I came across a new post by a Nokia employee. Link to the post behind the image.
Summary.
Overview: Solving the Data Center Crisis
The writing discusses the “data center crisis” caused by artificial intelligence (AI), where current infrastructure is hitting the limits of power consumption, heat generation, and data transfer speeds. Butts uses mathematical models (Bass Diffusion Modeling) in his analysis to predict when new technologies will move from research to practice.
The analysis is divided into two main groups (clusters):
Cluster 1: AI Fabrics Operating at the Speed of Light
The research focuses on transforming network structures from static electrical networks into dynamic and optical ones.
- Photonic Interconnects: Moving from copper to light (photonic rails) is essential for eliminating latency and increasing bandwidth.
- SmartNICs: Offloading tasks from the server’s main processors (CPUs) to smart network interface cards frees up power for the AI computation itself.
- Dynamic Topologies: Networks that can change their shape in real-time according to traffic.
- Timeline: This technology is maturing (81% of the research phase has been passed). Hyperscalers (such as Google, Meta, Microsoft) are expected to deploy these widely over the next 12–18 months.
Cluster 2: Coordination of Energy and Computing
This area is only at the beginning of its lifecycle (only 9% of potential reached), but it is critical for satisfying the massive hunger of AI.
- Carbon-Aware Computing: Software and architectures that adjust AI workloads in real-time based on the state of the power grid and carbon intensity.
- TCO Optimization (Total Cost of Ownership): Decision-making is driven by cost-effectiveness over the entire lifecycle, including the use of FP8 precision, which reduces inference costs.
- Social Impacts: The research considers the strain data centers place on local power grids and water resources (the “Cloud Next Door” phenomenon).
Key Takeaways for Investors and Technology Leaders:
- The bottleneck has shifted: The problem is no longer just chip speed, but how chips talk to each other (interconnect).
- Optics are a necessity: The research strongly signals that optical solutions (which Nokia and Infinera are developing) are moving from the research phase to the mainstream very quickly.
- Energy efficiency is the new currency: The growth of data centers is no longer dependent on money, but on whether they can get enough electricity and manage its environmental impact.


