Latest / Minds, Bodies, and Terawatts / SpaceX Secures $30B Google Computing Deal Ahead of $1.78T IPO
Transcript
- 0:00[Host] This is Minds, Bodies, and Terawatts, where the rocket company becomes the server farm. The New York Times reported today that SpaceX has secured a thirty billion dollar deal to supply Google with AI computing power, ahead of an IPO valuing the company at one point seven eight trillion dollars.
- 0:18[Guest] Google is renting roughly a hundred and ten thousand Nvidia GPUs from xAI's Colossus data centers in Memphis, chips Musk originally built for his own AI lab, now leased to a direct rival.
- 0:30[Host] Google is one of the world's largest cloud builders. Why rent from a rocket company?
- 0:36[Guest] Because demand has outrun what even Google can build. Morgan Stanley puts total hyperscaler capex above six hundred billion for 2026 and the infrastructure still isn't keeping up. A Google Cloud spokesperson told reporters this is bridge capacity to meet surging customer demand.
- 0:52[Host] Bridge capacity that runs through 2029. That's a long bridge.
- 0:55[Guest] Either side can walk after December on short notice, so Google isn't locked in forever. But Musk gets an anchor contract that de-risks the largest IPO in history right before shares start trading.
- 1:07[Host] xAI competes directly with Google's Gemini. Musk is selling compute to his own rival.
- 1:13[Guest] Both sides accepted that conflict because the shortage forced the handshake. And xAI needs the revenue badly. The IPO filing revealed the lab lost six point four billion dollars last year on three point two billion in revenue.
- 1:27[Host] Losing two dollars for every dollar earned. And the IPO is asking the public to fund that gap.
- 1:33[Guest] Harrison Rolfes at PitchBook called the financials reckless. But his read is that the whole bet is on timing.
- 1:40[Host] Timing how? They're torching billions a year right now, and they're asking ordinary investors to trust a payoff that may never land.
- 1:48[Guest] By the time rivals build their own clusters, xAI will have moved the goalposts. If it pays off, the burn was a bargain. If not, his words, the largest venture-funded correction in history.
- 2:01[Host] They're reportedly raising seventy-five billion in the offering, the largest ever.
- 2:05[Guest] And most flows into compute. The growth story in the filing isn't rocketry. It's contracted GPU revenue from customers who can't build fast enough.
- 2:13[Host] Is Wall Street pricing this as a rocket company or a compute company?
- 2:18[Guest] Compute. Starlink proved the recurring revenue model. The Google deal proves SpaceX can sell AI infrastructure at scale. Strip that contract away and a trillion-plus valuation is hard to justify on launches alone.
- 2:30[Host] Musk also controls Tesla's robots, announced Terafab for chip fabrication, and operates the only reusable launch system. Chips, rockets, robots, and AI under one person.
- 2:40[Guest] One person spanning all four layers hasn't happened before. The IPO converts that vertical reach into publicly traded equity, and millions of retirement accounts will depend on it staying consolidated.
- 2:51[Host] What happens when making AI cheaper threatens the stock price?
- 2:54[Guest] Every GPU-hour routed to xAI training instead of paying Google customers is revenue SpaceX doesn't book. That tradeoff sat with one person in a private company. Post-IPO, it becomes a quarterly earnings question.
- 3:07[Host] Patrick Deglon argues in Unscarcity that this is exactly what the Enterprise EXIT was designed to prevent. Capture doesn't arrive through the machines. It arrives through the ownership structure.
- 3:17[Guest] Once abundance is collateral in everyone's retirement portfolio, making it cheaper becomes an attack on the portfolio itself. For more on how ownership reshapes the path to abundance, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.