AI’s power problem: the slow grid
Explores the critical bottleneck of power grid capacity for AI data centers, highlighting transmission constraints and costly workarounds.
David Mytton is the CEO of Arcjet, building developer-first security as code tools, and author of the Console devtools newsletter. He is also a sustainable computing researcher at Oxford and an angel investor in developer-focused startups.
9 articles from this blog
Explores the critical bottleneck of power grid capacity for AI data centers, highlighting transmission constraints and costly workarounds.
Google's report details the measured energy, emissions, and water consumption of a single Gemini AI text prompt in production.
OpenAI CEO Sam Altman reveals ChatGPT's energy and water usage per query, sparking discussion on AI's environmental impact and data transparency.
Analyzes challenges in measuring video streaming energy use, advocating for rigorous measurement over modeling to improve sustainable computing research.
Analyzes the public health costs of data center pollution, critiquing a study on AI's environmental impact beyond just carbon emissions.
Analyzes why data centers are unlikely to adopt demand response programs despite academic interest, focusing on power constraints and uptime requirements.
Analyzes the limitations of using GPU manufacturer TDP for estimating AI workload energy consumption, highlighting real-world measurement challenges.
Analysis of updated US data center energy consumption trends and projections, highlighting the impact of AI growth and server efficiency.
Argues that effective security and sustainability require systemic changes, not user behavior modification, drawing parallels between tech and environmental efforts.