Infrastructure has never been more popular with investors. Funds run by firms such as Brookfield, Macquarie and KKR raised close to $300 billion in 2025, a record, according to the trade publication Infrastructure Investor.
Physical AI startups raised another $47.4 billion in the first half of this year, according to Crunchbase, including $16 billion for Waymo and a $1.45 billion valuation for a robotic gripper.
That capital is under fresh scrutiny.
On Thursday, the market cratered as investors rethought the growth path of the AI sector after the midterm elections.
Democrats’ gains, including governorships in data-center hubs such as Texas, Ohio and Georgia, and a wave of voter-backed restrictions on new facilities have raised the prospect of slower buildouts, tighter permitting and higher costs.
Chipmakers and data-center operators led the sell-off; the Nasdaq entered correction territory as roughly $2.5 trillion in market value was erased.
The AI boom, like traditional infrastructure, rests on physical assets whose condition, readiness and local tolerance are not fully known.
All of that money rests on roads, bridges, and the ground beneath them. And much of what anyone knows about the condition of those assets is out of date.
Most bridges in the United States are inspected every two years. That rhythm made sense when inspection meant an engineer with a clipboard, and the inspections themselves are careful work. But a lot can happen in two years, especially after a flood or hurricane, and the people who price these assets are paying close attention to condition.
Fitch’s rating criteria for infrastructure projects weigh operating and maintenance risk, including the cost and timing of renewing aging assets.
S&P Global sells investors data that scores individual assets for exposure to floods, hurricanes and other physical hazards. The result is that billions of dollars are being priced on a snapshot that may be two years old.
I have seen what that snapshot can miss.
On a county road in Alabama, a bridge was inspected in good condition, rated seven out of nine. Satellite radar showed the ground on the stretch of road beside it was sinking. The inspection was right about the bridge. The risk sat next to it, in a place no single report was designed to look. Nobody would have known from the paperwork, and that is exactly the kind of gap a valuation model cannot see.
Insurers are learning a version of this lesson with robotaxis. Insurance Business reported in January that real-world incidents are putting autonomous vehicle risk models, built on controlled trials, under pressure. A risk model is only as good as what its owner knows about the real world, and for infrastructure, the real world is the asset’s condition today. The same constraint now applies to the data-center buildout that investors are re-pricing: power, water, land and grid connections are physical realities that models and filings often capture imperfectly.
The knowledge to close that gap exists, but it is scattered. The engineering firms that design, inspect and maintain infrastructure hold decades of it, and almost all of it is sold by the hour.
AECOM booked $16.1 billion of revenue in FY2025, and 38% of its contracts were cost-reimbursable.
Across the big firms, software and digital products make up between 2% and 5% of revenue. A few are trying to turn what they know into products, such as Jacobs and Arcadis, but Arcadis’ software margin fell from 10.1% to 4.8% last year, a reminder of how hard the shift is. AI tools that could monitor assets continuously remain largely outside the core business model.
There is also a question the industry has barely started to discuss: who owns the data.
Much of the best condition data was produced for public agencies, under contracts written long before anyone imagined training software on it. Can it be used to build tools that would help every owner? On what terms, and who benefits? Without clear answers, the most valuable record of how our infrastructure ages will stay locked in filing systems and project archives.
The fix starts with better questions from investors. When was this asset last inspected, and what has happened since? Does the owner know how it is changing between inspections? Is the valuation built on that knowledge, or on a report that is two years old? For the data centers and power assets now under political and market pressure, the same questions apply: what is the true state of the supporting infrastructure, and how durable is the path to growth once local constraints are fully priced?
The first reason to ask is safety.
Catching a fault before it becomes a failure protects the people who drive across these structures every day. The money follows, because knowing how an asset is ageing decides where maintenance budgets go and how long it keeps earning. Investors have poured record sums into the physical world. They should know what condition it is in.
Saar Dickman is CEO and co-founder of Dynamic Infrastructure, an Engineering AI Agent platform for critical civil assets, which helps owners and operators of roads, bridges and other infrastructure understand the condition of their assets. The company works across 15 US states and internationally. diglobal.tec
