Federal OneGov AI deals: Grok, Gemini, and ChatGPT for under $1 per agency

What OneGov actually is

OneGov is GSA’s attempt to treat the federal government as a single buyer instead of hundreds of agencies negotiating duplicate contracts. The framework pre-negotiates enterprise terms — security baselines, unlimited-seat structures, reseller channels through Carahsoft and similar partners — so a program office can adopt a frontier model without running a six-month procurement from scratch.

The AI wave of OneGov deals accelerated through 2025 and into 2026 as the administration pushed agencies to comply with OMB memoranda on federal AI adoption. Josh Gruenbaum, head of GSA’s Federal Acquisition Service, framed the xAI agreement as fulfilling “President Trump’s promise that America will win the global AI race,” according to the September 2025 GSA press release. The result is a curated menu where five frontier labs compete side by side on a public price board — something no Fortune 500 procurement team has ever seen at this scale.

The price menu — and why the numbers are symbolic

GSA publishes list pricing on its Buy AI page. As of June 2026, the promotional tiers look like this:

• Perplexity Enterprise Pro for Government: $0.25 per agency through April 2027
• xAI Grok for Government Teams: $0.42 per agency through March 2027
• Google Gemini for Government: $0.47 per agency through September 2026
• OpenAI ChatGPT Enterprise: $1.00 per agency through August 2026 (executive branch)
• Anthropic Claude Enterprise: $1.00 per agency through August 2026

Carahsoft’s reseller documentation for Grok Enterprise lists a commercial list price of $360 per annual license with a 99% discount under the OneGov promotion, according to Carahsoft’s xAI OneGov page. The $0.42 figure is not a mistake. It is a loss-leader designed to get Grok onto every federal desktop before renewal negotiations begin at real enterprise rates.

The same logic applies across the board. OpenAI and Anthropic are not trying to earn margin from a $1 agency fee. They are buying reference customers, political goodwill, and training data on how government workflows actually use frontier models — at precisely the moment both companies are preparing IPO roadshows where government revenue diversification is a slide every investor expects to see.

Inside the xAI Grok deal — the longest contract on the board

xAI’s OneGov agreement is notable for duration as much as price. At 18 months through March 2027, it is the longest-running frontier-model OneGov deal GSA has signed, according to Nextgov/FCW reporting. Participating agencies receive access to Grok 4 and Grok 4 Fast with unlimited seats grouped into a single order per department or bureau.

The contract includes more than API keys. xAI committed a forward-deployed engineering team to help agencies integrate Grok into existing workflows, plus introductory training and a path to higher-security tiers. FedRAMP-authorized and Department of Defense Impact Level-aligned subscriptions are available at undisclosed additional pricing — the upgrade lane where real revenue eventually lives.

For xAI specifically, the deal matters because Grok is simultaneously landing in three distribution channels: federal workers, X’s social platform, and the SpaceX IPO narrative that treats xAI as an enterprise AI stack rather than a standalone model lab. A $0.42 agency fee buys mindshare that no Super Bowl ad could match. If a USDA analyst or VA claims processor defaults to Grok because it was free and pre-approved, that habit persists when the promotional period ends.

Security tiers, FedRAMP, and the gap between promo and production

The promotional OneGov tiers are not automatically cleared for classified workloads or sensitive personal data. GSA’s documentation describes baseline enterprise access; agencies handling controlled unclassified information or higher classifications must purchase upgraded tiers with FedRAMP authorization and DoD Impact Level alignment.

That two-tier structure mirrors what private enterprises are discovering: the cheap seat gets you a chatbot; the expensive seat gets you audit logs, data residency guarantees, role-based access control, and the compliance paperwork your CISO will actually sign. OpenAI’s Lockdown Mode rollout this week — restricting agentic browsing and file access for security- conscious deployments — is the commercial mirror of the same problem federal CISOs face when enabling frontier models agency-wide.

The voluntary 30-day frontier cyber review framework adds another layer. Agencies can request classified-benchmark testing of covered models, but participation is voluntary and the timeline is tight. OneGov gets models into buildings fast; cyber review decides whether those models stay connected to sensitive systems. Procurement velocity and security review velocity are deliberately decoupled.

