Chinese large language model DeepSeek has shown that state of the art generative AI capability may be possible at a fraction of the cost previously thought.
AI is not a uniform workload — the infrastructure requirements for a particular model depend on a multitude of factors. Systems and silicon designers envision at least three approaches to developing and delivering AI.
SMRs promise to usher in an era of dispatchable low-carbon energy. At present, however, their future is a blurry expanse of possibilities rather than a clear path ahead, as key questions of costs, timelines and operations remain.
Rapidly increasing electricity demand requires new generation capacity to power new data centers. What are some of the new, innovative power generation technology and procurement options being developed to meet capacity growth and what are their pote...
Agentic AI offers enormous potential to the data center industry over the next decade. But are the benefits worth the inevitable risks?
The cost of low-carbon green hydrogen will be prohibitive for primary power for many years. Some operators may adopt high-carbon (polluting) gray hydrogen ahead of transitioning to green hydrogen
Dedicated AI infrastructure helps ensure data is controlled, compliant and secure, while models remain accurate and differentiated. However, this reassurance comes at a cost that may not be justified compared with cheaper options.
AI infrastructure increases rack power, requiring operators to upgrade IT cooling. While some (typically with rack power up to 50 kW) rely on close-coupled air cooling, others with more demanding AI workloads are adopting hybrid air and DLC.
Data center infrastructure management software is widely used but rarely utilized at full potential. Adopting the latest capabilities and optimizations could achieve better resiliency and efficiency.
Power and cooling requirements for generative AI training are upending data center design and accelerating liquid cooling adoption. Mainstream business IT will not follow until resiliency and operational concerns are addressed.
Dedicated GPU infrastructure can beat the public cloud on cost. Companies considering purchasing an AI cluster need to consider utilization as the key variable in their calculations.
The Uptime Intelligence research agenda includes a list of published and planned research reports for 2025, and is focused on Uptime Intelligence primary coverage areas: 1) power generation, distribution, energy storage; 2) data center management sof...
Cloud providers need to win AI use cases in their early stages of development. If they fail to attract customers, their AI applications may be locked-in to rival platforms and harder to move, which can have serious repercussions.
Nvidia’s dominant position in the AI hardware market may be steering data center design in the wrong direction. This dominance will be harder to sustain as enterprises begin to understand AI and opt for cheaper, simpler hardware.
The cost and complexity of deploying large-scale GPU clusters for generative AI training will drive many enterprises to the public cloud. Most enterprises will use pre-trained foundation models, to reduce computational overheads.
John O'Brien
Max Smolaks
Daniel Bizo
Andy Lawrence
Jay Dietrich
Peter Judge
Owen Rogers
Tomas Rahkonen
Jacqueline Davis
Douglas Donnellan