The Resource Costs of AI

The Networking Field Day 39 roundtable, led by Tom Hollingsworth, dove straight into the massive resource drain caused by the current AI boom. The delegates discussed how the industry is already behind the eight ball on power, with AI’s exponential demand making the existing, outdated power grid’s problems significantly worse. This isn’t just a data center issue. It’s causing widespread component shortages for essentials like RAM and GPUs, affecting everyone from enterprise users to home gamers. The conversation highlighted that consumers are ultimately paying for this AI race, not just through new tech, but through soaring power and water bills as AI companies with deep pockets outbid ordinary consumers for finite resources.

In response to this power crisis, the discussion shifted to solutions. The delegates noted a serious new investigation into small modular nuclear reactors and even restarting old plants like Three Mile Island, things unthinkable just a few years ago, alongside more hopeful developments in solar. On the networking side, this resource demand is forcing the creation of entirely new, expensive technologies like Ultra Ethernet and massive 800-gig switches just to keep these AI data centers fed. These come with huge R&D costs, which will inevitably be passed down, raising the question of whether networking costs are about to go through the roof for reasons outside the network engineer’s control.

Finally, the panel debated the future of AI itself, noting the buzzword is becoming meaningless as the industry pivots from massive LLMs to more efficient, domain-specific smaller models (SLMs) and agentic AI. The group observed that AI might move from giant cloud data centers to running locally on devices with new NPUs, or even on standard laptops. Ultimately, the delegates concluded that the future of AI won’t just be decided by the hyperscalers. It will be shaped by consumers and engineers through the products they choose to use and the companies they criticize for wasting resources.