started · updated
AI infrastructure demand shifts focus to power, cooling, and memory
The rapid expansion of artificial intelligence is shifting the industry focus from GPU procurement to a broader competition for infrastructure, including memory, power, and cooling solutions. As AI moves from the training phase to the inference phase, the demand for high-performance servers, specialized memory like HBM, and advanced storage is increasing. Companies such as Samsung Electronics and SK Hynix are expanding their offerings to meet this demand, with SK Hynix committing approximately 54 trillion won to production facilities in South Korea.
Data center design is facing significant challenges due to the extreme power and heat generated by AI workloads. Traditional air-cooling methods are reaching their physical limits, leading to a projected rise in liquid cooling adoption, which is expected to reach 60% by next year. This shift is forcing operators to rethink floor plans and electrical distribution to accommodate racks that can exceed hundreds of kilowatts.
In the United States, the demand for computing power is physically altering the landscape, as large-scale data centers are increasingly built on agricultural and forested land. Meanwhile, technological advancements like NVIDIA’s DSX MaxLPS aim to optimize efficiency by allowing more GPUs to operate within existing power budgets through dynamic power sharing.
Entities
Akamai · Akamai Technologies · Anthropic · Nvidia · SK Hynix · Samsung Electronics
Claims
What the coverage asserts, and how many sources carry each claim.
- [○ 1 SOURCE] Akamai signed a $11.6 billion cloud service contract with Anthropic for a 7-year term. www.kgnews.co.kr
- [○ 1 SOURCE] Akamai expects to require approximately $5.5 billion in capital expenditures to meet AI computing demand. www.kgnews.co.kr
- [○ 1 SOURCE] Liquid cooling adoption for AI chips is projected to reach 53% in 2024 and 60% in 2025. www.kgnews.co.kr
- [○ 1 SOURCE] NVIDIA DSX MaxLPS can allow up to 40% more GPUs to be deployed within the same power budget. developer.nvidia.com
- [○ 1 SOURCE] Approximately 33 percent of large data centers were constructed on agricultural land. www.index.hr