The national conversation about artificial intelligence has focused on how it should be regulated, whether America can build enough data centers and generate enough power to run them, and whether today’s extraordinary investment boom can last.
Much less attention has gone to another piece of infrastructure AI will increasingly depend on: spectrum.
As AI moves into manufacturing, logistics, energy, and other parts of the physical economy, machines will need to exchange large amounts of information reliably and in real time. That requires wireless capacity. Unlike computing power or electricity generation, spectrum is finite. We cannot build more when demand rises. Instead, we count on federal regulators to manage the country’s limited spectrum resources effectively.
The National Telecommunications and Information Administration is preparing to study options for opening the 4.4–4.94 GHz band, more than 500 megahertz of prime mid-band spectrum that today supports a wide range of military and other federal operations. Congress and the president have directed NTIA to examine whether portions could be reallocated for commercial licensed use as part of America’s push toward 6G.
My firm, Valo Analytica, recently analyzed the band, and we found that spectrum sharing deserves to be part of NTIA’s consideration. Some industry advocates have suggested that the right path for the 4.4 GHz band is the traditional “clear and auction” method of making spectrum available, but this approach can create serious disruption for government users providing critical services, forcing them to relocate elsewhere, buy new equipment and endure uncertainty in their operations.
There is another way.
Sharing spectrum between government and commercial users could bring more usable spectrum online sooner, put it within reach of more businesses, and avoid a significant portion of the cost and disruption of relocating federal systems.
Those advantages matter for AI.
A manufacturer may need enormous wireless capacity inside one production complex. A port may need dedicated connectivity across its terminals. A utility may need a highly reliable network around a handful of facilities. These businesses need intensive spectrum access where they operate. They generally do not need exclusive control of the same frequencies across an entire state, let alone the entire country.
Sharing can make spectrum available on a more localized basis, widening access to manufacturers, logistics companies, utilities, research institutions, and others that will build and deploy AI-enabled systems.
It could also reach them much sooner.
CTIA recently released its own 4 GHz report, with a roadmap that does not reach spectrum identification until 2028, FCC service rules until 2029, and an auction until 2030 — before accounting for the time required to transition federal systems.
Our analysis finds that sharing could allow commercial use to begin without first waiting for every incumbent to move. It could also make more of the band available where federal systems are not using it, rather than limiting commercial users to whatever portion is eventually cleared.
For an AI economy moving this quickly, years matter.
I have spent much of my career working on spectrum-sharing systems. The engineering is more sophisticated than simply deciding that either the government or commercial users get a frequency.
A Navy operation concentrated around a port can be protected where it operates. A weapons range can receive protection within a defined area. Fixed communications links can be protected through automated coordination. Different missions can receive different levels of protection without requiring an entire band to remain unavailable everywhere.
We already do versions of this at scale. More than 440,000 CBRS base stations operate today alongside protected military radar, while automated frequency coordination manages commercial access around incumbent links in the 6 GHz band. The details at 4 GHz will be different, but the underlying capability is proven.
There is also a fiscal reason to examine sharing.
Spectrum auctions tend to be judged by their headline proceeds. But if federal systems must first be relocated, those transition costs must be accounted for. Federal law generally conditions an auction of eligible federal frequencies on recovering at least 110% of estimated relocation or sharing costs.
The relevant comparison is therefore net, not gross, revenue.
Our analysis finds that a shared, auctioned tier could carry much lower transition costs, reach the market sooner, and attract a broader range of bidders through smaller licenses. A cleared exclusive auction might generate a larger headline figure while producing a smaller net return after relocation costs.
Congress explicitly left room to consider both paths. The spectrum provisions enacted in 2025 direct NTIA to identify 500 megahertz of federal spectrum for nonfederal use, for shared federal and nonfederal use, or for a combination thereof, for full-power, commercially licensed use.
That should be the starting point for NTIA’s 4 GHz study. Given the scarce public resources at stake, NTIA should carefully consider the options for this band.
WE FAILED OUR CHILDREN ON SOCIAL MEDIA. AI IS OUR LAST CHANCE TO GET IT RIGHT
The agency should model both exclusive and shared approaches and publish the comparison: how much usable spectrum each would produce, how quickly businesses could access it, what each would cost, what each would net the Treasury, and how well each would protect federal missions.
America is already thinking hard about whether it has enough power, computing capacity, and capital for the AI economy. It should bring the same urgency to the spectrum and make sure every viable option for using this finite resource is on the table.
Andrew Clegg is co-founder and chief technology officer of Valo Analytica.