Project Horizon: Poolside AI Announces 2 GW AI Campus in West Texas
Poolside AI, in partnership with Nvidia, is launching Project Horizon, a 2-gigawatt AI campus in West Texas purpose-built for frontier-scale model training. The announcement, detailed in Poolside’s official blog post, underscores that building advanced AI requires massive infrastructure investments alongside model research. The project aims to deliver long-term control over power, cost, and scale through vertical integration as demand for compute continues to surge across the industry.
The initiative addresses a core challenge in the AI sector: while public attention focuses on model architectures and training breakthroughs, the physical infrastructure needed to power frontier models is equally critical. Poolside’s co-founders emphasized in the announcement that the company is taking a ground-up approach to data center development to avoid constraints faced by other labs competing for limited resources.
The Infrastructure Bottleneck in Frontier AI
According to Poolside’s blog, “When people ask what it takes to build frontier AI, the focus is usually on the model—the architecture, the training runs, the research breakthroughs. But that’s only half the story.”
The company states that Project Horizon is explicitly designed to power the next generation of frontier-scale training. The 2 GW facility will eventually match the entire power output of the Hoover Dam, according to Eiso Kant, co-founder of Poolside, as reported by multiple outlets covering the announcement.
This scale is significant in an industry where training runs for the largest models now routinely require hundreds of megawatts of continuous power. By developing the campus from the ground up in West Texas, Poolside aims to secure dedicated power infrastructure rather than relying on strained regional grids or existing data center providers.
Vertical Integration and Long-Term Control
Poolside describes Project Horizon as “a ground-up data-center development designed to give us long-term control over power, cost, and scale through vertical integration.” This strategy reflects a growing trend among leading AI labs seeking to reduce dependency on third-party cloud providers and hyperscalers.
The campus will be rolled out gradually “to avoid overloading any systems,” Kant noted. This phased approach suggests careful coordination with local utilities and grid operators in Texas, a state known for its deregulated energy market and abundant natural gas resources.
Reports indicate the project is expected to cost approximately $16 billion and will be gas-powered, aligning with Texas’s energy infrastructure strengths. The involvement of Nvidia, a key backer of Poolside, points to deep technical integration with the company’s GPU platforms, though specific hardware configurations were not detailed in the initial announcement.
Competitive Landscape and Industry Context
Poolside’s move comes as major players across the AI industry race to secure compute capacity. Companies like OpenAI, Anthropic, Google DeepMind, and xAI have all signaled plans for massive data center builds or partnerships to support next-generation training runs.
CoreWeave, a specialized AI cloud provider, has also been expanding aggressively, as referenced in coverage of the announcement. However, Poolside is betting that owning and designing its own power infrastructure will provide advantages in both cost predictability and operational control compared to leasing capacity.
The choice of West Texas leverages the region’s available land, existing energy production capabilities, and relatively streamlined permitting processes compared to more densely populated areas. Texas has become an increasingly attractive hub for large-scale data center projects due to its business-friendly environment and energy resources.
Technical Ambitions and Phased Deployment
While the announcement focuses primarily on power capacity rather than specific model details or benchmarks, the 2 GW target signals Poolside’s intention to train models at a scale that would be difficult to achieve through conventional cloud arrangements. Frontier training clusters today often consume 100-500 MW; a 2 GW campus could theoretically support multiple such clusters operating simultaneously or enable even larger single training runs.
The gradual rollout is a pragmatic acknowledgment of the engineering and grid integration challenges involved in bringing online such a massive facility. Poolside has not yet disclosed the exact timeline for reaching full 2 GW capacity or the initial phase sizes.
Nvidia’s involvement is particularly noteworthy given the company’s dominance in AI accelerators. The partnership likely includes not only hardware supply but also co-design of the data center infrastructure to optimize for Nvidia’s latest GPU and networking technologies, such as NVLink and InfiniBand or Ethernet-based fabrics optimized for large-scale AI workloads.
Impact on Developers, Users, and the AI Industry
For developers and researchers, Project Horizon represents a potential new source of frontier-scale compute capacity. If Poolside follows through on its vertical integration strategy, it could offer more predictable availability and potentially competitive pricing compared to spot markets or capacity-constrained cloud providers.
The broader industry may see this as validation of the “build your own” approach to AI infrastructure. Several leading labs have already begun exploring similar strategies, from Microsoft’s and OpenAI’s reported Stargate project to Meta’s investments in custom data centers.
However, such massive projects also raise questions about energy consumption, environmental impact, and grid strain. Poolside’s choice of natural gas power in Texas will likely draw scrutiny from environmental groups even as it provides a faster path to deployment than waiting for new renewable capacity.
For the Texas economy, the $16 billion investment promises significant job creation during construction and ongoing operations, along with increased tax revenue. The project could further solidify the state’s position as a major hub for AI infrastructure.
What’s Next
Poolside has not provided a detailed timeline beyond the commitment to gradual rollout. Industry observers will be watching for updates on initial power capacity targets, specific GPU cluster sizes, potential partnerships for excess capacity, and any sustainability measures the company may adopt.
The success of Project Horizon could influence how other AI companies approach infrastructure. If Poolside demonstrates meaningful advantages through vertical integration, more labs may pursue similar large-scale, purpose-built campuses rather than relying exclusively on cloud providers.
As frontier models continue growing in size and capability, the ability to secure reliable, large-scale power and compute infrastructure may become as important a competitive advantage as model architecture itself.
The announcement also highlights the increasing capital intensity of the AI race. A single 2 GW campus at $16 billion represents an investment that only the best-funded players can realistically pursue, potentially accelerating industry consolidation around those with access to significant capital and energy resources.
Technical Details and Future Implications
Though the initial blog post focuses more on strategic rationale than granular technical specifications, the 2 GW figure itself provides important context. Modern AI training clusters using Nvidia H100 or Blackwell GPUs can consume tens of megawatts per thousand GPUs when including networking, cooling, and power delivery overhead.
A facility capable of delivering 2 GW of compute power could theoretically support clusters orders of magnitude larger than today’s largest publicly disclosed training runs. This positions Poolside to potentially train models with trillions or tens of trillions of parameters at scales that would be challenging for competitors without similar dedicated infrastructure.
The West Texas location may also offer advantages for cooling, given the region’s climate, though water usage for evaporative cooling in large data centers remains a consideration in arid areas.
Poolside’s emphasis on “long-term control over power, cost, and scale” suggests the company may explore various energy strategies over time, potentially including renewable integration or advanced efficiency technologies as they become commercially viable.
Conclusion
Project Horizon represents a significant bet by Poolside AI on the importance of infrastructure ownership in the AI arms race. By partnering with Nvidia and committing to a massive purpose-built campus in Texas, the company is signaling confidence in its ability to execute at a scale few others have attempted.
As the AI industry grapples with the physical realities of powering ever-larger models, announcements like this one highlight that the next breakthroughs may depend as much on electrical engineering and energy strategy as on advances in machine learning algorithms.