converts into revenue and earnings over the next several quarters, since Musk will naturally frame the opportunity in years rather than quarters. Sell-side consensus is scattered, largely because analysts have no reliable handle on capital spending into new data centers, though the run rate is likely consistent with prior years and possibly a step higher. Separately, 911 million shares become available to be sold two days from now (this past Thu), which will drive near-term price action independent of anything said on the call. Our long-run view is unchanged at $5 trillion plus, and we would treat quarterly volatility accordingly. [SpaceX subsequently provided guidance that Dec 2026 annual run rate revenue will be $100 billion, +220% in just 6 months, and 2030 full year revenue is forecasted at $1.0 trillion. The CFO also noted that SpaceX return on AI data center capex is less than one year due to the overwhelming demand of AI compute.]
The Boring Company is reportedly raising at a $20 billion valuation. Musk has a demonstrated ability to convert physically constrained businesses into technology businesses, but tunnels do not scale the way software does. A software company can serve a billion users in a year. It cannot dig a billion holes. The Las Vegas Loop, the only operational system, cost roughly $48.7 million to construct, and the economics include variable revenue on ridership rather than a flat construction fee, which gives the asset a toll road profile with annuity characteristics that persist long after the capital is spent. The constraint is the customer, because this is a government-driven revenue model and permitting cycles move at government pace regardless of how fast the operator executes. The bull case requires a commercial channel, and the most interesting version is smaller-bore tunneling for fiber optic and communications infrastructure rather than passenger loops. Businesses that serve only governments are interesting, but businesses that serve governments and commercial customers carry a materially larger TAM (total addressable market), and that is the question we would put to management. Absent that, we see better risk-adjusted uses of capital, with credible hyperloop-class execution the one development that would change our view.
Prediction markets are repricing faster than almost anything else in the private market. Polymarket is out at a $20 billion valuation after closing a round in April at $15 billion, on annualized revenue above $1 billion, while Kalshi marked $22 billion in May and is now reportedly seeking $40 billion. The most telling data point came from Robinhood, which reported more Q2 revenue from prediction markets than from stock trading. Investors should understand the profit distribution underneath that growth: roughly 0.4% of participants capture about 70% of profits, and 84% lose money, with quantitative trading firms extracting the edge much as they do in equities. Sports still accounts for roughly 89% of activity, so this is a gambling business today whatever the framing. The larger and underappreciated opportunity is international, and the analogue is Interactive Brokers, where 60% to 70% of customers are non-US clients opening US brokerage accounts, a structure that built a company with a founder worth roughly $40 billion. Starlink follows the same pattern, and the pattern is consistent: take a service Americans treat as ordinary and extend it to the four billion people who have never had access. Institutional hedging is the other vector worth watching, with event contracts used to hedge input costs such as produce prices for restaurants, and that is where the category graduates from entertainment to infrastructure.
Zack Dell's company, Base Power, raised $1 billion at a $13 billion valuation and has installed 23,000 residential batteries across Texas and the Chicago area at a pace near 100 installations per day. The model is clean: own and operate batteries in deregulated markets, partner with utilities in regulated ones, charge overnight when grid prices are low, and discharge at peak demand, reducing strain on the grid while lowering the homeowner's cost. The macro backdrop does much of the work here, as the forward price to secure power on PJM, the largest US grid, is up 800% over the past year. Distributed storage is one of the few near-term levers that addresses peak constraint without waiting on new generation, and it requires no rooftop solar to function. Pairing with residential solar is the obvious extension, since panels generate nothing at night and storage solves precisely that gap. The broader thesis is that the home acquires a new utility layer alongside HVAC and hot water, with local power storage and eventually local compute becoming standard residential infrastructure. We would expect the cost curve in electricity to compress the way inference costs have, and companies that own the distributed asset base will capture the spread.