A dozen radios and no network
Watch a nine-person team walk out the gate. Count the transmitters. Every phone carries cellular, WiFi, and Bluetooth. Two people wear watches that talk to those phones. Someone has a tablet. Someone has a laptop. That is a dozen radios or more moving in a loose column, and a 1990s battalion staff would have gone to war for that much RF capability. The team treats almost all of it as dead weight.
The comms plan rides on one or two purpose-built radios. When those work, the team has a voice net and maybe a thin data link. When those fail, the team has hand signals. Meanwhile the dozen other radios sit in pockets, fully charged, each one capable of reaching its neighbors, none of them enrolled in anything. The most redundant network the team will ever own is already distributed across their bodies. It just has no idea it exists.
This is not a hardware gap. The hardware was bought at retail, works out of the box, and gets replaced every two years on personal upgrade cycles the government never has to fund. The gap is that no software treats those devices as one coordinated fabric. That gap is the whole story.
What the programs bought
Tactical radio programs have consumed billions of dollars across decades of development. The most ambitious of them promised a universal software-defined radio and was cancelled after more than a decade without fielding one. The radios that did reach the force are capable and hardened, and they arrive years after the requirement was written, at a unit cost that guarantees scarcity. A team gets two. A team never gets twelve.
Scarcity shapes behavior. When the network lives in one box, the network dies with that box. Ukraine made the opposite bet under fire. CSIS documented a military digitizing on whatever hardware its soldiers already owned, building situational awareness tools that ran on commercial devices because those were the devices actually present at the front. The lesson was not that consumer gear beats military gear on any single spec. The lesson was that the fielded device beats the unfielded one every single time.
And the spectrum those programs assumed is gone. The Washington Post reported in May 2024 that Russian jamming was degrading GPS-guided American weapons in Ukraine, systems built by the best-funded procurement pipeline on earth. A network that depends on one link, one band, or one box is a network with a single point of failure standing in front of a peer jammer. Diversity of paths is not a luxury anymore. It is the survival requirement.
The hobbyists proved the physics
While the programs iterated, hobbyists built mesh networks for fun. A $30 LoRa board reaches kilometers on milliwatts. Volunteer communities strung those boards into off-grid text networks spanning cities, with store-and-forward relaying built in, running on hardware a teenager buys with allowance money. Hikers pass messages through nodes they will never see. The physics of a self-healing, low-power, carried mesh is settled. Amateurs settled it.
So the honest comparison is uncomfortable. On one side, a procurement system spending billions to field scarce radios. On the other, weekend builders getting multi-kilometer resilient mesh from parts that cost less than a case of water. The hobbyist networks are toys by military standards. No priority scheme, no mission boundaries, no guarantees about what arrives when. But they demonstrate the floor, and the floor is already high enough to matter.
What separates a toy mesh from a mission network is software discipline. Store-and-forward with delivery guarantees. Traffic priority, so the medevac request preempts the resupply chatter. Graceful degradation, so losing three nodes narrows the network instead of killing it. Enrollment and exclusion, so the fabric knows exactly which devices belong to this mission and no others. None of that requires new radios. All of it requires software that does not exist.
The nervous system is already fielded
This post joins a thread. Local AI needs a local network, because a model at the edge is only as useful as its reach to the sensors and teammates around it. Mortal data needs a network that respects mission boundaries, because data that must die at mission end cannot ride on infrastructure that remembers forever. Both arguments dead-end without a fabric to run on.
The fabric is already walking around. Every dismounted team is a mesh network that nobody switched on, a dozen paid-for radios waiting for the software that enrolls them, prioritizes them, and lets them fail gracefully. Whatever fills the gap in the stack will have to treat that carried mesh as its nervous system, because there is no other network guaranteed to be present when everything else is jammed, severed, or left behind. The hardware shipped years ago. Someone has to ship the discipline.