So, microalgae you say? Needs a lot of sunlight...AI is gobbling up the grid and water supply, apparently. The Pilbara data centre industry is taking off using Twiggy's excess solar for energy supply. How much groundwater would it need? Could Atmospheric Water Harvesting be used? Why not recycling water using biochar filtration? What about seawater to grow the microalgae? Why not base the data centres along the coastline? Desalination of water for liquid cooling? What about nasty chemicals added to the cooling fluid? Why not go back in time to water based liquid cooling? What about copper pipe corrosion? What about an anti-corrosion chemical that could be effectively removed by biochar? What about diesel generators? What about logistics problems getting diesel to the site? Why not produce Ultra Low Sulfur Biodiesel from the VFMAP biorefinery and use it in tuned diesel generators as a second backup option? How much bioelectricity could the VFMAP generate during pyrolysis? Which heat engine/Stirling engines to use? What about the VFMAP hardware and integrated control software using Machine Learning to guide efficient pyrolysis? How much R&D is needed? Could Twiggy or someone else fund it? What about the downstream tech value adds? In my opinion, it's all possible!!!
But - the idea of building a data industry in a hot desert, which presumably will get hotter with climate change over time, will take an enormous amount of power just to stay cool which means millions of PV panels. Could a cooling fluid be designed that can be specifically filtered with possibly modded biochar be achieved for cooling fluid recycling? And, ergo sum, what is the liquid actually cooling eg. CPUs, GPUs, RAM and SSDs? Might want to consult the tech giants eg. Bill Gates on that one...maybe NVIDIA, Intel or AMD have some ideas? Count me in if you need a liquid cooling biochar experiment...And then there's the cooling towers. Fuck. If the cooling towers go down, possibly at night time (more of an issue over the hotter months), the whole data centre will go down too until the power problem is fixed. Maybe giant installation of Fe redox flow batteries could work (plenty of Iron), assuming they're not going to fail too. I doubt there could be enough electricity from the backup bioelectricity and biodiesel, though both backup options can switch on with little delay. Failing that, there could then be outages of AI in Asia via a submarine cable that makes no difference without data flowing through it though many of us don't need AI to live. But if there's too much resistance in the cities (which should be based on fact - not uninformed fear as I believe it's a tech problem, not a Nimby problem), what other options are there? Or not do it in the desert either?
I guess, after all this argy bargy, I'm looking for Twiggy/other to come up with a rock solid AI data centre platform that is sustainable at it's core and can be adapted to climate change, extremely important for desert installations, and in the cities, without draining the civilian grid and water supply (which would be unpeaceful AI). But don't panic. There's a lot more science and engineering to do plus energy efficient data centre/cloud architecture can be achieved...but then, if many of us use cloud AI and keep using it plus an expanding human population that will also presumably use it too, AI will be increasingly at the centre of your use cases which will grow faster than just about any industry ever invented in the Planet's history. In my mind, the giant question is whether or not humans in the future can live peacefully alongside AI? Also, how much AI is needed, does it make us any more intelligent, and seriously, how intelligent does one need to get? I mean, I doubt any Gov can ethically determine how intelligent their population should get? I like to think AI is both on the path of human-machine convergence and augmented intelligence (I still have a brain)...but could a human ever become purely 'Artificially Intelligent'? Will AGIs take over? Can humans become AGIs? What about edge AI with efficient LLMs and low-powered hardware - AI without cloud? Maybe we need the 'best of both worlds' (Midnight Oil).
And so many other questions...I think it's just starting to get interesting. Time to tune in.
Write a comment