Microsoft's Emissions Spike 29% as AI Gobbles Up Resources
pcmag.com
pcmag.com
I haven't paid attention to these kinds of optional disclosures. Never thought about it but were I asked I would have said these are advertisements. I don't dislike sustainability, but I thought those function as advertisements because you can expect to get more sustainability "for free" over time, because of many things (Moore's Law, ephemeralization, societal investment). So of course savvy corporations publish sustainability reports that say, "We're doin' great : )."
Therefore I'd argue their commitment to sustainability is shown by their disclosure of the increase.
... noticed that it doesn't even define "indigenous"
Didn't they just learn a lesson about the additional bad press you can be subjected to if you try to hide the severity of your security issues instead of just owning up to what's happening?
Alternatively, one could argue the increase shows their commitment to profit at the expense of the environment and the voluntary disclosure shows a commitment to greenwashing.
The US currently has 947 GW of renewable generation capacity that is not hooked up to the grid [1]. To put that number into perspective, the US currently has 1160 GW of generation capacity in total [2]. We could almost double our countries generation capacity if we could just hook up the power to the grid.
Unfortunately it takes 10 years to build new transmission lines in the US [3]. Changes are going to need to be made to the permitting process, which may include giving more responsibility and agency to FERC to oversee the review process.
Congress needs to act on this. If you care about this, you can educate your self and get involved. https://community.citizensclimate.org/topics/clean-energy-pe...
[1] https://emp.lbl.gov/utility-scale-solar
[2] https://www.eia.gov/energyexplained/electricity/electricity-...
[3] https://yaleclimateconnections.org/2022/10/permitting-americ...
Seems more likely to me that the utilities are expecting a large growth in consumption in the coming years and built out capacity ahead of time. Seems reasonable.
The 947 GW is the total proposed capacity for solar projects that are in various planning stages. They have not been built yet, and most of them will never be built -- historically only about 14% of proposed capacity ends up entering service [1].
This isn't to say that it's not important to invest in the grid to handle future needs. But it's not like there's a bunch of renewable energy just sitting there idle.
If most power generation is blocked by a lack of connectivity to the grid then it would stand to reason that most projects are abandoned or fail as they've no path to entering service (unless I've misunderstood).
I read the first comment quite literally. That there is over 900 GW of power that currently exists but just needs to be “hooked up”. i.e. a solar farm that is built, but through some bureaucracy there’s no line running from it to the grid.
With the context of the reply, that is not the case. Most of the cited 900GW isn’t even built yet. This seems like an important distinction when considering the scope of work needing to be done.
so nuclear wouldn't affect those 14%
What stops training AI's during sunlight hours only and hibernating as power decreases?
I assume that most cloud venders are over-provisioning because their margins are so good that they'd rather overspend on inventory than miss out on revenue most of the time.
It's not inconceivable that there's an answer that meets needs based on only running the most energy consuming operations during daylight and migrating those operations about the globe during sunlight hours while having less demand operations run at night.
There's a real issue that needs addressing surrounding ever increasing global emissions, giving lip service while continuing to increase emission related activity is double think.
> The vast majority comes from Scope 3 (96%), which includes the supply chain, data centers, and "the use of our products across millions of our customers."
> Yet the data centers remain an issue with no easy answer. The race to secure them is in full swing, and it would take years to switch them to renewable energy, experts tell PCMag.
That would only mean that this renewable energy would no longer be available for other uses. Like, for instance, replacing or reducing the use of fossil fuels.
What went up 30% is what they call Scope 3 (indirect emissions from suppliers, transport, construction and customers):
"Indirect emissions from all other activities up and down the value chain such as upstream and downstream transportation, materials, and end-of- life impacts, as well as all suppliers’ direct Scope 1 and 2 emissions.
Scope 3 represents over 96% of Microsoft’s annual emissions in FY23. Our Scope 3 emissions result primarily from the operations of our tens of thousands of suppliers (upstream) and the use of our products across millions of our customers (downstream).
Tackling Scope 3 means decarbonizing industrial processes such as steel, concrete, and other building material production for use in our campus and datacenter construction, as well as jet fuel for business travel and logistics."
Full report PDF: https://query.prod.cms.rt.microsoft.com/cms/api/am/binary/RW...
This is not the case. As long as electricity is fungible on the grid, and there's some fossil electricity production close enough for transmission, changes in renewable electricity use mean increases/decreases in fossil electricity use at the margin.
