Maven is now a $1.5B program of record. It still serves HQ.
In March 2026, Defense Secretary Feinberg issued a directive making Maven Smart System the Pentagon's primary AI platform, consolidating AI investment under a single program of record. The budget trajectory: $103M in FY2025, $240M in FY2026, over $1.5B requested for FY2027. Combined with more than $2B requested for CJADC2, the Pentagon is betting nearly $4B per year on AI-enabled command and control.
That money buys a system designed for combatant commands and operations centers. Maven ingests satellite imagery, signals intelligence, and multi-source feeds, then delivers fused intelligence products to analysts and commanders with network access. It is a cloud and HQ tool. The architecture assumes reliable connectivity, centralized compute, and users who sit at workstations.
The gap starts where the link ends
A dismounted operator 40km from the nearest relay, running a surveillance position in contested spectrum, gets nothing from Maven when the SATCOM link drops. No threat assessment, no pattern analysis, no ISR fusion. The $1.5B investment evaporates at the point where AI matters most: the operator making decisions under fire without reachback.
The January 2026 DoD AI strategy memo requires AI that operates 'on-board, in real time, and often without any sort of connectivity or centralized compute resources.' Maven does not meet that requirement. It was not designed to. Ukraine and the South China Sea have already disproven the assumption that connectivity will be available when it counts.
Scaling budget does not scale reach
Doubling Maven's budget does not push AI capability one meter closer to the operator. The architecture is centralized. More money buys more cloud compute, more analyst seats, more data fusion at the operations center. It does not buy local inference on a MacBook in a hide site or threat detection on a phone at a checkpoint.
The Foreign Affairs Forum covered America's algorithmic arsenal spending. The money goes to echelons that already have bandwidth, power, and compute. The lower echelons, where decisions happen fastest and links are worst, get nothing. A different architecture is required.
What the operator needs when Maven goes dark
When the link drops, the operator needs local object detection on the device. Local threat analysis correlating cameras with RF. Local transcription of intercepted comms. Local mission state so the team knows where everyone is. All generating evidence with chain-of-custody metadata, not disappearing when the connection times out.
EdgeLance runs that stack on consumer hardware. A MacBook Pro is the command node with full inference. iPhones are field nodes with detection and mesh relay. It does not require Maven, does not compete with Maven, and does not replace what Maven does well. It covers the operator Maven cannot reach. The ISR gap between enterprise platforms and the tactical edge is not closing.