Data center emissions probably 662% higher than big tech claims
theguardian.com
theguardian.com
If we as a planet need to be using better sources of power so that our planet literally doesn't melt our food sources, and then us, then our governments need to make it more cost effective to use better sources. Carbon credits or energy swaps don't work to actually reduce the amount of polluting energy produced.
It’s about the last place you want to start using intermittent power sources.
Chemical battery storage is simply not feasible today for such uses cases. If you go renewable your choice currently is to back that intermittent source with either coal or natural gas at nearly a 1:1 basis.
Nuclear starts looking pretty compelling for such use cases.
The dirty secret of renewables is that they are currently directly linked to natural gas peaker plants. This was obvious a decade ago, and still is just as true. My portfolio shows I was completely correct in this prediction.
Chemical storage at these tens to hundreds of megawatt scale is currently a pipe dream. Current deployments are measured in single digit hours of storage. Perhaps another decade will show things have changed, but I doubt it. Nuclear is currently the best battery tech we have available. Especially considering how greenwashed methane production has been.
And that’s not even getting into seasonal energy storage which we simply do not have any answers for short of hydro. Interconnections only work so long until you run out of other people’s coal or nuclear power sources.
I suppose you could redesign your facility to operate the emergency diesel backup generation on a regular basis but that’s even worse.
I’m very into solar and battery storage where it makes sense - even if not the most economical. I think anyone who can afford it should have a solar setup plus a few days worth of batteries installed at their place of residence. Focusing on datacenter power use is about the last thing we should be looking at after we optimize the rest. It’s like focusing on aviation as your priority for co2 reduction. Why go after the highest hanging fruit when so much low hanging fruit currently exists?
It is true that peaker plants have grown alongside renewables. And that combination has reduced CO2 emissions, which was the whole point. And mostly it isn't peaker plants but more conventioanl CCGT's.
>> Chemical storage at these tens to hundreds of megawatt scale is currently a pipe dream. Current deployments are measured in single digit hours of storage.
There is over 80GW of battery storage capacity in the pipeline for the UK alone. And GW scale schemes are already active and being built. And whilst most schemes are still small (in MW and MWh terms) the grid is perfectly well designed to handle that.
>> Interconnections only work so long until you run out of other people’s coal or nuclear power sources.
And particularly renewables where there is excess generation in geographically distant places.
>> I suppose you could redesign your facility to operate the emergency diesel backup generation on a regular basis but that’s even worse.
Or just connect to a grid with a healthy mix of renewables, storage, interconnection, and yes nuclear. It is not an all or nothing question.
>> Why go after the highest hanging fruit when so much low hanging fruit currently exists?
Grid scale generation and usage is the low hanging fruit. That is why companies focus on it so much! A solar farm with 80k panels is cheaper to develop than 20k rooftop installations. A large operator like Google can make a significant difference on their own.
The US doesn't like it because you end up with plutonium doing that. It's contrary to their geopolitical/non-proliferation goals if we start using it because we simultaneously want fairly arbitrary foreign states to use nuclear power (so they don't feel as compelled to start wars over energy), and we don't want arbitrary states having the ability to create nuclear weapons. So pretending that reprocessing fuel isn't the best way to approach long term nuclear power generation is what the DoE/DoD/DoS are stuck on, even domestically.
https://blog.rootsofprogress.org/devanney-on-the-nuclear-flo...
Asking for taxes to go up is fine, but nobody wants to pay more taxes.
I feel like you underestimate the energy that sacrifice releases. I also feel like you don't have children. Because if you did a sentence like "you know you're going to be dead before the end of the century, right?" would probably not be what you were thinking. Exactly when you have something to lose (the life of your children and grand children), it generates a desire to change this miserable state that the world is in. This is not a depressed or negative state. It's the opposite. It feel like diving into the issues of the world and not looking away.
And I think the injunction to "Enjoy!" is exactly what's wrong with the world and what leads you away from inner peace. This is what psycho analysts have discovered quite a while ago and what religious people are able to avoid by having higher goals than your own enjoyment. My own experience with this is that since I have given up trying to enjoy my life and I'm trying to wrestle with my own negativity, life is better.