Commoditization day — when every frontier lab sits on the same shelf

The strategic shift is not that government got a discount. It is that five frontier AI providers now compete on a single GSA price board with functionally identical procurement mechanics. Perplexity at $0.25 is cheaper than Grok; Gemini at $0.47 sits in the middle; OpenAI and Anthropic anchor the $1 tier. Agency buyers can pilot multiple models for less than the cost of a office coffee run.

This is the AWS Bedrock dynamic arriving in government form: the platform that hosts every major model becomes the power center, not any individual lab. GSA is not building Bedrock, but it is creating price transparency and side-by-side comparison that erodes the moats labs built on exclusive API relationships. When a program manager can swap from Claude to Grok by changing a Carahsoft order line, switching costs collapse.

For developers, the implication connects directly to agent tool-use architecture: applications that hard-code a single provider’s API will face procurement pressure to become model-agnostic. The federal government is effectively mandating multi-model thinking at the organizational level, even if individual apps still call one endpoint.

Politics meets pricing — equity stakes and penny contracts

The irony of June 2026 is thick. On Friday, Trump told reporters the government may take equity stakes in OpenAI, Anthropic, and xAI — making taxpayers literal shareholders. Simultaneously, those same companies sell unlimited enterprise access to the government for less than a dollar per agency. The two policies pull in opposite directions:

• Equity stakes imply the government should capture upside from AI wealth creation.
• OneGov penny contracts imply the government should subsidize AI adoption and treat models as infrastructure utilities.

Senator Bernie Sanders’ proposed 50% sovereign wealth fund tax targets the same three companies named in Trump’s equity-stake comments. Whether either policy advances, the political environment for AI IPOs just became more complex: roadshow slides about government partnership now sit next to headlines about government confiscation and government giveaway pricing in the same news cycle.

Congress may also weigh in through the Great American AI Act draft, which proposes federal preemption of state AI laws alongside frontier-model audit requirements. OneGov deals accelerate deployment; legislative frameworks try to govern what gets deployed. The gap between buy fast and regulate carefully is where the next year of enterprise AI policy will be fought.

What private enterprise buyers should take away

Fortune 500 procurement teams watching OneGov should note three lessons:

1. Promotional pricing is a customer-acquisition strategy, not a market price. When renewal comes, the $360-per-seat list price reasserts itself. Budget for post-promotion economics, not launch-window optics.
2. Unlimited-seat structures change usage patterns. When marginal cost per user is zero, adoption explodes — and so do inference bills on the back end if rate limits or overage tiers kick in on upgraded security plans.
3. Multi-model procurement is becoming the default. If the federal government — not known for agile IT — can run five frontier models in parallel, private enterprises clinging to single-vendor contracts look increasingly behind.

The weekend pause in equity markets gave traders time to digest Friday’s semiconductor selloff. It also gave policy watchers time to notice that Washington finished assembling a complete frontier-AI marketplace while Wall Street debated whether the AI trade still has legs. The models are not getting cheaper because they are less capable. They are getting cheaper because distribution is the battle, and the biggest single buyer in the world just set the reference price at pocket change.

Bottom line

GSA’s OneGov AI menu is the clearest signal yet that frontier models are commoditizing faster than their builders’ IPO valuations imply. Grok at $0.42, Gemini at $0.47, and ChatGPT at $1.00 per agency are not revenue strategies — they are land grabs for default-tool status inside the world’s largest enterprise customer. Security upgrades, FedRAMP tiers, and post-promotion renewals are where the real money waits.

For anyone building on AI APIs, the message is equally blunt: model choice is becoming as swappable as cloud regions. The value accrues to integrators, orchestration layers, and workflow-specific agents — not to locking in a single lab’s endpoint. Washington figured that out before most of Silicon Valley finished its Q2 board decks.

Sources: GSA — Buy AI / OneGov pricing; GSA — xAI OneGov partnership release; Nextgov/FCW — GSA Grok deal; Carahsoft — xAI Grok Enterprise OneGov. Related on Solana Garden: Trump AI government stake talks, frontier cyber review framework, AI agents and tool use, Great American AI Act analysis.