In August 2020, the U.S. Department of Defense's Joint Artificial Intelligence Center awarded Scale AI a blanket purchase agreement for data labeling and AI infrastructure work. The five-year BPA carried a ceiling of $249 million, a figure set by the JAIC and published in the government's procurement database. The contract has since expired. The JAIC itself was dissolved in 2022 and absorbed into the Chief Digital and Artificial Intelligence Office. Scale AI went on to win a far larger award from the CDAO in 2023: a deal to evaluate large language models, publicly valued at over $1 billion by the Pentagon.
The BPA was one of the largest known contracts for commercial AI data work inside the DoD at the time. It let military components place orders against pre-negotiated terms instead of striking separate agreements. For the Pentagon, the award signaled a shift from pilots toward embedding commercial machine learning tools in operational workflows. Scale AI, founded in 2016 by Alexandr Wang, had built its business supplying labeled datasets for training models. The JAIC contract validated that model and gave the startup a defense foothold it would later widen substantially.
What the JAIC Wanted and How It Procured It
The JAIC stood up in 2018 with a mandate to speed AI adoption across the DoD. It operated as a central hub, picking high-impact use cases and funding projects that individual branches lacked the resources to pursue alone. By 2020 the center had launched work in predictive maintenance, humanitarian assistance, cybersecurity, and disaster response.
Data labeling was a persistent choke point. Defense AI applications need accurately annotated datasets that are secure and reflect operational contexts. Off-the-shelf commercial platforms often could not meet security requirements or lacked domain fluency. Scale AI pitched itself as a supplier that could deliver labeled data at volume while complying with DoD security protocols.
The contract used a BPA structure, not a sole-source lock or a winner-take-all competition. A blanket purchase agreement allows multiple orders over a period of time without renegotiating terms for each task order. The $249 million ceiling, as stated in the award notice, represented the maximum the DoD could spend. It was not a guaranteed minimum, and whether the full amount was drawn down has not been publicly confirmed.
Reactions Inside and Outside the Company
The award drew notice from both advocates and critics of defense AI. In Silicon Valley, it registered as a milestone for startups willing to work with the military establishment.
The deal arrived three years after the employee protests at Google over Project Maven, a DoD drone-imagery analysis contract that Google declined to rebid. Scale AI's leadership expressed no hesitation. Alexandr Wang said the company was proud to support national security applications of AI. Some employees and outside observers raised ethical concerns about military use of the technology, but no large-scale internal protest tied directly to this BPA was ever reported. The company did not face the kind of workforce backlash that hit Google.
The broader policy argument continued. Critics contended that data labeling for defense purposes could feed autonomous weapons or enable targeting decisions with thin human oversight. Supporters answered that the DoD already deploys AI for defensive missions and disaster relief, and that a commercial partner subject to public scrutiny is preferable to classified in-house development.
Scale AI's Trajectory Before and After the JAIC Contract
The startup at the time of the award
When the BPA was signed, Scale AI was a four-year-old company that had raised venture funding from Accel, Y Combinator, and others. The business was valued at roughly $1 billion in a 2019 funding round. The JAIC agreement offered a recurring revenue stream from a creditworthy counterparty and a proof point that the firm's offerings could clear government bars.
From labeling to LLM evaluation
After the award, Scale AI kept raising capital and broadening its product line. It moved beyond data labeling into model evaluation, synthetic data generation, and AI testing. In 2023, the CDAO awarded Scale AI a contract for evaluating large language models that the Pentagon priced at over $1 billion. That award dwarfed the 2020 JAIC deal and reflected the government's surging demand for AI testing infrastructure.
The contract's end
The five-year JAIC BPA period that began in August 2020 has since run its course. Whether the DoD exhausted the $249 million ceiling is not publicly established. The BPA structure meant spending could have landed anywhere beneath that cap if demand failed to materialize.
What Happened to the JAIC
The reorganization
The office that signed the BPA no longer operates on its own. In 2022 the DoD merged the JAIC, the Defense Digital Service, and the Chief Data Officer into the Chief Digital and Artificial Intelligence Office. The CDAO received broader authority over data, AI, and digital transformation across the department.
Why the Pentagon folded the JAIC
The restructuring grew from a recognition that the JAIC's centralized model had limits. The Army, Navy, and Air Force had each developed their own AI capabilities, and the DoD wanted to cut duplication. The CDAO was designed to set standards, coordinate investments, and furnish shared platforms while letting the services run their own AI programs.
What the dissolution meant for Scale AI
The transition from JAIC to CDAO did not interrupt Scale AI's relationship with the Pentagon. The company's subsequent contracts were with the CDAO. The 2020 BPA stayed in effect until its term expired, even though the office that issued it had been folded. The episode shows how rapidly the Pentagon's AI governance changed in a period when commercial AI capabilities were also accelerating.
