Trump presses OpenAI to throttle GPT-5.6—while Big Tech fights over AI-driven power prices

The Trump administration has asked OpenAI to roll out its next powerful AI model in a staged manner, with an initial restriction of GPT-5.6 to a small set of government-approved partners before any broader public release. Reporting indicates the administration’s rationale is security, with officials seeking tighter control over who can access frontier capabilities first. The request is framed as a pre-release governance step rather than a full ban, but it signals a more interventionist posture toward model deployment timelines. In parallel, the broader AI ecosystem is moving toward more operational integration, as Anthropic tests mobile support for Claude Cowork to manage long-running tasks from phones. Geopolitically, this is a governance-and-leverage story: frontier AI access is becoming a strategic resource that governments want to meter, audit, and potentially align with national security priorities. The administration’s approach suggests a shift from “innovation-first” release norms toward “controlled access” regimes that can be used to reduce risk while also shaping the competitive landscape among vendors and government contractors. That dynamic is likely to intensify competition between U.S. tech leaders and their ability to secure favorable procurement or regulatory treatment. Meanwhile, Oracle’s lawsuit against Wisconsin over efforts to protect utility ratepayers from data-center-driven price hikes highlights how AI policy collides with domestic economic distribution, creating political friction that can feed back into federal tech and energy agendas. Market implications span both AI software and the power/utility complex. If GPT-5.6 access is limited initially, demand may concentrate among approved partners, potentially shifting near-term revenue visibility toward government-adjacent deployments and enterprise contracts rather than mass consumer adoption. On the infrastructure side, the Wisconsin dispute underscores that AI-driven load growth can translate into higher electricity rates, raising the risk of cost pass-through debates for data centers and cloud operators. While the articles do not provide specific price figures, the direction of pressure is clear: utilities and ratepayers face upward cost exposure, while hyperscalers and AI compute providers face political scrutiny and potential regulatory constraints. In the background, product moves like Figma’s “full-stack canvas” for AI-era creation point to continued investment in AI-native workflows, which can support software valuations but also increase compute intensity across the digital economy. What to watch next is whether the administration formalizes these access limits into enforceable procurement or licensing conditions, and whether OpenAI publicly confirms a phased timeline for GPT-5.6. Key triggers include any expansion of the “government-approved partners” list, changes to security review procedures, or indications that other frontier labs face similar constraints. On the energy front, monitor court filings and state regulatory actions in Wisconsin, because outcomes could set precedents for how AI data-center costs are allocated. For the broader market, watch for signals that mobile and desktop-like agent tooling (such as Claude Cowork) accelerates usage patterns that increase compute demand. Escalation would look like broader restrictions or compliance requirements across the sector, while de-escalation would be reflected in faster approvals for wider releases and clearer, predictable cost-sharing frameworks for power.

Geopolitical Implications

• 01

  State leverage over frontier AI deployment timelines is increasing via controlled access regimes.

• 02

  Security-gated releases can reshape competitive dynamics among AI vendors and government contractors.

• 03

  Domestic energy-cost disputes may constrain AI scaling and intensify regulatory bargaining.

• 04

  Agentic, mobile tooling raises the likelihood of stronger monitoring and audit demands.

Key Signals

• —Formal policy or contract clauses tying GPT-5.6 access to security reviews and approved partners.
• —Changes in the size and criteria of the government-approved partner list.
• —Wisconsin court/regulatory outcomes affecting data-center cost allocation.
• —Adoption metrics for Claude Cowork mobile that indicate rising compute demand.