And I think it’s no coincidence that this is when we’re seeing a big AI push. We’ve had machine learning for at least 20 years now but have had to be precious about the resources we can afford to go to that kind of compute (I still remember “donating” my screensaver compute time to SETI and genetics processing). Now that it’s not completely unfeasable thanks to renewable energy, every tech company and their mom is all in. We truly need an energy budget for the planet or we’re just going to be circling the drain indefinitely.
We truly need an energy budget for the planet or we’re just going to be circling the drain indefinitely.
Sure would be nice, unlikely without a world government, which would likely be bad as centralization of power tends to be, but likely better than Pax Americana. Solidarity and sympathy.
Now that it’s not completely unfeasable thanks to renewable energy
Are we living in the same reality? Renewable energy has nothing to do with machine learning or especially LLMs being feasible. The data centers used/being built for AI use some of the dirtiest form of energy you can use, because it’s cheaper or easier available.
I think you’ve misunderstood my point here. I’m not saying that data centers care about renewable energy, quite the opposite. I’m saying that the buildout of renewables has lead to enough surplus energy (in the intention to replace fossil fuels) such that tech companies said, what if we use that cheap power instead, and built up the demand for that power such that fossil fuels are not decommissioned but even increased like you pointed out. Off-grid power for data centers is because this demand is outstripping even the excess supply right now because everyone is demanding a piece of the pie. It’s wholly unsustainable and when it hits a wall it is going to be ugly.
Then I misunderstood that, but I still don’t agree. The “surplus energy” of the renewables is not really there. Granted, in some regions (not really sure about the USA, though, I don’t have data on that), renewables can provide > 100% of the required electricity for a few hours per day, but that’s not really a lot of surplus energy. If it was, they wouldn’t have to build power plants just for the data centers.
You even kind of mention that with
because this demand is outstripping even the excess supply
Electricity price definitely is a factor for data centers, but having electricity 24/7 is even more important to them. Also, what made AI feasible was the usage of GPUs or even more specialized hardware, that is tailored to those tasks. Machine learning as a concept has been around since the 1980s or even earlier, not just the last 20 years. Paradoxically, that increase in efficiency caused a drastic increase in total power consumption for AI.
And I think it’s no coincidence that this is when we’re seeing a big AI push. We’ve had machine learning for at least 20 years now but have had to be precious about the resources we can afford to go to that kind of compute (I still remember “donating” my screensaver compute time to SETI and genetics processing). Now that it’s not completely unfeasable thanks to renewable energy, every tech company and their mom is all in. We truly need an energy budget for the planet or we’re just going to be circling the drain indefinitely.
We won’t circle the drain indefinetly, we’ll collapse at some point
Sure would be nice, unlikely without a world government, which would likely be bad as centralization of power tends to be, but likely better than Pax Americana. Solidarity and sympathy.
Are we living in the same reality? Renewable energy has nothing to do with machine learning or especially LLMs being feasible. The data centers used/being built for AI use some of the dirtiest form of energy you can use, because it’s cheaper or easier available.
Point in case: Data centers are being built off-grid using gas turbines. In fact, the planned amount of gas-powered data centers use more electricity than Germany.
Also, they are not even using proper gas power plants for that, mostly just gas turbines, which are cheaper and faster to install, but way less efficient.
They don’t care about the power use of their data centers, and especially not whether they are renewable.
I think you’ve misunderstood my point here. I’m not saying that data centers care about renewable energy, quite the opposite. I’m saying that the buildout of renewables has lead to enough surplus energy (in the intention to replace fossil fuels) such that tech companies said, what if we use that cheap power instead, and built up the demand for that power such that fossil fuels are not decommissioned but even increased like you pointed out. Off-grid power for data centers is because this demand is outstripping even the excess supply right now because everyone is demanding a piece of the pie. It’s wholly unsustainable and when it hits a wall it is going to be ugly.
Then I misunderstood that, but I still don’t agree. The “surplus energy” of the renewables is not really there. Granted, in some regions (not really sure about the USA, though, I don’t have data on that), renewables can provide > 100% of the required electricity for a few hours per day, but that’s not really a lot of surplus energy. If it was, they wouldn’t have to build power plants just for the data centers.
You even kind of mention that with
Electricity price definitely is a factor for data centers, but having electricity 24/7 is even more important to them. Also, what made AI feasible was the usage of GPUs or even more specialized hardware, that is tailored to those tasks. Machine learning as a concept has been around since the 1980s or even earlier, not just the last 20 years. Paradoxically, that increase in efficiency caused a drastic increase in total power consumption for AI.
I don’t think you’re reading what I’m writing, so I’m going to leave this here. Have a good one!