The network that turns data centers into AI brains
In just two decades, the cloud has transformed from Amazon’s first storage service to a global platform that powers most consumer and business applications. As the cloud has evolved, so has the infrastructure behind it. Data centers are booming. Their count has jumped from ~250 a decade ago to over 10,000 today. They are getting larger, more complex and are placed closer to the users they serve.
All of this makes connectivity within and between these distributed data centers critical. And that’s where the network emerges as a vital element in the evolution of the cloud and data center. This evolution has strategic implications for all network operators – from hyperscalers and neocloud providers to communications service providers (CSPs), enterprises, and operators of mission-critical networks.
AI and sovereignty redefine the data center
Artificial Intelligence (AI) has radically changed what a data center is and how it is built and run. In parallel, rising global uncertainty has pushed nations to heighten security at their digital borders to safeguard privacy and assert data sovereignty.
Both of these change factors create extraordinary constraints on compute power, energy supply and cooling systems.
The graphics processing unit (GPU) has become the new currency for data centers. It is the ideal solution for running the complex calculations that power AI. An AI factory for intensive training purposes can house up to a million GPUs.
One GPU consumes about 1 kW of power. To put this into perspective, this is the average hourly consumption of a citizen in developing countries like Egypt, Vietnam or Peru. A full rack will soon reach 1 MW – pushing further power-density requirements. Meta just started the construction of its 29th data center designed to scale up to 1 GW of capacity.
Because servers turn most of that energy into heat, cooling systems must keep pace, necessitating the adoption of liquid cooling. Some companies are even placing data centers underwater to exploit ocean currents as natural coolant.
To overcome energy and heat dissipation limits, leaders like Jeff Bezos (Amazon) and Eric Schmidt (former Google) have floated the idea of orbital data centers that could harvest abundant solar energy and enjoy efficient cooling in space.
But as the industry debates innovations in compute, energy, cooling and location, let’s not forget another essential component of the data center: the network.
The hidden network powering data centers
Data centers are nothing without the network.
Clusters rely on the network to form the brain that creates and runs AI. Just as neurons and synapses give the human brain its power, the combination of compute and network makes artificial intelligence possible.
Networks are the connective tissue within and between the world’s data centers, supporting the delivery of data, digital services and applications:
- Inside a data center, the network forms the platform that moves the massive volumes of data between compute resources. It contributes to maximize efficiency and minimize job completion time.
- Between data centers, the network stitches multiple facilities together into a cohesive AI infrastructure, allowing workloads to be shared and collaboration to happen across sites.
- Between data centers and their users, the network ensures the data can reach devices quickly and reliably, delivering the performance modern applications demand.
Why the network must evolve too
As the cloud evolves to support AI and data sovereignty, network design must respond. This will take more than an incremental upgrade; it calls for a profound and foundational shift:
- AI models shuffle terabytes of training data between storage, compute nodes and external sources. Without sufficient bandwidth, speed and reliability, the most expensive GPU farms sit idle, turning multi-million dollar investments into costly waiting rooms.
- Computationally intensive tasks move between data centers to work around space and power limits, generating cascades of data requests and rapid traffic bursts. Therefore, architecting for increased scale and adaptability becomes critical.
- Model training iterates quickly, and real-time inference must deliver instant insights, requiring responsiveness and low-latency transmission.
- AI and data-sovereignty requirements for safety, defense, finance and healthcare services, demand extreme availability and robust security.
And yet, it’s possible to satisfy these demands with architecture and technologies that deliver ultra-high scale, performance and flexibility.
Done well, we can level-up data security, cut the environmental impact of AI processing, and unlock the economic, competitive and social benefits that AI promises.
What you should expect from your data center network?
Data center networking solutions should feature a unique set of capabilities to thrive in today’s AI-driven environment.
- High performance: The network should provide ultra-high speed, low-latency, loss-less connectivity, leveraging Ethernet to slash cost and power per bit. 800G Ethernet is the new state of the art, with 1.6T Ethernet coming next year.
- Quality-first design: Network hardware and software must be built to best-in-class standards and undergo rigorous testing to guarantee reliability and continuous business operation.
- Operational efficiency: Precise, reliable and predictable automation should let the network adapt dynamically to changing demands and to unexpected events.
- Deep security: A defense-in-depth approach featuring quantum security and built-in DDoS detection and mitigation must protect the network today and into the future as threats evolve.
Take the next step
Networks built this way are no longer optional; they’re the only way to keep sovereign AI workloads from becoming bottlenecks. They become strategic differentiators that turn data center infrastructure into a competitive edge.
If you want to learn more about the critical role that the network plays in the evolution of the cloud and data centers, start reading our white paper Network the cloud, or visit our web page Critical connectivity for modern data centers.
If you are planning to build a mission-critical data center infrastructure, Nokia can help. Join the many cloud specialists who rely on our proven track record. Our quality-first philosophy is a key reason customers trust us to deliver data center networks that simply work and are easy to operate.
