The AI economy has a new bottleneck, and it is not a chip

The rate-limiter on the AI economy in 2026 is not silicon. It is the people who hang conduit, pour rebar, and wire switchgear. Chip shortages still get the headlines, but the capex cycle that turns models into product now runs through job sites the industry cannot fully staff.

That is the world behind Eric Wu's newest company, NavigateAI, which came out of stealth in late May with a $25M seed at a $225M post-money valuation. Wu calls the product a hands-free expert coach for the people physically building things. The bet is software can stretch a thinning crew. The trade group Associated Builders and Contractors has said roughly 349,000 additional construction workers would be needed this year just to keep pace with demand. That number is the load-bearing detail. The gap is structural: aging crews, immigration enforcement, and mega-projects that each need thousands of builders. Meta's Hyperion campus in Louisiana has been reported to need around 5,000 construction workers. OpenAI's Stargate site in Texas has involved 6,400. A Kelly survey found 90% of data center operators cite staffing as a critical constraint.

The reusable category is a capex cycle that runs out of bodies before it runs out of ideas. A coaching copilot can compress training time and lift a junior crew toward a senior crew's output. It cannot invent the senior crew. The frame travels because the next productivity frontier is not the desk. It is the job site, and the trades are now the binding constraint on the AI buildout.