The most important tools in this “creative accounting” when it comes to data centers are renewable energy certificates, or Recs. These are certificates that a company purchases to show it is buying renewable energy-generated electricity to match a portion of its electricity consumption – the catch, though, is that the renewable energy in question doesn’t need to be consumed by a company’s facilities. Rather, the site of production can be anywhere from one town over to an ocean away.
Suppose you can get solar power in Kansas City, but can't transport it to Seattle, maybe due to lacking interconnect capacity, maybe lack of contracts among the power companies, it doesn't matter. What you do is buy 1GW of solar in Kansas, sell that same power as fossil power in Kansas and buy fossil power in Seattle.
The Guardian thinks it's creative accounting, IMO it's a fair and good hack: When you open a new site and use that hack to buy power for the new site, the corresponding new power generation is solar. It's just solar with two additional accounting hops.
There are other variants of the same hack.
In the case where I came across it, the problem was contractual. Only one vendor could supply power to that site. So by applying that hack, ① new renewable energy generation was built ② a conventional power plant lost a contract with Otherco and ③ another conventional plant gained a contract of the same size. After a while the contractual problem was solved.
IMO it doens't make sense to choose site based on something as flimsy as a contract. Contracts are changeable on a timescale of months or years, buildings last longer.
If we had a world without wind it would be different, but we do have wind on this planet.
The damage is done regardless of where the emissions happen. Shuffling the location of the emissions around does not help.
The Guardian blames companies who use regenerative power, and who pay to have regenerative power generation expanded faster than would otherwise be possible. IMO that blaming hurts.
So in KC you're +1 solar, -1 fossil. In Seattle you're +1 fossil. Net: +1 solar.
If you buy existing clean energy to compensate for your emissions, overall emissions will remain the same, no matter which hacks or accounting tricks you use. That clean energy would be produced and sold anyway, because the marginal costs are so low. If you can't avoid emissions and want to compensate them, you must invest in new renewable capacity that replaces fossil fuels somewhere.
It only works as a clean bypass of you buy so little you don't impact the local fossil market, but this is a discussion about large data centres so that's basically never satisfied.
I buy my power from a company that supplies 100% regenerative power and gets most of its power from new generation facilities, 350 of them so far. Those 350 deliver power to about a half-million people, which obviously took market share away from existing fossil plants and pushed the market prices down.
So by that test, what action does not prolong the use of fossil fuel? You're making the test much too strict.
In Seattle you've increased demand without increasing generation, raising the price there. Yes, fossil generators may be built to satisfy this new demand, but that's unlikely in 2024. In Seattle in particular, excess demand is generally met by buying hydro from BC.
If I’m understanding this correctly:
1. You buy 1MW of solar power in location X, and don’t use it.
2. You sell 1MW of emissions producing power in location X, that someone else uses
3. You buy 1MW of emissions producing power in location Y, and use that.
Net result: 2MW of emissions producing power used, 0MW of solar power used, and a claim of net-0MW of emissions produced?
Yes, it’s a “clever hack”, but accounting sleight of hands are just that. Most charitably magic, less charitably outright falsehoods.
> IMO it's a fair and good hack: When you open a new site and use that hack to buy power for the new site, the corresponding new power generation is solar. It's just solar with two additional accounting hops
I don’t understand how you got to this outcome from the first position. Can you help me understand what steps I missed for this to actually have good outcomes resulting in solar power actually being used?
I’d love to be wrong and for this to actually be actually good outcomes combined with clever accounting - those are absolutely the most satisfying solutions, both to design and read about (for me personally, anyway): where clever people are coming up with creative solutions to solve bureaucratic bullshit, while also achieving good outcomes.
EDIT: Child comment has helped me spot the error in my comprehension leading to incorrect math: “ What you do is buy 1GW of solar in Kansas, sell that same power as fossil power in Kansas” - I didn’t read ‘same power’ properly in my first read. I’ve left my comment as-is despite it now being completely wrong as a result of that, and glad to have been wrong
No, the 1MW solar power was used. It was used by someone who believed that it created emissions. But the emissions were instead produced in Seattle rather than Kansas. The accounting relies on the person buying power in Kansas not to care that it's fossil fuel power that they're purchasing. They know that there's emissions from their power use, but they don't know where those emissions are.
Thank you!
