Santa Clara, California – February 14, 2022 – Artificial Intelligence data center provider Colovore continued its strong growth in 2021 and announced a 2 MW expansion of its Santa Clara facility to meet the growing and unique demands of AI and HPC infrastructure deployments.
Colovore’s expansion brings much-needed high power-density colocation capacity to the Silicon Valley market, which is facing a significant shortage of data center supply amidst the explosion in Artificial intelligence, Machine Learning, and Big Data applications and services.
AI and GPU Servers Are Rendering Legacy Data Centers Obsolete
Because of how much compute is now packed into a standard GPU server, AI and ML platforms require much more power and cooling per server rack unit than was ever anticipated when the vast majority of data centers were built. Limitations on the amount of available power and cooling per cabinet due to insufficient air-based cooling systems have resulted in significant waste inside the legacy data center, with customers now only being able to fill up 10 – 25% of an entire rack. Expensive, unusable empty space, and I.T. inefficiency are now the norm in legacy colocation data centers. In addition, these legacy environments present significant barriers to AI performance and data efficiency, due to latency resulting from the need to spread out the server footprint in order to manage heat loads.
The Rise Of Purpose-Built AI Data Centers
Just as the explosion in AI applications and services relies upon purpose-built AI chipsets and server configurations, purpose-built data center environments are necessary to support AI applications in a reliable, scalable, and cost-efficient way. The incredible compute densities in these AI platforms require similar high power and cooling densities in the data center. In terms of specifications, an AI Data Center offers:
- 35 – 50 KW+ of usable power per cabinet (vs. 3-5 kW per cabinet in legacy colos)
- Robust liquid cooling via rear door heat exchangers or direct liquid cooling (vs. inefficient air-based cooling)
- Data center floor loading of 3,000 pounds+ per cabinet vs. 1,500 pounds (as AI and GPU servers are very heavy)
Colovore was built from the ground-up with liquid cooling and market-leading power and cooling densities per cabinet in Silicon Valley to offer a highly-scalable, AI-optimized data center environment for the most data-intensive, modern server platforms. The end result is that customers can fill up an entire AI rack with servers from top-to-bottom, which is necessary to optimize AI performance. This small, compact, compute-dense footprint allows customers to achieve the highest application performance and lowest latency, lowest TCO, and highest I.T. operating efficiency in Northern California.
“The growth in data and compute intensity has been astonishing in just the past five years,” said Sean Holzknecht, President and Co-Founder of Colovore. “We now support a large amount of AI and GPU systems for Fortune 500 companies down to Silicon Valley startups. The AI revolution is here, it’s happening across all industry sectors and company sizes, and these servers require highly-customized data center environments. We’re proud to be on the forefront supporting this AI wave.”
“You don’t race a Ferrari on a go-kart track,” said . “The amazing innovations at the chip and silicon layers have resulted in these incredibly dense servers with very different technical requirements from prior generations—and yet most data centers haven’t changed their specs in years. Just finding the right home in which to deploy these AI and GPU servers has become a huge bottleneck for many companies, and we’re excited to continue to meet this growing market demand with our highly-differentiated, liquid-cooled data center in Santa Clara.”
Colovore’s AI data center in the heart of Silicon Valley features state-of-the-art liquid cooling, power densities of 50 kW per rack, and the lowest TCO and highest scalability in San Francisco Bay Area colocation. Colovore supports Fortune 500s to startups, from single rack to multiple-MW deployment sizes.