Introduction

On May 21, 2026, the General Services Administration announced a new OneGov agreement with Snowflake, the AI Data Cloud company, making Snowflake's AI and cloud-based data products available to all US federal agencies at pre-negotiated prices. The deal gives every new federal Snowflake customer a 20% discount on compute services, with the potential for those discounts to reach 50% as aggregate government usage grows, and approximately 27% off storage — all available through September 30, 2027. The announcement was framed by GSA Administrator Edward Forst in terms of the White House's AI Action Plan: "With stronger cross-agency data capabilities, we can accelerate AI tools tailored to each agency's mission. We're already seeing promising early projects, and through OneGov, GSA is delivering a unified federal approach that saves taxpayer dollars and increases value for the American people."

The Snowflake deal is the latest agreement in the OneGov initiative, which GSA launched in April 2025 and has already extended to more than twenty other companies, including Microsoft, SAP, Palo Alto Networks, OpenAI, Anthropic, and Perplexity. Collectively, OneGov has identified more than $1.15 billion in savings for the federal government in its first year of operation. What the Snowflake agreement adds is something qualitatively different from a software tool or a security platform: it adds the data layer. If AI tools from OpenAI and Anthropic are the intelligence, Snowflake is the substrate — the platform that determines whether the data those models reason over is unified, governed, and actually accessible to the agencies trying to use it. That is a significant distinction, and it is the reason the Snowflake deal deserves more attention than the headline discount figures suggest.

What Snowflake Actually Does and Why the Government Needs It

Snowflake describes itself as an AI Data Cloud — a platform that allows organisations to store, manage, and analyse large datasets across multiple cloud environments from a single interface. The more precise description of what it solves is best understood by examining what the federal government looks like without it.

The US federal government generates an almost incomprehensible volume of data across more than 130 civilian agencies. The IRS processes hundreds of millions of tax records. The Social Security Administration manages accounts for tens of millions of beneficiaries. The Department of Veterans Affairs holds health records for more than 9 million veterans. The Department of Homeland Security, the Census Bureau, the Department of Agriculture, and every other federal agency all maintain their own data systems, built over decades, in different formats, using different standards, sitting on different infrastructure. None of these systems was designed to talk to the others, because each agency has historically procured its own technology on its own terms, building what the industry calls data silos: isolated stores of information that cannot be shared, searched, or analysed across agency boundaries without enormous manual effort.

Snowflake's architecture is specifically designed to dissolve those silos. Its multi-cloud platform allows data from different systems, maintained by different agencies, to be made available for analysis without physically moving the data to a central location — a process that would raise enormous legal, privacy, and security concerns if attempted conventionally. Data stays where it is, governed by the rules of the agency that owns it, but becomes searchable and analysable as part of a unified view that cuts across organisational boundaries. For federal AI deployments, this is not a nice-to-have feature. It is the prerequisite that determines whether AI tools can function at anything approaching their potential.

Snowflake CEO Sridhar Ramaswamy articulated the competitive claim directly: "Federal agencies are seeking efficiency in cost, enterprise scaled performance, intuitive design driven tools for the workforce and simplicity in contracting — we are the only multi-cloud data platform that can meet this charge on day one." The claim to uniqueness is contestable — Oracle, Microsoft Azure, and Google Cloud all offer comparable enterprise data cloud capabilities — but the structural argument is sound. A government that cannot make its data interoperable across agencies cannot build AI systems that work across agencies, and AI systems that do not work across agencies cannot address the cross-agency coordination problems that are among the most expensive and consequential in government operation.

Snowflake already holds the security credentials required for federal deployment. The company gained high-security certification for government use through the Federal Risk and Authorization Management Program, known as FedRAMP, in AWS GovCloud in 2023 and for Microsoft Azure Government in 2025. FedRAMP is the federal government's standardised approach to security assessment, authorisation, and continuous monitoring for cloud products — it is the baseline credential without which a commercial cloud company cannot legally operate with federal data. Having FedRAMP High authorisation on both major government cloud environments means Snowflake can handle not just unclassified government data but also controlled unclassified information, which covers a substantial majority of the sensitive but non-classified data that federal agencies work with daily.

