.jpg)
According to a recent New York Times article, the rise of AI represents “the most fundamental change to computing since the early days of the World Wide Web.”
Written by Cade Metz, Karen Weise, Marco Hernandez, Mike Isaac, and Anjali Singhvi, the in-depth piece examines how artificial intelligence is driving a seismic shift in the design and infrastructure of data centers. The key to this monumental transformation is GPUs, which power the massive and growing computational needs of AI systems. This has led to the rise of data center-sized supercomputers that use vast amounts of electricity and water, triggering a fundamental revamp in how computers are built and operated.
Cirrascale is prominently featured in the article, specifically for our partnership with the Allen Institute for Artificial Intelligence (Ai2)—an AI research lab committed to building breakthrough AI to solve the world’s biggest problems—through open-source models, data, and tools. The data center where we manage Ai2’s compute resources clearly illustrates the infrastructure demands of modern AI. Originally a traditional 139,000-square-foot data center powered by 5 megawatts—enough for about 3,600 homes—the facility was gutted and rebuilt to support GPU-intensive workloads. Today, that same original 5megawatts can only power eight to 10 rows of GPU-packed computers, with each row typically consisting of multiple racks filled with high-density servers, highlighting how energy-hungry AI hardware has become.
The challenge isn’t only power consumption, however. Cooling is also a significant hurdle because AI chips generate intense heat. The article discusses how, instead of relying on traditional air cooling or water-intensive systems like Google’s evaporative cooling, Cirrascale uses massive chillers—industrial-scale air-conditioners that cool and recycle water. While this does increase power consumption, it ultimately eases pressure on local water supplies, underscoring the trade-offs between environmental sustainability and the relentless demands of AI computing.
As the industry quickly transforms, Cirrascale’s experience and leadership in the space have positioned the company as the new standard in data infrastructure: ultra-dense, ultra-powerful, and carefully engineered to meet the unique demands of AI.
How Ai2 Scaled Open-Source AI with a Custom Infrastructure Partner
As the NYT story makes clear, this transformation isn't just theoretical—it's already reshaping the world’s approach to AI development. The Ai2 example showcases how the evolving demands of AI are successfully met with purpose-built infrastructure.
Ai2’s mission was to build the most capable open-source AI models with visual capabilities, but a fragmented infrastructure hampered these aspirations. As a nonprofit research institute, its mission is to develop open AI that improves lives, meaning the organization needed more than just compute power. It needed a strategic partner.
That’s where Cirrascale comes in.
Having already deployed an initial NVIDIA GPU cluster for Ai2, Cirrascale understood the organization’s evolving needs. But the project wasn’t about selling cloud hours—it was about designing a scalable, high-performance solution tailored for Ai2’s long-term vision.
Cirrascale built a custom infrastructure using NVIDIA Hopper GPUs, optimized storage and a design focused on future growth. This flexible model allowed Ai2 to own key hardware components while relying on Cirrascale for setup, management, and ongoing support. Live and continuous communication bolstered collaboration between the teams, ensuring rapid response and proactive scaling.
"We have expertise across the entire stack, but we were lacking the right infrastructure to meet our needs," said Michael Schmitz, Ai2's director of engineering. “Cirrascale has been fantastic because of their willingness to customize every aspect of our setup.” This included reserved expansion space in the data center and the deployment of additional clusters, including the latest NVIDIA HGX B200, allowing Ai2 to grow without disruption.
Today, Ai2 is training models like Molmo, OLMo and Tülu 3 faster and at greater scale. The organization is now supported by infrastructure that evolves with its needs, meaning it’s not just keeping pace with AI innovation—it’s leading it.
Ultimately, Cirrascale didn’t just help build a data center. Together with Ai2, we developed an infrastructure foundation and a launchpad for open-source AI that’s changing the world. To read the complete Ai2 case study, visit https://www.cirrascale.com/ai2-case-study. To learn more about Ai2 and their open-first approach to empower researchers and developers, visit https://www.allenai.org.
