CAISO Proposes New Grid Connection Rules for Large AI Loads
California grid operator sets 50 MW threshold and cost protections
The California Independent System Operator (CAISO) has approved a comprehensive proposal establishing new rules to accelerate and regulate the connection of large electrical loads, such as AI data centers, to the state's transmission grid. The measure aims to expedite energization for high-demand digital infrastructure while safeguarding grid reliability and protecting residential consumers from unfair cost shifting.
Key details
The CAISO board approved the package to govern end users with a peak electrical demand of 50 megawatts (MW) or more. CAISO plans to file the formal proposal with the Federal Energy Regulatory Commission (FERC) ahead of a November 16 deadline, responding to a federal order directing regional grid operators to reform large-load interconnection tariffs.
The package introduces five primary mechanisms designed to balance growing computing demand with electric system stability:
- 50 MW classification: Formally classifies end users requiring 50 MW or more as large loads, establishing clear operational oversight.
- Remote disconnection capability: Authorizes CAISO to remotely disconnect large loads during grid emergencies to prevent widespread outages and maintain overall system reliability.
- Ratepayer cost protections: Mandates that participating transmission owners establish strict tariff terms to prevent infrastructure costs from being shifted onto utility consumers.
- Colocated generation rules: Exempts purely on-site generation that serves only an adjacent facility from comprehensive ISO studies, while requiring studies for any generation that feeds power into the broader grid. Colocated loads must also be metered separately from offsetting power sources to ensure they pay required ancillary service charges.
- Flexible and phased energization: Permits large loads to connect early and receive electrical service capped at dynamic thresholds set by grid conditions before permanent transmission upgrades are finished. CAISO also plans to introduce permanent flexible interconnection frameworks and proximate generation mechanisms by 2027.
Why this matters
Data centers connected to CAISO's network already account for approximately 8 gigawatts (GW) of demand across high-tech hubs like Silicon Valley. As artificial intelligence models scale in complexity, individual computing facilities routinely request hundreds of megawatts of continuous power. Without enforceable standards, sudden large-load connections risk overwhelming transmission capacity and driving up electric rates for regular households.
Context
CAISO operates across nearly 26,000 circuit miles (41,842 km) of transmission lines, serving more than 30 million people across 80 percent of California and parts of Nevada. The operator's initiative follows growing federal scrutiny over how regional transmission organizations manage surging data center pipelines. By establishing structured interconnection options and remote curtailment safeguards, California is moving to become the first major US grid operator to gain board approval for post-order large-load tariff reforms.
Risks and open questions
While the proposal offers developers an expedited path through early, condition-dependent grid access, it leaves key operational questions unresolved. The remote disconnection authority means operators of energy-intensive AI training facilities must accept the financial and computational risks of curtailment during grid stress unless they invest heavily in on-site backup generation or battery storage. Furthermore, the specifics of long-term flexible interconnection frameworks will remain in development through 2027.
What happens next
CAISO will submit its proposal to FERC by the November 16 deadline for federal review and approval. Transmission owners and regulatory stakeholders across California will work with the grid operator to define specific tariff language, while developers prepare for independent metering and curtailment readiness across planned 50 MW-plus data center campuses.
Source: Data Center Dynamics Published on AI Usage Global, author: AUG Bot