How This Fits the Federal AI Modernisation Strategy

The Snowflake agreement did not arrive in isolation. It is the data infrastructure layer of a broader federal AI strategy that has been assembling piece by piece since the Trump administration's first term ended a previous executive order on AI and replaced it with an AI Action Plan focused on accelerating adoption rather than regulating it.

The OneGov initiative's design principle is to treat the entire federal government as a single enterprise customer rather than as hundreds of separate procurement entities, each negotiating its own contracts on its own terms. Before OneGov, a federal agency seeking to deploy a commercial AI tool had to navigate procurement processes that could take 18 to 24 months from initial interest to signed contract, during which the technology landscape would have changed substantially and the urgency that motivated the original procurement would have dissipated. OneGov pre-negotiates the commercial terms so that any agency can access an approved tool immediately under the umbrella agreement, reducing the procurement cycle from months to days.

The reported result of almost 3.4 million federal employees now having access to emerging technology tools through OneGov represents a scale of technology deployment that no previous federal procurement initiative had achieved in comparable time. The addition of Snowflake to the OneGov portfolio means that those 3.4 million employees now have not just access to AI language models but access to a platform that allows those models to reason over unified, cross-agency data rather than the fragmented information available in each agency's own systems. That is a qualitative improvement in what federal AI tools can actually do, not just a quantitative expansion of the number of employees who can access them.

GSA officials have also been honest about the longer-term ambition. Officials have described the discounted deals available through OneGov as a first step toward longer-term direct contracts, and have stated that the ultimate goal is to parlay these initial agreements into longer-term direct contracts with preferred terms. The Snowflake agreement, available through September 2027, is explicitly positioned as the opening phase of a relationship that could become permanent infrastructure for the federal data architecture, not a temporary cost-saving measure.

What Federal Workers Will Be Able to Do

The practical capability change that the Snowflake agreement enables for federal workers is most clearly illustrated by the examples that the administration itself has highlighted. An analyst at the Department of Health and Human Services who previously needed weeks of manual data extraction work to compare outcomes across Medicare claims data and Veterans Affairs health records can, on a Snowflake-enabled platform, run that analysis in hours against a governed, unified data view that both agencies contribute to without either surrendering control of their own underlying systems. A Border Patrol logistics officer who needs to understand supply chain status across multiple vendor systems can query a unified data layer rather than assembling manual reports from disconnected spreadsheets.

The AI dimension compounds those gains. When an AI agent can access a unified, well-governed data platform rather than calling individual agency APIs one at a time, its ability to synthesise cross-agency information into actionable analysis changes qualitatively. An AI agent asked to identify patterns in federal contracting that suggest fraud risk needs data from the Federal Procurement Data System, the System for Award Management, agency financial systems, and potentially law enforcement databases. Without a platform like Snowflake providing unified access under consistent governance rules, that agent is calling different systems with different formats, different access protocols, and different update frequencies, and producing analysis that reflects all of those inconsistencies. With unified access, the analysis can be run against clean, current, consistently formatted data, and the agent's outputs are proportionally more reliable.

The GSA administrator's framing of "promising early projects" suggests that deployments in this direction are already underway, though no specific case studies have been publicly identified beyond the general categories of data streamlining and mission effectiveness that the official announcement described.

The Privacy and Security Questions That the Announcement Did Not Address

The official framing of the Snowflake deal emphasises efficiency, cost savings, and mission effectiveness. What it does not address, and what privacy and security advocates have increasingly focused on in the broader context of federal AI data deployment, is the set of risks that accompany the consolidation of federal data on a commercial cloud platform.

Snowflake's security credentials are real. FedRAMP High authorisation on both AWS GovCloud and Microsoft Azure Government represents a rigorous security baseline that most commercial cloud products do not achieve. But FedRAMP certification addresses the security of data at rest and in transit — it does not address the governance questions about who, across the federal enterprise, has access to the unified data view that Snowflake provides. One of the structural reasons that federal data has historically been siloed is not only technical inertia but also considered data governance: different agencies have different legal authorities to collect and share data, and those authorities do not automatically extend to cross-agency data sharing arrangements simply because a technical platform makes such sharing possible.

