The current era of artificial intelligence has moved past the phase where a clever architectural tweak or a more efficient attention mechanism can define the state of the art. We have entered the age of the compute wall, where the primary constraint on intelligence is no longer just the quality of the data or the elegance of the code, but the raw, physical availability of electricity. For the developers and researchers pushing toward the next generation of frontier models, the conversation has shifted from floating-point operations to gigawatts. The battle for AGI is now being fought not just in the cloud, but in the soil and the power grids of the American South.

The Scale of Project Camellia

OpenAI is addressing this physical bottleneck with the launch of Project Camellia, a massive infrastructure undertaking located at the Savannah Gateway Industrial Hub in Effingham County, Georgia. The scale of the project is defined by a staggering power requirement: 3.2GW of electricity. To put this in perspective, a gigawatt is one billion watts, and securing 3.2GW for a single data center project represents one of the most ambitious energy acquisitions in the history of the AI industry. This power capacity is the direct ceiling for the computing power OpenAI can deploy, which in turn dictates the intelligence level, training speed, and operational capacity of the models that will emerge from this site.

To realize this vision, OpenAI has entered into a strategic agreement with the Georgia Power Company. The delivery of this power will not happen overnight but will follow a phased rollout beginning in 2028 and extending through 2032. This staggered approach allows the company to synchronize power availability with the physical installation of server racks and the scaling of the cluster, ensuring that the local grid remains stable while the facility grows. Notably, the entire project is funded through private capital. OpenAI is bearing the full cost of infrastructure construction and electricity services without relying on government subsidies or local tax incentives. This financial independence is designed to accelerate the pace of development and maintain operational autonomy. Detailed progress and official updates on the project are hosted at www.projectcamellia.com.

Engineering Around the Resource Conflict

Historically, the arrival of a hyperscale data center in a rural or industrial area has been viewed with suspicion, often seen as a parasitic relationship where the facility drains local water tables and spikes electricity costs for residents. Project Camellia attempts to flip this narrative through specific engineering choices that decouple industrial growth from community depletion. The most significant shift is the implementation of a closed-loop water cooling system. Unlike traditional data centers that continuously draw fresh water from the environment and discharge it as waste, a closed-loop system functions like a car radiator, circulating the same coolant repeatedly to dissipate heat. This design ensures that water consumption is limited to basic facility maintenance and office use, effectively removing the threat of water scarcity for the surrounding Effingham County community.

Beyond water, OpenAI is introducing an active power control system to manage the grid's volatility. During peak demand periods—such as extreme heatwaves when residential air conditioning usage spikes—the data center is designed to preemptively reduce its own power consumption. By dynamically shedding load, the facility prevents the grid from becoming overwhelmed, ensuring that residential customers do not face instability or brownouts. Because OpenAI is funding the entire grid expansion and service cost privately, these upgrades do not translate into higher utility bills for local citizens. The tension between the energy-hungry nature of frontier models and the needs of a local community is thus resolved not through policy promises, but through technical constraints and financial absorption.

This shift reveals a deeper truth about the current trajectory of AI development. The competitive advantage of a lab is no longer just its talent pool, but its ability to manage civil engineering and energy logistics. When the minimum requirement for a training cluster reaches the gigawatt scale, the AI researcher must become a power grid expert. The infrastructure at Effingham County is a blueprint for this new reality, where the physical footprint of the AI is as carefully engineered as the neural network it supports.

Investing in the Human Layer

Recognizing that physical infrastructure alone does not create a sustainable ecosystem, OpenAI is pairing Project Camellia with a significant investment in local human capital. The company has committed a total of $80 million to a community benefit fund throughout the project's lifecycle. These funds are earmarked for essential public services, including local schools, public safety, and healthcare. To prevent the common pitfalls of corporate philanthropy, the fund will be managed by an independent entity and subjected to annual public audits, ensuring that the economic windfall of the data center reaches the people of Georgia directly.

Parallel to the financial grants, OpenAI is launching a massive educational initiative targeting students at universities, community colleges, and technical schools across Georgia. The company is providing up to $71 million in Codex credits, with each eligible student receiving $100 in credits via their ChatGPT accounts. Codex is an agentic coding tool—a system that does not just suggest lines of code but can plan and execute complex software projects based on high-level goals. By providing this access, OpenAI is attempting to bridge the gap between theoretical AI knowledge and practical application in fields ranging from manufacturing and medicine to entrepreneurship and skilled trades. Students can access these credits through the dedicated portal at chatgpt.com/codex/georgia-college-students/.

This educational push serves a dual purpose. While it provides immediate value to the students, it also cultivates a local workforce capable of operating and maintaining the very infrastructure Project Camellia is building. The project is expected to employ thousands of construction and permanent staff, with a stated preference for local contractors and businesses. By becoming one of the largest taxpayers in the county, the facility transforms from a mere server farm into an economic anchor for the region.

The New Standard for Frontier Infrastructure

Project Camellia is not an isolated experiment but part of a broader trend toward massive, centralized compute hubs. The existing Abilene campus in Texas has already demonstrated the viability of this model, supporting the training of frontier models and creating thousands of jobs. However, the move toward 3.2GW projects signals that the baseline requirements for the next leap in AI capability have shifted upward. We are seeing the emergence of a new industrial standard where the ability to secure and control gigawatt-level power is the primary gatekeeper to the next generation of intelligence.

To ensure transparency and community alignment, OpenAI held a public meeting on July 23, allowing residents to engage directly with the project team. This process of creating a community agreement is now as critical as the technical specifications of the GPUs being installed. When a project reaches this scale, the social license to operate is just as important as the electrical permit. The integration of closed-loop cooling, active load shedding, and direct educational investment represents a strategic attempt to minimize the friction of hyper-growth.

As the industry moves forward, the distinction between software companies and infrastructure companies is blurring. The success of the next frontier model will not be decided solely by the loss function or the dataset, but by the stability of the power grid in places like Effingham County. Project Camellia proves that the path to AGI is paved with high-voltage lines and sustainable cooling systems.