I think we should hold off on building data centers.
Current data center architecture is predicted to be obolete in 36 months.
AI data centers can be compsred to interstellar travel. In astronautics, this is sometimes called the wait calculation or the incessant obsolescence problem.
The idea is that launching a spacecraft too early can be counterproductive. A faster spacecraft launched decades later might overtake the original spacecraft and reach the destination first.
The same principle could apply to AI infrastructure.
Under a cooperative, socialist development model, society could optimize for long-term collective benefit rather than maximizing the return on each company's investment.
But there's an important distinction: We can mathematically demonstrate when waiting for better technology is advantageous. We cannot assume that socialism will automatically produce better technology. That depends on how research, investment, accountability, and innovation are organized.
If another spacecraft launches 30 years later at 50% of light speed, it arrives just 50 years after the original launch—beating the first spacecraft by 50 years.
The AI equivalent is building a massive computing facility now, only to discover that a much smaller, more efficient system built later can accomplish more at a lower lifetime cost.
The important engineering question becomes: At what rate of technological improvement does waiting produce a better cumulative outcome than building and upgrading continuously?
I think that is the calculation worth developing. It could incorporate processor efficiency, hardware replacement cycles, cooling, electricity consumption, capital costs, and the social value of having computing available earlier.
It might reveal an optimal construction schedule that looks quite different from today's rapid expansion.
AI data centers can be compsred to interstellar travel. In astronautics, this is sometimes called the wait calculation or the incessant obsolescence problem.
The idea is that launching a spacecraft too early can be counterproductive. A faster spacecraft launched decades later might overtake the original spacecraft and reach the destination first.
The same principle could apply to AI infrastructure.
Under a cooperative, socialist development model, society could optimize for long-term collective benefit rather than maximizing the return on each company's investment.
But there's an important distinction: We can mathematically demonstrate when waiting for better technology is advantageous. We cannot assume that socialism will automatically produce better technology. That depends on how research, investment, accountability, and innovation are organized.
If another spacecraft launches 30 years later at 50% of light speed, it arrives just 50 years after the original launch—beating the first spacecraft by 50 years.
The AI equivalent is building a massive computing facility now, only to discover that a much smaller, more efficient system built later can accomplish more at a lower lifetime cost.
The important engineering question becomes: At what rate of technological improvement does waiting produce a better cumulative outcome than building and upgrading continuously?
I think that is the calculation worth developing. It could incorporate processor efficiency, hardware replacement cycles, cooling, electricity consumption, capital costs, and the social value of having computing available earlier.
It might reveal an optimal construction schedule that looks quite different from today's rapid expansion.
