What's Happening?
The AI bandwidth boom is reshaping network connectivity, with data centers requiring unprecedented compute capacity and energy consumption. According to a webinar hosted by Ciena, GPU clusters for AI model training are expected to reach 300,000 GPUs this
year, potentially increasing to a million GPUs per cluster by 2026. This growth is leading to the development of gigawatt data centers by 2028. Communications service providers (CSPs) are poised to benefit from the AI WAN traffic boom, driven by hyperscalers, neocloud providers, and enterprises. CSPs anticipate significant traffic demand from AI in both metro and long-haul networks over the next three years, with AI expected to account for the majority of traffic demand.
Why It's Important?
The rapid advancements in AI are transforming data centers and redefining network connectivity, presenting significant opportunities for CSPs. Enterprises are expected to be the biggest traffic drivers on CSP networks, ahead of hyperscalers. CSPs are set to benefit from enterprise AI inferencing, which is associated with using pre-trained AI models. High bandwidth wavelength services are predicted to grow the most from AI over the next three years, offering CSPs a chance to expand their revenue streams. The evolution from AI training to AI inferencing impacts connectivity, with CSPs anticipating increased demand for high bandwidth services.
What's Next?
As AI data centers move to gigawatt consumption, new fiber routes and additional capacity must be established to connect data centers to the cloud. CSPs are expected to play a broader role in AI connectivity, particularly in serving enterprise customers with high bandwidth wavelength services. The rise of enterprise AI traffic is anticipated to drive significant growth in CSP networks, with CSPs adapting their networks to support the growing demands of hyperscalers, enterprises, and other participants in the AI ecosystem.
Beyond the Headlines
The historic energy demands of AI clusters are driving changes in data center locations, with access to reliable energy and land dictating where they are built. This shift may lead to more distributed designs that span multiple energy grids, requiring extended data center interconnect (DCI). The AI bandwidth boom is not only transforming data centers but also presenting ethical and environmental considerations related to energy consumption and infrastructure development.