Sounds good. Reciprocally though, my skepticism in various green credit programs has only grown over time. It seems unclear how many efforts out there are green washing credits, to not yet mine or make some display that doesn't help greatly. But I do think if there's any hope for green economy, a $3T company probably has swagger & personel to eventually refine & drive towards effective spending. Hopefully. If they can be true to these declarations. Let us hope.
Carbon credit schemes may do SOMETHING to offset climate change but less than they advertise.
Note that I recognize many (possibly most?) carbon credits programs are bogus or worse, but trying to draw attention to circumstances where it's not always so cut and dry.
Even some of these, hilariously enough, despite how insanely quantifiable their benefits are, you see phenomenon like phantom credits like where Alberta, Canada gave Shell double credit for each tonne of Co2 they actually reduced.
I'm sure subsidies for green energy projects do reduce emissions, I just think the accounting is BS.
That being said, maybe Microsoft won't, and will buy the low quality "we totally were going to cut this forest but now we won't pinky promise" kind of credits, but I'd at least give them the benefit of the doubt over the average greenwashing corporation.
https://www.microsoft.com/en-us/corporate-responsibility/sus...
"In economics, the Jevons paradox occurs when technological progress increases the efficiency with which a resource is used (reducing the amount necessary for any one use), but the falling cost of use induces increases in demand enough that resource use is increased, rather than reduced. Governments typically assume that efficiency gains will lower resource consumption, ignoring the possibility of the paradox arising"
As there's a huge crowd who seems very convinced of the idea that the solution to handle our resource problems is to be found in various growth and tech oriented solutions.
Thinking of entertainment, it probably IS a climate win if you spend an hour at home watching Netflix or chatting with GPT, as opposed to driving around town or jetting across the world. Supposedly a GPT-4 query costs 0.01kWh - meanwhile, a Tesla consumes 0.35kWh a minute at freeway speed.
Nope.
Look back to the GP's paradox and the energy consumed watching even just the single most popular youtube video . . .
That isn't "energy saved" from "otherwise people would be flocking miles in cars to watch Despacito and Baby Shark Dance in theatres.
The AI training loads are over and above the already existing supercomputer modelling of land|sea|air fluid flows vie regular means - it's questionable whether LLM's et al even add anything of values in that domain (despite a plethora of papers asserting it to be so).
I expect that if you were to calculate "incremental energy per request" - how much "extra CPU compute" each request adds, you could probably get to that sort of value.
But odds are good that figure ignored all the training data collection, all the processing on that, storage of that digested information, retrieval of it, etc, and that sort of number tends to also skip things like "storage systems running to have the information available."
If I've got an entire datacenter running to provide services, and the request consumes, say, 3 GPU-minutes of time across all the nodes, sure. This is a sane value. It just ignores a lot of the other resources dedicated to the task at various points.
Microsoft isn't using 30% more energy than their current footprint on 10Wh/request AI answers.
Math like this is very much a "Tell me what answer you'd like, and I'll make it work out!" sort of scenario. An increase in data center use by 30% is harder to fudge.
'Is 10w/h per request the marginal cost of a request? Or does it factor in the fixed energy cost of the whole facility? Does it include training costs'
I'm inclined to lean towards it including at least some fixed costs. It seems rather high to be marginal cost. I have no gut feel for training costs though.
So, assuming it does include at least some fixed cost, using it more will reduce 'cost per use' while at the same time driving up actual consumption.
There is a group of people that trust LLMs for all searches but most tech-literate people know they hallucinate, and understand roughly how an LLM works and that hallucinations are indistinguishable from the truth to the model. And so we still Google things and read books.
The same applies to AI-generated art, music, voiceovers, and code. They are technically impressive, but rarely meet client expectations in any industry. And so most artists and programmers still produce the work themselves.
AI is fantastic for low-quality content, and this explosion of AI is emblematic of how people often lack a filter for quality of information they put in their brains. This isn't new; YouTube has always had more low-quality content than high-quality pieces, probably 1000:1 if not a higher contrast. However, AI shows the scale like no tech has shown it to us before (and maybe this was known by some people at Google or Bing but not so publicly). Almost everyone consumes AI work — blogs, movie posters, art, music, and code. The market is massive but all AI does is reproduce the existing work it was loss-minimized on and the loss of quality is mathematically necessary unless the model is not a neural net, but a database of literally the entire dataset itself that is being searched.
So this is very interesting. When did we become so okay with feeding ourselves a fast food diet of content? Isn't that causing more harm than good? The scale of this phenomenon is truly (and literally) industrial. I myself can hardly imagine what the modern life would look like if people turned to the internet, let's say, only to look up facts and information they already seek. We are so far from that. And AI both illustrates it, and exacerbates it.
