Many operations during US Space Command’s March 8-12, 2021 Global Lightning exercise were based in the command’s joint operations center. (US Space Command photo: Lewis Carlyle)
WASHINGTON — The Pentagon’s outreach arm to Silicon Valley, the Defense Innovation Unit, has selected San Francisco-based Scale AI to lead what it calls Thunderforge, a prototype project to build AI to speed up large-scale military planning, the company and DIU announced this morning.
Scale AI will work with fellow defense tech upstart Anduril, IT giant Microsoft, and other as-yet undisclosed “global” subcontractors, the company and DIU said. The new AI toolkit will combine Anduril’s Lattice data-sharing system with Large Language Models (LLMs) developed by both Microsoft and Scale AI itself.
The first prototype tools will go to the four-star headquarters tasked to deter America’s most powerful adversaries: Indo-Pacific Command, which the Trump administration has identified as its top priority in a new Cold War with China; and European Command, whose decades-long focus on Moscow has been complicated by Trump’s reversal of previous anti-Russian policies.
The Thunderforge project is just one piece of a much larger multi-front effort to use AI algorithms, big data, and long-range communications to coordinate US and allied forces across the military “domains” of land, air, sea, space, and cyberspace. The jaw-breaking official jargon for this mega-project is Combined Joint All-Domain Command & Control (CJADC2).
All the armed services and many defense agencies have been building their own, hopefully compatible pieces of this AI-powered global meta-network, coordinated by Pentagon’s Chief Digital & AI Officer. (However, the Trump administration is reportedly considering major changes to the role of CDAO, a Biden-era creation). Working closely with CDAO, DIU has played a major role in linking all these efforts to cutting-edge companies in the commercial sector, whose latest and greatest AI tools the military wants to adopt and adapt for its own purposes.
RELATED: Tech bros to the rescue? Reagan Institute 2025 scorecard dings Pentagon on ‘scaling’ innovation
Those military applications can be remarkably mundane, in stark contrast to the killer-robot apocalypses of pop culture. Planning a major combat operation, or even a peacetime redeployment, requires going through vast amounts of data: latest locations of potential targets, unit readiness reports, transport routes and timetables, the load-carrying capacities of road bridges and seaports, stockpiles of everything from ammunition to spare parts, and on and on and on.
Today, much of that data is scattered across multiple, incompatible, and increasingly geriatric computer systems, which means human staff officers have to labor for hundreds of hours to pull it all together, let alone analyze it and calculate an optimal course of action. So the Pentagon is intensely interested in any kind of AI that can accelerate this laborious data-crunching and give everyone at headquarters more time to actually think.
“Today’s military planning processes rely on decades-old technology and methodologies, creating a fundamental mismatch between the speed of modern warfare and our ability to respond,” DIU’s Thunderforge program lead, Bryce Goodman, said on the agency’s blog. “Thunderforge brings AI-powered analysis and automation to operational and strategic planning, allowing decision-makers to operate at the pace required for emerging conflicts.”
Specifically, DIU and Scale told Breaking Defense, Thunderforge will automate many traditional staff processes, assist in drafting plans, and even run AI wargames of alternative courses of action. It will also handle classified information at multiple levels.
While today’s announcements didn’t disclose many details, including the amount of funding the government will provide, recent military experiments suggest that the Large Language Models provided by Scale AI and Microsoft will serve two major functions: extracting key details from masses of written reports and other documents — so-called “unstructured” data, which is notoriously hard for older forms of AI to make sense of — and generating draft planning documents for staff officers to review.
But, as Scale emphasized in its announcement, these AI tools will be “always under human oversight.”
