Why building two data centers a week won’t fix AI’s bottleneck (www.techradar.com)

The AI race is intensifying as major hyperscalers like Meta, Apple, Microsoft, Amazon, and Alphabet are projected to invest around $600 billion in AI infrastructure by year-end, with two new data centers launching weekly to meet soaring demand. However, a critical bottleneck is emerging not in computing power, but in connectivity. While thousands of data centers are being built, the data transfer networks that interlink them are lagging, risking the potential of deployed AI technology. Despite substantial investments in AI chips, the systems depend heavily on reliable, high-capacity data pipelines to continuously feed workloads. To address these connectivity challenges, a shift towards wireless optical communication is gaining traction. This technology utilizes directed beams of light to establish high-capacity links over considerable distances, bypassing the limitations of traditional fiber networks that are slow to deploy. As AI demands evolve rapidly, innovations like optical communication can facilitate more agile network expansions, ensuring that data moves swiftly and efficiently between computing resources. The future of AI will hinge not just on powerful hardware and extensive data centers, but on how well these resources can be interconnected to enhance performance and meet the increasing operational demands of AI systems.