The $750 Billion Question: AI CapEx Has Gone Off the Rails
I keep coming back to the same thought: who is this all for?
OpenAI just ballooned its projected spend through 2030 to $750 billion, up 25% from their 2026 estimate. The first campus alone, Project Camellia in Georgia, is a 1,400-acre data center drawing 3.2 gigawatts of power from Georgia Power. They're targeting deployment between 2028 and 2032. That's a half-billion-billion-dollar bet on a computing paradigm that most consumers still treat as a novelty chatbot.
Google Cloud reports revenue surged 82% year-over-year to $24.8 billion in Q2 2026, driven almost entirely by enterprise AI solutions. Alphabet's capex sits at $180-190 billion for 2026 with a cloud backlog of $514 billion. CEO Sundar Pichai signaled the payoff won't come until 2027. Half a trillion dollars in the pipeline and nobody's sure where it ends.
Nvidia isn't just sitting here watching their GPU business. They're deploying Jetson chips on a Lunar Outpost rover that will be the first GPU on the lunar surface. An orbital partner called Firefly Aerospace is putting Jetson on a satellite, and NASA is backing private lunar exploration campaigns. We are putting AI accelerators in space and I cannot pretend I saw that coming.
The valuations keep climbing. AI chip startup Etched hit a $10.3 billion valuation as of July 23, betting everything on specialized inference chips instead of general-purpose GPUs. Ex-Uber CEO Travis Kalanick's robotics company raised $1.7 billion led by a16z. Money men building robots and a startup going to the moon with edge AI silicon.
The scale is real and I don't know what to do with it. OpenAI wants to spend $750 billion across four years. Google's backlog is half a trillion with no clear product at the end of it. Enterprise AI adoption is happening -- Google Cloud proves that -- but consumer-facing AI still feels like cathedral-building for a faith nobody has finished worshipping yet.
Nobody knows the answer to whether any of this makes sense. And if they do, they aren't talking.
Related
More from the blog
AMD Just Launched a Direct Challenge to Nvidia's AI Monopoly
AMD just launched Helios, a rack-scale AI system challenging Nvidia's monopoly — with OpenAI, Meta, and Anthropic as early customers. Why infrastructure competition matters for the future of AI.
Anthropic's Building Its Own Chips. The Nvidia Dependency Problem Just Got Real.
Anthropic's hiring a custom silicon team, joining the growing list of AI companies building their own chips to reduce Nvidia dependence and optimize for their models.
Nuclear got a $1 billion vote from Sequoia, and AI's appetite is why
Valar Atomics just raised $1 billion, led by Sequoia, to build miniature nuclear reactors for AI workloads, a clear sign the compute buildout now runs on power.
CoreWeave's $2.58 Billion Quarter Proves AI Is Devouring Crypto's Money
CoreWeave reported $2.58 billion in Q2 revenue with a $104 billion backlog, proving AI infrastructure demand is shifting capital away from crypto toward compute.