I think it's a really clever hack. An admirable way to drive forward installation of regenerative power plants, faster than the infrastructure/contracts would otherwise allow.
The effect is to either shift the attribution of emissions to some other company who doesn't care, or to cause renewables to be double-counted.
Consider a toy grid, with two users A and B and generators C(lean) and D(irty). All connect to the same point. A buys a certificate saying that 100% of their energy comes from C. But that does not inform B about anything - they might reasonably look at the grid and then claim that 50% of their energy is clean, when in fact by arithmetic 100% of it is dirty because all the clean stuff has been bought by A.
See e.g. https://industrydecarbonization.com/news/the-trouble-with-eu...
If you have on-site generation then this is a bit different because you can at least reduce your grid consumption. But otherwise it's not hugely different to buying carbon offset credits and assuming the trees will exist in 100 years.
I mean on paper it's a good idea, but there's not enough restrictions. Emission rights, for example, should be bought (from the government) in the first place, at an exponential price (so the more you need the more you pay). Supply should be reduced every year, with a goal being zero. They should not be tradeable across country (or in the case of the US, state) borders. Same with these renewable energy certificates mentioned, the further away the source, the higher the price should be.
But that's all bookkeeping shenanigans; it doesn't mean that e.g. data centers or industry will reduce their energy usage. That's what's broken about the push for renewable energy; it isn't a replacement of existing power production, it's on top of, and any added capacity is bought up and used.
Internally we had an interesting discussion, about travelling. Some one brought up that it might not look good that we fly to frequently, or at all. That's true, but it's also a very visible and easy to understand thing to focus on. One engineer does some quick napkin math and suggests that we optimize our code instead, as that would save more CO2 yearly, compared to us not flying. To me that's a really good indication of something that's wrong in the industry, we're not optimizing our code anymore, we're barely able to get rid of rarely used code and infrastructure, because keeping it running isn't really that expensive anymore.
Again, data centers shouldn't lie about emissions, but what difference would it make it they didn't? We're the ones buying the capacity, rather than moving to on-device computing, rather than an always running cloud. We spin up way more infrastructure than we should, because it's easy, and it doesn't cost that much.
Legacy tech stack are especially wasteful: HW + OS + App running 24/7 regardless if it used or not.
My intuition is that AWS claims they can reduce CO2 emissions from your onprem stack by 80% because *IF* it was rewritten into a serverless app, it would consume energy only when it runs.
That's one big if! https://aws.amazon.com/de/blogs/mt/how-to-drive-the-discussi....
We tried to move so many customers to some sort of container, auto-scaling EC2 (i.e. turning the instances off when not in use), Aurora, Lambda, you name it, anything that doesn't necessarily run 24/7. The customers don't like it. They don't feel safe, it's no familiar to them, they don't understand how they should manage it or make up reasons as to why they need to run MariaDB one EC2 or why EFS or S3 won't be able to replace their EC2 backed NFS share.
When it's done correctly, seeing an entire business running without a single VM or physical server is really a amazing thing to see.
To measure it, we use your detailed billing data and local kWh carbon intensity.
To reduce it, we are offering a cloud governance tool based on Cloud Custodian to identify and remediate wasteful or/and unsecure cloud resources.
Thought it was quite good: https://emmlab.info/Resources_page/Data%20Center%20Fights_di...
Individual frugality is the precondition for collective degrowth, both practically and morally. Otherwise you just open business opportunities for datacenter growth.
IMO both should be pursued, make one and don't omit the other.
Demand will fall in the countries with dirty electricity.
Would need a solar farm with about 50x-100x the land area.
Even if you assume 1 floor and a roof covered entirely in solar and the lower end of the power estimate, that's only covering 13% of the power requirements. If you look at the higher end of power estimates and assume a 3 storey data center with 50% of the roof covered, a more realistic set of numbers for a hyperscaler, that's only 0.4%.
As for being synergistic for emergency power, batteries would help here, but peak power usage is probably around dusk when solar won't be that useful but when power is most likely to be disrupted.
Peak power usage for large scale facilities is not much different than off peak. Most megawatt plus builds I’ve been part of have less than 10% difference between peak and valley in terms of watts used. Perhaps the hyperscalers see a much larger difference but I have my doubts.
and to think of all the emissions being spent on useless AI