For example I was looking for the name of a windows API command. I "knew" the command must exist, but didn't gave a clue what it was called. Asked Cgpt for an example program, and there's the name of the API. (Which I can then Google for docs.)
I also had a complicated-to-ask question about sun movement which it explained to me, along with site links to actual data.
I'm not using it as a Google replacement, but more of a Google supplement when the question is long-winded to write.
In my work niche, which at least 10,000 other software engineers do each day, ChatGPT 4 and 4o almost never give correct answers. Usually, they are misleading. I do find myself trying the LLMs when I'm faced with a challenging problem, but in that scenario they have not been helpful once.
Granted, there are areas of work that are much more popular and LLMs will be more helpful there. But these are individual scenarios, and we can find many where LLMs are fantastic and many where LLMs are awful if we wanted to cherry-pick.
Overall, the quality of the content is lower than what a human professional would do in their area of work, including flexibility, interactivity, and accessibility. This extends to books written by professionals and lectures given by them, as well as work carried out by them. It applies to home carpentry as much as it does to neurosurgery.
The more specific the knowledge has to be, the more this is true. The more generic, the less. But all-in-all, I think it's still very evident humans produce higher quality knowledge and content. And any quantized model of that content and knowledge will be unable to reproduce it at the same fidelity or quality.
With that said, I hope I expressed this enough — I do see your point in some circumstances. Just not overall.
I work in an industry adjacent to generative AI, so my views are very specific, and kind of beside the point. But I think the broader public does think what you describe. Much (and probably most) of AI-generated content that ends up on the internet is very barely changed by a human.
I am not saying gen AI is not useful, but that it's not efficient in the process of making high quality content. It's simply not steerable enough in practice. In creative industries, people are going pretty wild about how much work AI can replace, but all I've seen is mediocrity and failure when it is involved. Or frustration, as you say, that what it outputs is very difficult to turn into a high-quality product.
The interesting use case which has emerged is that there are a lot of times where it would be really helpful to me to have a short conversation with an expert on a topic adjacent to my own expertise. And it turns out that for those conversations, talking to ChatGPT is much better than talking to no one; it can help me with the kind of things someone would learn in the first few months on the job in that area, things a little too hard to google but where a human expert is not readily available.
I think this is the best, maybe the only, professional use case for GenAI right now -- advice and limited assistance in areas just outside your area of expertise, such that you don't need to depend directly on its output and can easily check/integrate the work.
A lot of feel-good articles and tweets will often talk about the plummeting costs of solar, or how the share of renewables is higher than ever, or stuff like that, which is, from a global warming point of view, absolutely irrelevant.
The only thing that matters as far as warming is concerned is how much CO2 is there in the air. The cumulative number. It doesn't matter if the growth is slowing, it doesn't matter if there's more solar than ever, etc etc.
So yeah, we're producing more clean energy than ever, but we're also using more carbon than ever.
[0]: https://ourworldindata.org/grapher/annual-co2-emissions-per-...
A company may be losing money today, but if they are growing revenue on a path to profitability, that is what ultimately matters.
But for purposes of global warming, you care about the total amount of carbon in the air. The cumulative sum. The integral.
If a company becomes profitable today, it survives.
If emissions were to magically stop today...warming continues!
So no, it's not about lack of renewable electricity.
https://www.economist.com/the-world-this-week/2024/05/16/bus...
There was definitely a window, maybe fifty years ago, where widespread adoption of nuclear energy would have stopped and reversed climate change. We may have had an increase in Chernobyls with the proliferation of non-modern reactor designs, but in this hypothetical reality maybe people would have been OK with that.
The problem today is that renewables are getting too cheap and too good, and the storage problem shrinks every day. Meanwhile, it takes upwards of a decade to license and build a single reactor. France's fancy new reactors won't be online until 2040. Nuclear is just too slow.
I feel like 10 years from now it won't even be a debate or a contest, nuclear will just be the most expensive option by a country mile. Greenhouse gases will peak within the next 2 years[0]. The nuclear lobby missed their chance, which does mean we'll have to deal with some effects of climate change we could have avoided.
My feeling right now (and I kinda flip-flop every few years) is that nuclear lost and it's economically infeasible to try again. Change my mind?
0. https://climateanalytics.org/publications/when-will-global-g...
https://www.ans.org/news/article-5842/amazon-buys-nuclearpow...
