In September, air traffic controllers at Dulles International Airport outside Washington began working with a piece of software that tries to see a delay before it forms. It reads flight paths, aircraft positions, weather and staffing levels, and proposes routing changes while there is still time to make them. The Federal Aviation Administration plans to extend the trial to the country’s major commercial airports.
The company behind that system, ASI, said Tuesday that it has raised $325 million in a Series C round that values it at $3 billion. Spark Capital led the financing. Meritech Capital, Craft Ventures, Energy Impact Partners and Andreessen Horowitz joined, along with affiliates of Citadel Securities. Gradient, an early backer that led the startup’s pre-seed round, confirmed the raise.
The Boston company, formerly called Air Space Intelligence, does not build chatbots. Its models simulate physical systems: national airspace, electric networks, military operations. The output is a forecast of how those systems will behave in the next hours, days or weeks, and a set of interventions that an operator can choose from before a failure spreads.
That work earned ASI its largest contract in June, when the FAA awarded it an $875 million agreement running twelve years to deploy the platform across the national airspace system. The company won the work ahead of Palantir Technologies and Thales, both of which bid aggressively for the program, according to reporting by Bloomberg. For a startup that had been selling to airlines, a twelve-year federal contract is a different kind of asset than a subscription.
Phillip Buckendorf, the founder and chief executive, argues that the distinction between his company and the model makers is where the consequences land. The technology industry, he said, spent hundreds of billions of dollars teaching machines to manipulate words, while his firm concentrated on systems where a bad prediction damages equipment or endangers people. He described the target as physical infrastructure with no margin for error.
The demand is real and easy to locate. Grid operators in the United States face interconnection queues that stretch for years, aging transmission equipment, and a surge of load from data centers that utilities did not forecast. Airlines run schedules with less slack than they did a decade ago, and a single storm cell at a hub can cascade through the system for two days. Air traffic control facilities are short of certified controllers, a problem the FAA has acknowledged for years.
Software that forecasts those failures competes with spreadsheets, institutional habit and a small group of industrial vendors that have sold simulation tools to utilities and defense agencies for decades. The pitch is that a model trained on operational data beats a rules-based system written by an engineer in 2009. The counterargument is that operators trust systems they can inspect, and that a probabilistic forecast is harder to defend in a post-incident review than a fixed procedure.
The defense business carries its own logic. Military planners buy forecasting tools for logistics and air operations, and they pay for accreditation and security clearances that consumer software companies never obtain. Contracts are long, margins are defended, and the customer does not switch vendors mid-program. The trade is a sales cycle measured in years and a dependence on budget cycles set in Washington.
The valuation implies investors expect the FAA deployment to become a template. Air traffic management is a market with few buyers, most of them national governments, and the incumbents are large: Thales, Indra, Leonardo, and in the United States the federally funded MITRE, which has advised the agency for decades. Winning one national system is a reference. Winning four is a company.
There is a timing question the round does not answer. The FAA’s operational testing began in September and the wider rollout has not been scheduled in public. Federal procurements of this size are also exposed to appropriations fights, and a twelve-year agreement is only as reliable as the annual budget that funds it. ASI has not disclosed revenue, and the $3 billion valuation is being applied to a business whose largest customer is the United States government.
The comparison investors will reach for is the defense technology cohort that has listed and raised at similar multiples over the past three years. Those companies sold hardware and software to the Pentagon and were valued on backlog rather than current revenue. ASI’s backlog now includes one very large number.
Buckendorf’s argument is that the next decade of computing will be spent on systems that act in the physical world, where errors cost more than a bad paragraph. The grid and the airspace are the two largest such systems in the country, and both are run by organizations that have been slow to replace anything.
The FAA trial will produce the first evidence either way. If controllers at a handful of airports keep using the software after the pilot ends, the case for a category gets stronger. If they turn it off, the round will look like a bet on a demonstration.
