As U.S. critical sectors such as finance, healthcare, and defense accelerate their transition into cloud-native infrastructures, the need for robust and adaptive cybersecurity frameworks has become a matter of national urgency. Traditional perimeter-based models have proven insufficient against sophisticated cyberattacks, particularly supply chain compromises exemplified by SolarWinds, which exploit trusted channels to bypass conventional defenses. Addressing this challenge requires a paradigm shift toward zero-trust principles, where continuous authentication, contextual verification, and strict least-privilege policies are fundamental to ensuring system integrity. This paper proposes a zero-trust cloud security architecture enhanced by AI-orchestrated policy enforcement and microsegmentation to safeguard high-value assets in critical infrastructure. At a broad level, the framework combines identity-centric security with automated decision-making models that continuously monitor user behavior, system activity, and network flows. AI-driven orchestration enables real-time privilege adjustments, proactive anomaly detection, and automated policy enforcement across distributed cloud environments. By embedding intelligence into access and workload controls, the system not only prevents lateral movement by malicious actors but also adapts dynamically to evolving threat landscapes. Focusing on application within the U.S. context, this work demonstrates how the architecture supports resilience against insider threats, ransomware, and supply chain attacks targeting essential services. The framework establishes a scalable, national-security-grade model capable of supporting compliance with federal cybersecurity mandates while maintaining operational efficiency. Narrowing the scope, the study underscores that the integration of zero-trust and AI is not merely a technical upgrade but a strategic necessity for securing critical infrastructures that underpin national stability.