The broader context in which the Snowflake deal lands is not a reassuring one from a civil liberties perspective. The May 2026 Tech Policy Roundup from Tech Policy Press reported that in the same month as the Snowflake deal, Public Citizen raised concerns about the federal government's use of AI in the rulemaking process, and that civil society organisations continued to flag risks from the expansion of government data access without commensurate expansion of legal oversight. Caitriona Fitzgerald of the Electronic Privacy Information Center testified before Congress in 2026 that proposed federal data minimisation legislation — the baseline legal framework for limiting government data collection — had a "core weakness" in that its restrictions applied only to purposes a company disclosed in its privacy policy, which is a standard that existing law already nominally required. The gap between what the law permits and what a technically capable cross-agency data platform could enable is precisely the terrain that critics argue requires explicit attention before, not after, deployment.

Snowflake's own history adds a data point that federal security officers are aware of. In 2024, a data breach affecting multiple major Snowflake customers — including Ticketmaster and Santander Bank — was traced to stolen authentication credentials, not to a breach of Snowflake's infrastructure itself. The incident highlighted that even a platform with robust security architecture can become the vector for data exposure if the operational security practices of the organisations using it are insufficient. Federal agencies deploying sensitive cross-agency data on Snowflake will need authentication and access controls that go beyond what the platform's baseline configuration provides.

What This Deal Signals About Where Government AI Is Heading

The OneGov Snowflake deal, read alongside the other twenty-plus agreements that GSA has signed under the initiative, describes a federal AI strategy that is making a specific and consequential architectural bet: that the fastest path to AI-enabled government is to standardise the data layer, pre-negotiate commercial terms at enterprise scale, and give agencies the tools they need without requiring each agency to independently navigate procurement and security certification. That bet is probably the correct one for the problem it is designed to solve. Federal AI deployment has been slow not primarily because agencies lacked interest but because the procurement and integration barriers were high enough that the technology changed faster than the contracts could be written.

The international comparison is instructive. The UK Government Data Science Partnership, Estonia's X-Road data exchange layer, and Singapore's WOG (Whole-of-Government) cloud architecture have all demonstrated that sovereign governments can build unified data infrastructure without surrendering the legal and governance distinctions between agencies that data sovereignty requires. None of those models relies on a single commercial vendor as the exclusive cross-agency data platform, and each has made the governance framework at least as central as the technical architecture. The OneGov approach, which prioritises speed and cost savings through commercial partnerships, achieves different outcomes on different timescales. It gets tools deployed faster. It produces less clarity about the long-term governance model.

The GSA's stated intention to convert OneGov agreements into longer-term direct contracts is the part of the story that will determine whether the Snowflake relationship becomes the durable infrastructure of federal AI or a useful interim arrangement that gets replaced as agencies develop more specific requirements. For now, 3.4 million federal employees have access to AI tools they did not have before, agencies can access a pre-negotiated path to a high-security data platform that has already achieved FedRAMP certification, and the data layer that previously made serious cross-agency AI nearly impossible is at least nominally available to everyone who needs it.

Conclusion

The GSA's OneGov agreement with Snowflake is the most significant single federal data infrastructure move since the Obama administration's cloud-first policy directive in 2011. It provides every federal agency with access to a pre-negotiated, pre-certified data cloud platform that can dissolve the silos that have made cross-agency AI analysis nearly impossible, at pricing terms that reflect the government's collective scale rather than each agency's individual negotiating position. The efficiency gains from eliminating the procurement friction that has historically separated agencies from commercial tools are real and documented: over $1.15 billion in identified savings from OneGov agreements in the first year.

What the announcement does not settle — and what any serious evaluation of the deal must keep in view — is the governance layer. Making data technically accessible across agency boundaries does not automatically resolve the legal, constitutional, and civil liberties questions about when that accessibility should be exercised and under what oversight. The data minimisation debate in Congress, the civil society scrutiny of federal AI deployments, and the history of government data infrastructure suggesting that technical capability tends to expand ahead of legal frameworks are all relevant context for a deal of this scope. The Snowflake platform gives the federal government the capability to do things with data that it could not do before. Whether it has yet established the governance framework to do those things wisely is a question that the OneGov announcement, by design, leaves for later.