Terafab is a planned chip manufacturing complex in Grimes County, Texas, backed by SpaceX and Tesla, meant to make the chips behind Optimus robots, Cybercab robotaxis and SpaceX's planned space-based data centres. The latest reporting, from August 2026, puts the initial investment at $16.8 billion for a site of over 100 million square feet employing at least 3,000 people, covering manufacturing, packaging and testing of advanced logic and memory chips.
Why the cost figures keep changing
Reported numbers have moved a lot:
- March 2026 announcement: at least $20 billion.
- May 2026 public notice: an initial $55 billion, rising to as much as $119 billion if every phase is built.
- August 2026 report: an initial $16.8 billion.
These probably describe different phases and scopes rather than contradictions, but treat any single figure with caution until the companies publish firm plans.
What it's for
SpaceX cites demand for more than a terawatt of compute across robots, vehicles and orbital data centres. Earlier reporting described separate fabs, each dedicated to a single chip design: a vertically integrated approach that avoids competing for TSMC's capacity.
Why it matters
- More AI chip capacity in the world, eventually.
- Vertical integration is spreading. Large AI users increasingly want their own silicon instead of competing for the same foundry capacity.
- Execution risk is real. Building a leading-edge fab is extremely hard, even for well-funded newcomers. Timelines often slip.
The takeaway for software builders
Hardware investment on this scale keeps pushing the cost of AI down over the long term. For businesses, the smart move is to build AI features now, designed to swap in cheaper and better models as they arrive.
At NextGen Fusion, our AI and software development company in Lucknow, that's how we build: AI features that get better and cheaper as the hardware market evolves.
