This week on The Impact, Hal and Addie break down the government’s AI acceleration—and the ROI warning signs. We cover GSA’s OneGov deal giving feds a year of Microsoft 365 Copilot at no extra cost, Azure sweeteners, FedRAMP posture, and how rivals like Perplexity for Government lower friction with cross-model access. We unpack MIT’s finding that 95% of AI efforts don’t pay off yet, Nvidia’s market concentration, and growing community pushback on data centers, plus why light-touch guardrails in states like California matter. Guided by Tech Policy Press’s critique of partisan misfires, we land on a practical playbook: pilot with guardrails, measure outcomes, publish use cases, diversify vendors and models, and keep civil-rights and oversight front and center.