Over the past few years, the IT industry has been laser-focused on AI, with those deploying new AI solutions looking at GPU supply chains, power shortages, and increased timelines for planning and deployments. As AI deployments grow, use cases to support new solutions emerge as companies move beyond testing what AI can do to production-level deployments, inference, and measuring real results.
As we move forward on our AI journey, we’ll see breakthroughs in more accurate and efficient healthcare, personal robotics that enhance our daily lives, smarter manufacturing with better quality and efficiency, and innovative climate solutions that will change how we tackle global challenges.
Infrastructure and market challenges
To reach the pinnacle, we will need a combination of capabilities, including energy, land, software, hardware, and network. These elements must be integrated with the ability to execute and operate on a large scale within a complex environment, with accurate data to support the 10x energy (power and cooling) requirements of next-gen AI workloads.
“Electricity consumption from data centres, artificial intelligence (AI) and the cryptocurrency sector could double by 2026.”
IEA
Over the coming decade, we can expect server power needs to grow dramatically. We are nearing the end where we think about servers, but rather rack scale architectures and designing for SuperPOD-scale deployments.
However, major market constraints have caused the planning cycle to grow tremendously, with power being the new denominator for data center deployments and our ability to get workloads up and running.
That wasn’t always the case. Data centers evolved from network access points or NAPs that grew out of the early days of telecom and the ARPANET, which became the Internet we know today. Finding a data center that worked for scaled applications required dense network connectivity or easy access to fiber to nearby carrier hotels. Network assets were constraints, scarce in some ways, and drove placement and expansion decisions.
The driving energy
Today, energy is the primary constraint. At Flexential, we stay ahead of our planning and expansions by engaging deeply with our energy partners, local governments, developers, and customers. Our focus on planning has shifted from 2-3 years to 5 or more years in major markets to understand supply and demand. In many cases today, facilities are being sold faster than they can be built. Enterprises, just like data center builders and operators like Flexential, need to plan much further out than ever before.
Planning the needs of future data centers and the power to support their growth is relatively new. Local grids are reaching capacity, and the supply chain, permitting process, and access to resources slow the growth of putting more power on the grid in meaningful ways. Not to mention, the growth rate is driving fast consumption. Many utilities and data center providers are looking at new technology for on-site generation or near-site deployments of small-scale nuclear, gas-powered generation, and hydrogen.
The emerging network
AI is data-hungry, and to make AI useful, it needs context. Data provides this context, sometimes right or wrong, but that’s where GenAI and machine learning can help. We know it’s not perfect, but models are getting more accurate every day, and new platforms will emerge using multiple models to help fact-check and improve accuracy. However, data needs the network.
Meanwhile, another major transition has occurred, one we use every day and don’t even notice. It’s mobile. 5G has now enabled over 100 trillion megabytes transmitted last year, which is double from 2021 (CTIA, 2024). In 2013, it was 3 trillion MB. Mobile data consumption and generation is off the charts, and a lot of it is via the cellular network. That will continue to grow as 6G emerges and 5G expansion continues, including device transition.
Energy gives way to the network
Our shift to mobile cellular and fixed wireless, as well as the increased need for network bandwidth for AI’s data-hungry demand, are happening simultaneously. Therefore, the network capacity will return to being the denominator once we solve energy needs. So, now that we’re back to the network, we must ask. If mobile is eating all the bandwidth, what’s left for AI?
To be successful with AI, fiber, throughput, latency, diversity of routes, and other core network constraints and requirements will be at the forefront. Just like with energy partners, network and fiber are just as critical as we continue to shift to the new AI-capable world.
AI eats data for breakfast.
The network enables data.
AI is enabled by power and the network.
The network will be key to keeping models and training up to date in AI inference. It will be critical. It will be the next area of constraints that we need to partner for scale, capacity, and expansion. Fiber is being refreshed in many markets, and fiber strand counts are increasing. Older systems have 96 fiber strands. Newer systems will have 144 or 288 strands or more. These capacity upgrades are coming at the right time!
At Flexential, with data center expansions in key markets such as Portland and Atlanta, we are building state-of-the-art data centers to support demand that may not have scaled network capacity.
Explore Flexential Data Centers