Zuckerberg says part of the problem is not just the amount of power needed for a SOTA AI-training datacenter (~1GW), but the fact that you need the power at that particular location which makes co-locating a power plant the best option. The biggest solar power plants in China put out over 2GW, but the biggest in the US is Solar Star in CA, wihch occupies 12 km^2 and only puts out 58MW.
There are many solar plants in US well in excess of 58MW.
https://en.wikipedia.org/wiki/Copper_Mountain_Solar_Facility is one, at 800MW.
It was at 58MW in 2010... which might confuse a LLM, though.
It's closer to improv jazz than to factual authority; still super wonderful, and worthwhile listening to, but not really for the purpose of learning how the original sounded when it was first recorded. Sure, you might get a sense of the original, but that's all it is: an impression. When you ask for facts, you get an impressionist render of facts, which sometimes, maybe even often times, accurately depict the Truth. But sometimes they depict the Truth the way an artist depicts the truth: if it feels right, it's right.
I discussed various climate related things with chatgpt and it doesn’t make such crazy mistakes.
https://www.theecoexperts.co.uk/solar-panels/biggest-solar-f...
The WikiPedia page for Solar Star indicates a capacity of 579MW, which may have been mis-parsed by a human as 57.9MW?
This is the top Google search result for "largest solar farm".
"The 15 largest solar farms in the world 2024"
https://www.theecoexperts.co.uk/solar-panels/biggest-solar-f...
Another top result for Solar Star is this:
https://8billiontrees.com/solar-panels/largest-solar-farm/
Which indicates (correctly or not) a current capacity of 314MW for Solar Star 1 (still less than the 1GW needed for Zuckerberg's projected SOTA data center).
The point I was trying to make, echoing Zuckerberg (who noted that power, not chips or data, is the constraining factor for further LLM scaling), is that power needs to be near to the data center, else the lead time and red tape will be even longer. If we're considering clean power then solar is an option, which limits datacenter location to where that is viable on this 1GW scale.
It's funny that you're the second person in this thread to assume that just because the data was wrong (out of date as it happens) it must have come from an LLM (which it didn't). I guess this is the world we are moving into, where all content is suspect of being AI generated and therefore suspect!
Hence China is still building coal plants, and won't really stop. Especially if it means the future of AI is being bottlenecked by it.
I fear their breakeven in 2024 will go the same way as the one back in 2018. (HN on that https://news.ycombinator.com/item?id=37676263)
That said they are building away on their latest Polaris machine. It'll be interesting to see how it works.
>Helion sets an ambitious goal to begin producing electricity by mid-2024, utilizing its innovative Polaris reactor. https://www.linkedin.com/pulse/helion-aims-fusion-energy-bre...
Anyone who is thinking AI will replace writers and artists and voice-overs and who knows who else ... it's going to cost more money than you think, and it's all getting added to the considerable bill we're running up that will end up costing us everything.
2012:
https://www.datacenterknowledge.com/archives/2012/09/25/micr...
2018:
https://www.datacenterdynamics.com/en/news/microsoft-wants-t...
2020: I wonder if anyone will hold them to this in 2030...
https://www.datacenterfrontier.com/energy/article/11428860/m...
Remind me how much carbon is emitted by workers being forced to commute to serve coffee to social media managers who are selling fast fashion to influencers?
2. According to the EPA's greenhouse gas calculator, this year's increase is equivalent to an extra 800,000 gas-powered cars on the road.
Whataboutism is when two entities are doing shitty stuff with roughly commensurate negative impact.
Versus dinging an endeavor that might have net positive impact (AI might REDUCE environmental impact) against forces that 100x larger and have no positive direction, it's actually just: pointing out when journalism uses a relative value to make a tempest in a teapot and distract from the big picture.
We can't just zoom in on 1% stuff that's working for the common good (at least in principle, if not, let's make THAT issue the article) as a scare story, and not talk about how it's a fraction of what is really up.
> Whataboutism is when two entities are doing shitty stuff with roughly commensurate negative impact
I have different view. What about'ism can be akin: "oh, you don't like that i drive a ar, what about the millions of cars driven everyday? And what about al gore! I will not stop driving until he stops flying"
Perhaps my understanding is wrong of what-aboutism, wanted to point out that there not be common ground on the definitions of terms.
FWIW, while AI might reduce carbon impact, as you have stated, so far it has not. Data centers use a lot of electricity, they are significant, and have a lot of projected growth ahead. With that said and news the impact has ramped up, so far we have only increased the challenge.
(Specifically the "President of Microsoft", which is a super fake sounding title.)