On May 8, Scale AI quietly landed a much larger Pentagon commitment.
The Defense Department raised the ceiling on its contract with the company from $100 million to $500 million, a fivefold jump.
The news surfaced through ClearanceJobs, a channel aimed at cleared defense-intelligence professionals, and drew little mainstream attention. But the wording around the contract matters more than the headline number.
More money, but a sharper change in language
A contract memo obtained by ClearanceJobs put it this way:
"This is less about future experimentation and more about embedding AI into everyday defense operations."
The line came from CDAO, the Defense Department's Chief Digital and Artificial Intelligence Office. The key contrast is between "experimentation" and "everyday operations."
Until now, U.S. military AI adoption has largely been fragmented. The Army ran its own projects, the Navy ran its own, and the Air Force might sign a proof of concept with a Silicon Valley company to see what worked. CDAO was created in part to pull those scattered efforts into a more coherent structure.
By moving the ceiling from $100 million to $500 million, the Pentagon is effectively telling acquisition offices that this is no longer budgeted like an experiment. It is starting to look like infrastructure.
What the $500 million covers
The public scope of the contract spans several areas:
- Computer vision, including target recognition and satellite imagery analysis
- Generative AI decision support for situational analysis and option generation for commanders
- Data operations that turn scattered military data into trainable material
- Classified generative AI tools for intelligence users
- Data-labeling pipelines that support autonomous systems and targeting systems
The last item is the one to underline: data labeling for autonomous and targeting systems.
In plain terms, that means training data for unmanned systems that can fly, identify targets and, in some contexts, fire on their own. Anduril's Lattice and Shield AI's drone systems depend on this kind of data. Scale AI's work is to turn raw video, imagery and radar returns into human or semi-automated annotations so models can distinguish a tank from a civilian vehicle or a decoy.
On the same day, Anduril also received a contract modification of its own. A Space Force mesh-networking project doubled from a $100 million ceiling to $200 million and runs through the end of 2027. The listed work locations were Colorado Springs and Costa Mesa: the former is Space Force headquarters, the latter Anduril's Silicon Valley base.
OTA is not new, but it is becoming the default lane
The memo included another important detail: the deal uses an OTA, or Other Transaction Authority, structure.
OTA is the Pentagon's faster procurement channel, designed to bypass parts of the traditional military-acquisition process. A conventional program can involve an RFP, FAR procurement rules and DCAA audits, and it can take two or three years to move from launch to fielding. Many Silicon Valley companies cannot absorb that timeline. OTA gives CDAO the authority to sign and move capabilities toward units within months.
OTA was previously associated mainly with smaller innovation projects. A $500 million Scale AI deal using OTA turns fast procurement from exception into routine practice.
People who know defense acquisition understand the pattern: once an OTA contract works, the next large deal is likely to use OTA again. With that opening in place, much of the defense-AI money can move through the same channel.
TurbineOne's move to Virginia is part of the same signal
Another related development landed the same day: AI defense company TurbineOne moved its headquarters from California to Fairfax County, Virginia.
The move brings 22 new jobs and $424,000 in state incentives. The company also opened a research and training facility in Chantilly called T1 Edgeworks.
TurbineOne builds edge AI: models that keep running when communications are jammed or networks are cut. That means local inference for a frontline soldier's handheld device, a drone that has lost its uplink, or an isolated sensor operating on its own.
Why move from Silicon Valley to Virginia? The answer is next door to Fairfax County: the Pentagon.
It is a signal that defense-AI companies are physically moving into the Washington, DC orbit. A year ago, a company could choose between Palo Alto and Silver Spring. Now, if it wants a major defense contract, a Beltway address is becoming harder to avoid.
How the $500 million changes the defense-AI market
For Scale AI, the expanded contract helps its valuation story. The company had taken pressure from litigation tied to data-labeling workers and PTSD claims; a major government contract steadies the core business.
The larger impact is on the new defense-technology ecosystem. For years, AI defense budgets were divided mainly among incumbents such as Lockheed Martin and Northrop Grumman, along with Palantir. Now CDAO is using OTA to open a direct path for Silicon Valley-born companies such as Scale AI, Anduril and TurbineOne. Traditional contractors either adapt or risk being routed around.
Lockheed has already been discussing internal partnerships with Anthropic and OpenAI, while Northrop recently acquired a small company building military LLMs. Those moves all happened in the past six months.
$500 million is not a huge number inside a Defense Department budget of more than $800 billion. But the language around the contract, especially "everyday operations" and "less about future experimentation," matters more than the amount.
The U.S. military's posture toward AI is shifting from "let's try it" to "this is how it should be used."
Sources: CocoLoop, Pentagon Expands Scale AI Deal to $500M and AI Defense Firm TurbineOne Moves HQ to Northern Virginia (ClearanceJobs)