US Expands AI Strategy with Focus on Infrastructure and Controls
The United States is broadening its AI strategy to focus on securing infrastructure, accelerating regulation, and managing global competition. Key priorities include ensuring energy abundance for data centers and using export controls on AI chips and software as a foreign policy tool. These controls now extend to cloud-based inference platforms, requiring providers to navigate complex end-user and country-of-origin rules.
- The focus on securing energy for data centers is driven by projections that they could consume between 6.7% and 12.0% of total U.S. electricity by 2028, a significant increase from 4.4% in 2023. This surge in demand is causing significant delays in connecting new data centers to the power grid, with utilities in some markets quoting wait times of up to ten years. - Recent policy shifts include Executive Order 14318, which aims to accelerate federal permitting for large-scale AI data centers by streamlining environmental reviews. This is part of a broader "America's AI Action Plan" that also focuses on expanding grid capacity and securing AI-relevant infrastructure from foreign threats. - The renewed focus on export controls has shifted from broad prohibitions to a more targeted, end-user-focused strategy. However, this has created policy uncertainty, with recent decisions to grant licenses for sales of advanced chips to China, such as Nvidia's H200, sparking bipartisan criticism and concern among U.S.-based AI companies who benefit from weaker Chinese competitors. - In enterprise adoption, the industry is moving from monolithic models to multi-agent systems and agentic workflows that can autonomously plan, execute, and reason through complex tasks. This shift is forcing a change in API design, moving away from simple request-response to supporting persistent, stateful interactions required for autonomous agents. - Governance frameworks are now evolving to address the complexities of these multi-agent systems, requiring mechanisms to monitor not just individual agent performance but also the emergent, system-level behaviors and interactions between agents. Key frameworks gaining traction include the NIST AI Risk Management Framework and ISO/IEC 42001, which provide structured guidelines for managing risks like algorithmic bias and ensuring model explainability. - Venture capital funding for AI startups surged to $202.3 billion in 2025, capturing nearly 50% of all global venture funding. A significant portion of this, $80 billion, was directed towards foundational model companies alone, more than doubling the investment from the previous year. - Despite massive investment, a recent survey shows that while 77% of executives believe rapid AI adoption is necessary to remain competitive, only 25% are confident their current IT infrastructure can support scaling AI across their enterprise. Common barriers to successful adoption include poor data quality, lack of technical expertise, and difficulties integrating with legacy systems. - The geopolitical landscape of AI is heavily influenced by the semiconductor supply chain, which relies on U.S. chip design, Dutch lithography machines from companies like ASML, and Taiwanese manufacturing. U.S. export controls aim to leverage these chokepoints to slow China's AI progress, a strategy that has constrained but not halted advancements, as Chinese firms like SMIC have developed workarounds for 7nm chip production using older technology.
