This video features a discussion between the host (Stephen) and early SpaceX investor Gavin Baker about SpaceX's current valuation, its diverse ventures, and the future of AI.
**Key News and Updates from the Discussion:**
* **SpaceX Market Cap:** The discussion begins with SpaceX's market cap just under, then over, **$1.5 trillion**, described as a "sale" for long-term investors.
* **Investment Philosophy:** Both speakers emphasize a long-term investment approach, focusing on the company's technology, progress, and execution rather than short-term stock price fluctuations. Gavin states he doesn't "give a f*** about the stock price." The host champions "slow money" over "fast money," advising investors to focus on what the company does, not what the stock does.
* **Analyst Reports:** The host notes Wall Street analysts generally have "buy" ratings on SpaceX but dismisses their opinions as irrelevant for personal investment decisions, suggesting their job is to generate business, not necessarily be correct.
* **Host's Valuation Models:** Stephen reveals he's spent months building and refining his own **20-year SpaceX and Tesla valuation models**, which will soon be available on Patreon. Interestingly, he finds some analyst estimates for SpaceX to be *above* his own "base case" long-term projections.
* **SpaceX's Broad Portfolio (The "Empire"):** The company is highlighted for its diverse operations, including:
* **SpaceX** (launch business)
* **Starlink** (Global Communications, now expanding into cell service)
* **AI Compute Business** (Grok products, language model, chat, agents, voice, build, imagine)
* **X platform** (and X money in the works)
* **Grok and AI Products:**
* **Grok 4.5** is positioned as "dominating AI models in terms of bang for buck" and "winning on intelligence per cost."
* **Aggressive AI compute build-out** is continuing, with "more AI compute deals" anticipated.
* **Grok Voice** recently received a "massive overhaul."
* **Office agents launched "just today,"** integrating Grok into Microsoft PowerPoint, Word, Excel, Outlook, and Google Docs, Sheets, Slides, allowing an AI agent within spreadsheets and other office tools.
* **Grok Build** (for coding) and **Grok Imagine** (video and image generation) are also part of the suite.
* The rate of progress and improvement for Grok products is described as "staggering," with "near daily updates."
* **Starship Progress:** Starship is seen as a "massive unlock" that could be flying regularly in "just a few years." The host dismisses concerns over launch delays/scrubs, calling them normal for test flights of new technology with "hundreds, thousands of changes" between iterations. Gavin adds that for long-term investors, whether reusability is achieved in three months or twelve months is insignificant.
* **AI Compute as the Core Story:**
* Gavin argues SpaceX uniquely monetizes installed power at a premium, bringing it online "faster and cheaper" than competitors, making "the most profit per gig." He suggests **3 gigawatts of AI compute next year could generate $150 billion in unmodeled revenue**.
* The host emphasizes the AI compute business is "not widely appreciated" but is a "consistent, predictable infrastructure build out."
* Evidence: Google and Anthropic are paying SpaceX AI for compute, indicating SpaceX's superior speed and cost-effectiveness in getting compute online.
* The host's thesis: SpaceX AI will build maximum compute, retain what's needed for its own models, and lease the surplus to other companies, using the revenue to fund further expansion.
* The concept of **"orbital compute"** is mentioned as a future possibility.
* Gavin explains that open-source models shift margin to the infrastructure layer (compute providers) because tokens still require the same compute resources. He states the market is "structurally short compute," making "neoclouds" (compute providers) bullish.
* He notes the curious absence of Grok 4.5 from an on-screen graphic comparing AI model pricing by "intelligence per cost," despite Grok's leadership in that metric.
* Gavin introduces the "conductor and violinists" analogy: frontier models (e.g., Grok, Claude) act as conductors for complex tasks, while cheaper, open-source models serve as violinists for simpler, high-volume tasks, all requiring underlying compute.
* "Intelligence per dollar" is highlighted as the crucial metric in AI, where having the most efficient infrastructure matters.
* Grok 4.5 is described as being on the "Pareto frontier," meaning it offers the best intelligence for the cost, being "almost as intelligent as the frontier models... but a fraction the cost."
* Gavin asserts that Wall Street's current premium on "model supremacy over cost" is a "mistake."
* He mentions Grok 4.5 is a **1.5 trillion parameter model**, a **2 trillion parameter model is currently in training**, with future plans for 6 trillion and 10 trillion parameter models.
* The "speed in which you bring on a gigawatt [of compute] is literally cost" (quoting Jensen Huang), and data center builds are beneficial for "blue collar America" due to job creation for skilled trades.