Under Secretary Emil Michael: Pentagon gives 3 million employees Gemini access for 47 cents, bans DeepSeek

Summary

The Pentagon has deployed Google Gemini to all 3 million Department of Defense employees and military personnel at an introductory rate of 47 cents per user, marking the first time AI has been made available at scale across military networks. Emil Michael, Under Secretary of Defense for Research and Engineering, confirmed the rollout was completed in roughly three months, with Google engineering teams standing up dedicated war rooms to manage the launch day deployment across a classified, air-gapped network architecture specifically configured to prevent query data from feeding back into Google's general training models.

Data ownership is explicitly retained by the government. Michael's position is clear: the taxpayer owns the government, the government owns the data, and Google has no claim to it. The technical architecture was designed around that constraint, which he says was the primary complexity in delivering the deployment.

DeepSeek and other foreign AI models are being banned across DoD personnel and contractors. Michael cited legislation moving through Congress that would codify the prohibition into law, framing it as preventing adversaries from gaining intelligence on how the US military uses AI systems.

The Gemini deal is framed as a commercial entry point, not a permanent pricing structure. Michael expects to pay market rates over time, with the 47-cent arrangement functioning as a customer acquisition and use-case discovery vehicle for Google while giving DoD immediate broad access without token-level budget constraints.

The longer-term AI architecture targets all four designated national champions, specifically Google, OpenAI, Anthropic, and xAI, deployed simultaneously across classification levels. The rationale mirrors how commercial enterprises already operate, using Claude for coding tasks, Gemini for other workloads, and allowing model competition to benefit the end user. Michael describes the goal as consumerizing access while expanding capability with each security tier.

On the infrastructure side, an executive order has directed that data centers be built on military land for the benefit of these companies, though business models and allocation frameworks have not yet been determined. The administration is explicitly positioning this as a departure from what Michael characterizes as the prior administration's approach of constraining AI development to favor a single company.

Use cases span a wide spectrum. At the low end, AI handles document drafting and spreadsheet generation for a 3-million-person workforce. At the high end, Michael points to decades of accumulated satellite imagery and sensor data that could be used to train models capable of anomaly detection at a scale no human analyst team could match. War-gaming simulations and logistics planning round out the operational use cases being pursued.

Talent acquisition remains a structural challenge. Michael runs dedicated recruiting sessions every Tuesday, pitching the role as the largest technology deployment in the world with genuinely novel use cases unavailable anywhere in the private sector, and positioning DoD experience as a career accelerant given how aggressively commercial AI companies are now building federal practices.