AI-related data center is frequently falsely framed by media. Water is and has always been a local issue. But setting that aside, the long-term problems in aggregate are huge. The planned 2026 buildout of data centers is, in terms of metric tons of carbon dioxide added to the atmosphere, equivalent to putting 50% more vehicles on the road. There will be an exponential increase in the number of methane leaks associated with buildout of new natural gas infrastructure. Methane is, far and away, the worst heat-trapping greenhouse gas — trapping +40% more heat. Then there is the issue of toxic runoff from construction sites into local watersheds (e.g. heavy metals). Then there is the issue of negative health effects from low-grade noise emitted by data centers.
We can’t burn up the planet fast enough? This is not Nimbyism”; it’s a suite of serious long-term environmental damage.
Finally, concrete is at the top of the list for contributing to climate change. At-the-edge fractal-computing will serve most of AI use. The hardware costs are multiples lower, +100% more compute can be achieved for the same unit of energy, and diffusion/distribution of fractal-computing hardware provides more data security. Big data centers are future soft targets for terrorism.
AI is eating itself. Cheaper, good-enough AI will likely commoditize most of the market. There isn’t enough enterprise use for the expensive models (Anthropic, OpenAI, et alia). So, why the inefficient and expensive push of LLMs on old technology? It’s likely an attempt by the big hyperscalers to continue to control the compute market. Cripes, it may reach a point where cheaper, good-enough Chinese AI models may be used in the U.S. to defend data against hacks by the big hyperscalers.
AI talent is diffused and can be scaled for most uses by smaller, independent providers. Bespoke AI can be brought in-house.
AI-related data center is frequently falsely framed by media. Water is and has always been a local issue. But setting that aside, the long-term problems in aggregate are huge. The planned 2026 buildout of data centers is, in terms of metric tons of carbon dioxide added to the atmosphere, equivalent to putting 50% more vehicles on the road. There will be an exponential increase in the number of methane leaks associated with buildout of new natural gas infrastructure. Methane is, far and away, the worst heat-trapping greenhouse gas — trapping +40% more heat. Then there is the issue of toxic runoff from construction sites into local watersheds (e.g. heavy metals). Then there is the issue of negative health effects from low-grade noise emitted by data centers.
We can’t burn up the planet fast enough? This is not Nimbyism”; it’s a suite of serious long-term environmental damage.
Finally, concrete is at the top of the list for contributing to climate change. At-the-edge fractal-computing will serve most of AI use. The hardware costs are multiples lower, +100% more compute can be achieved for the same unit of energy, and diffusion/distribution of fractal-computing hardware provides more data security. Big data centers are future soft targets for terrorism.
AI is eating itself. Cheaper, good-enough AI will likely commoditize most of the market. There isn’t enough enterprise use for the expensive models (Anthropic, OpenAI, et alia). So, why the inefficient and expensive push of LLMs on old technology? It’s likely an attempt by the big hyperscalers to continue to control the compute market. Cripes, it may reach a point where cheaper, good-enough Chinese AI models may be used in the U.S. to defend data against hacks by the big hyperscalers.
AI talent is diffused and can be scaled for most uses by smaller, independent providers. Bespoke AI can be brought in-house.
As they say, a lot to unpack there, John!
Love this format. Interesting and digestable, and something I don’t think I get elsewhere. Thank you!
Thanks, David.