Alaska Airlines started letting an AI system suggest flight routes to its dispatchers in 2021, and five years later the same software company has a 12-year, $875 million contract to rebuild how the United States manages its entire sky.
That is either the fastest useful procurement the Federal Aviation Administration has managed in a generation, or it is a very expensive bet that a tool built for one airline’s operations center scales to a continent.
The company is Air Space Intelligence, a Boston startup that beat Palantir and Thales for the work. Transportation Secretary Sean Duffy and FAA Administrator Bryan Bedford announced the award in late June, and Forbes reported at the time that the agency was buying two things: a replacement for the aging Traffic Flow Management System, and a predictive layer that flags congestion before planes leave the gate. The first is called FMDS, for Flow Management Data and Services, and it becomes the data backbone of the FAA’s Air Traffic Control System Command Center. The second is SMART, for Strategic Management of Airspace, Routes, and Trajectories, and it runs inside the first. Nextgov reported that the award came as part of the broader air traffic control modernization push, with initial SMART deployment targeted for this fall and full rollout expected within one to two years.
What Alaska Airlines Actually Bought
The product that got ASI here is called Flyways, and its job is narrow. It reads live traffic, weather, wind, and airport conditions, then hands a dispatcher a ranked set of route options. NPR’s Joel Rose spent time inside Alaska’s network operations center and came back with the least glamorous possible description of what the system does, from network operations center manager Alek Mead: “It ingests all that information and really kind of puts us all together in a nice little package for the dispatcher.”
Alaska’s managing director of network operations control, Capt. Bret Peyton, was even more careful about the boundary. The software “doesn’t decide the route or waypoints,” he said, “but it does help provide suggestions that might be more efficient.” The airline puts the payoff at roughly a million gallons of fuel a year and tens of thousands of hours saved in the air.
Those are real numbers. They are also numbers produced inside a single airline’s decision loop, where one company controls the dispatchers, the fleet, the schedule, and the incentive to actually take the suggestion. Alaska could adopt Flyways because Alaska is the only party in the room.
The Distance Between a Dispatch Desk and a National System
The national airspace is not one company’s decision loop. It is a negotiation among carriers whose interests directly conflict, run by controllers at facilities that were never built to see each other’s traffic, under a regulator that has to justify every choice to Congress and to the flying public.
Bedford is refreshingly blunt about the starting condition. “What we’re seeing is a system that is woefully unorganized,” the administrator said. “We don’t have in any of our facilities the ability to see the entire national airspace picture in a unified way.” He is right, and that diagnosis is the strongest argument for the contract. The FAA’s problem today is not that its controllers make bad calls. It is that they make locally correct calls with a partial picture, and the sum of locally correct calls is a delay cascade that ripples from Newark to Denver by dinnertime.
But a shared picture is not the same product as a routing recommendation, and the honest version of this contract is that the FAA is buying both at once from a company that has shipped one of them. ASI’s client list has grown past Alaska to include Delta, United, the Army, and the Air Force, which is meaningful validation. Every one of those customers is still a single operator optimizing its own aircraft.
Why the Agency Skipped Its Own Program
The part of this story that nobody at the announcement wanted to dwell on is what the $875 million is replacing. The FAA has spent two decades and tens of billions on NextGen, the modernization program that was supposed to have delivered exactly this capability by now. When the agency turned around and handed a twelve-year mandate to a startup founded in the last decade, it was making a statement about its own program that no press release would ever put in words.
That statement has a price. Twelve years is longer than most administrations, longer than most FAA administrators last, and long enough that the agency is now structurally dependent on one vendor for the data backbone of its command center. Palantir and Thales lost the bid, which means the obvious second sources are the companies the FAA just told they were not good enough. Procurement people call this lock-in. Everyone else calls it having no plan B.
There is also a timing problem the contract cannot solve. The reason American flying got worse was not routing math. It was staffing, and it still is. When the agency ordered flight cuts at 40 major airports rather than run traffic with the controllers it had, it was rationing a workforce, not a dataset. Predictive software makes an understaffed system slightly more efficient. It does not staff it.
The Word Nobody Wants to Say
FAA officials keep repeating that SMART is meant to help controllers, not replace them. ASI chief executive Phillip Buckendorf frames the shift as behavioral rather than technical: “It’s actually a really fundamental change, a massive behavior change.”
Both things can be true and still leave the interesting question unasked. A system that tells a controller what the sky will look like in ninety minutes changes who is actually deciding. If the recommendation is right nine times in ten, the tenth becomes very hard to override, because overriding it means writing down why you disagreed with the machine. Alaska’s dispatchers reportedly pushed back at first over precisely this, worried Flyways would run itself. They came around. Dispatchers work for an airline. Controllers work for the public, and the public’s tolerance for “the software suggested it” as an explanation after an incident is going to be low.
None of that is an argument against the contract. The status quo, where a regulator cannot see its own airspace in one picture, is worse than a vendor-supplied picture. It is an argument for the FAA publishing, before the fall deployment, exactly what SMART recommends, how often controllers override it, and what happens on the days it is wrong. After a fatal ground collision at LaGuardia in March, the agency’s instinct was to explain the sequence of decisions. That instinct gets much harder to satisfy when part of the sequence happened inside a model.
The proof case here is genuinely good. It is also a million gallons of jet fuel at one airline, and the thing being purchased is the sky over 330 million people. Fall is not far away.