Both responses are appropriate.
At an event in Austin, Musk announced what he's calling Terafab, a joint venture between Tesla, xAI, and SpaceX designed to manufacture AI chips at a scale the world has never attempted. The facility will start with an advanced technology fab in Austin, Texas, capable of producing any chip type: logic, memory, and the lithography masks to design them, all under one roof. The goal is to compress chip design cycles from months to days.
Musk framed the ambition in planetary terms. Current global AI compute output sits at roughly 20 gigawatts per year. Terafab is targeting a terawatt, which means the combined output of every existing fab on Earth today represents about 2% of the stated goal.
To reach that number, Musk said the effort will require approximately 10 million tons of payload delivered to orbit per year, with space-based solar supplying the power. His argument: solar in space delivers at least five times more energy than ground-based installations because there's no atmosphere, no night cycle, no weather, and no neighborhood opposition.
That last point, "nobody wants it in their backyard," was delivered as a joke. But it wasn't entirely one.
Before the satellites and the mass drivers on the moon become real, there's a more immediate story playing out in Texas and wherever this supply chain lands. The Austin fab is being built now. That facility will consume significant amounts of power and water. It will generate heat. It will require grid interconnection at industrial scale.
The pattern is one the energy and facilities world has seen accelerating for three years. AI infrastructure demand is outpacing grid planning timelines. As data centers have already strained regional grids in Virginia, Texas, and parts of the Midwest, a chip fab of this ambition adds a new layer of pressure, not just on electricity supply, but on the semiconductor supply chain itself.
For energy and operations professionals, this isn't an abstract futurism story. It's a procurement and infrastructure planning signal.
The more speculative piece of the announcement is the case Musk made for space-based AI compute becoming cost-competitive with terrestrial infrastructure within two to three years. SpaceX already has over 10,000 Starlink satellites in orbit. The company understands heat rejection in space at a scale no one else does. The economics of launch are dropping.
If even a fraction of that vision becomes real, it reshapes assumptions about where energy-intensive AI workloads eventually run. That's a question for long-range capital planners, not just technologists.
For now, the story is a chip fab in Austin, an energy demand curve that keeps going up, and a reminder that the AI infrastructure buildout is nowhere near its ceiling.
