The Trump administration’s General Services Administration said Thursday that federal agencies can now buy access to xAI’s Grok models through a new OneGov agreement that runs to March 2027. The deal adds Grok 4 and Grok 4 Fast to the government’s menu of AI services and promises dedicated engineering support to help agencies embed the tools inside daily workflows.
In a statement, Elon Musk called xAI’s systems “the most capable in the world” and said opening Grok to every department will help the government “innovate faster and accomplish its mission more effectively than ever before.”
The agreement sets a nominal $0.42 per agency access price for an initial 18-month period, according to GSA, positioning Grok as a low-cost option agencies can test quickly before scaling to enterprise subscriptions that meet federal security baselines.
GSA framed the offer as the longest OneGov AI term to date and said it fits squarely inside the administration’s effort to expand AI use across civilian and defense bureaus. The xAI team will provide hands-on support to accelerate adoption and integration.
What the government is buying
Under the arrangement, agencies can procure Grok through established GSA channels rather than navigating one-off contracts. That matters inside a sprawling bureaucracy where procurement cycles can stretch for months.
The package includes the latest Grok reasoning models and an “upgrade path” toward compliance with the Federal Risk and Authorization Management Program (FedRAMP), which standardizes security authorizations for cloud services across the government. GSA said the aim is to simplify pilots, shorten time to production, and keep agency costs predictable.
Josh Gruenbaum, commissioner of GSA’s Federal Acquisition Service, said broad access to capable models is “essential” to building a government that is more efficient and accountable to taxpayers. He linked the announcement to the administration’s promise to “win the AI race” and to modernize core operations.
Gruenbaum added that xAI “stood out” for “world class technical talent,” and argued that the values embedded in frontier systems should reflect Western, American principles. Agencies, he said, are in the current phase of human augmentation, with autonomous agents handling more tasks in the near future.
xAI cofounder Ross Nordeen billed the offer as “Grok for Government,” saying the company will provide a dedicated engineering team to ensure mission outcomes. He described the engagement as working “hand in glove with the entire government” to deploy AI while learning agency needs and tailoring the models accordingly.
Why this fits the Trump AI playbook
The announcement arrives as the White House pushes a whole-of-government AI action plan and leans on GSA to make frontier systems easier to buy and easier to secure. In January, the administration launched the Department of Government Efficiency to streamline technology and procurement, followed by a July AI action plan that set timelines for agencies to adopt automation where it can demonstrably improve service delivery. Thursday’s deal gives departments a ready-made option to meet those timelines.
The tie-up also reflects the competitive pressure inside the federal marketplace. Major model developers are vying for seats on contract vehicles that promise large, multiyear workloads. The GSA said the OneGov strategy is designed to avoid agency-by-agency fragmentation and to give buyers access to several leading models through common terms.
Reuters reported the xAI agreement is effective immediately and sits alongside other model options already available to agencies. For additional detail, see GSA’s announcement and Reuters’ coverage of the contract and its March 2027 horizon.
Potential use cases and limits
Grok’s pitch centers on high-volume knowledge work, like drafting, summarizing, and answering domain questions against agency data.
Early adopters are expected to focus on customer service, records triage, claims processing, cybersecurity workflows, and code assistance. Because Grok 4 emphasizes reasoning benchmarks, program offices will test whether it handles multi-step tasks with fewer hallucinations and lower supervision costs than prior generations.
There are limits. Agencies must route deployments through privacy, security, and records frameworks. Sensitive workloads will require FedRAMP-authorized environments, strict data-handling rules, and audit trails. Procurement officers will also weigh model performance against risks of inaccurate or biased outputs, which advocacy groups and independent researchers continue to flag across large language models.
GSA framed the OneGov model as a way to evaluate those tradeoffs quickly without locking into single-vendor paths.
Musk, the federal market, and the wider AI race
For Musk, the pact validates xAI as a federal supplier after a year of rapid model iteration and heavy investment in compute. The agreement follows parallel government efforts with other vendors and coincides with the private sector’s multibillion-dollar Stargate data center build-out led by OpenAI and partners. The combined trend is clear. Agencies want to experiment with multiple models, and model makers want durable, recurring public-sector demand.
The United States is not alone. China and allied democracies are also racing to secure compute, talent, and energy for large-scale AI.
Gruenbaum said that reality argues for aligning deployment with American values, and for working with allies to shape the norms embedded in the systems that will lead the world. In practice, that means agencies will pay as much attention to content controls, safety tools, and provenance features as they do to raw accuracy scores.
