I don't see how this follows. Data center operators buy energy and this is almost their only operating expense. Their products are priced to reflect this. The fact that basic AI features are free reflects the fact that they use almost no energy.
I would be surprised if AI prices reflect their current cost to provide the service, even inference costs. With so much money flowing into AI the goal isn't to make money, it's to grow faster than the competition.
I think it’s more instructive to look at providers like AWS than to compare with other AI labs. What’s the incentive for AWS to silently subsidise somebody else’s model when you run it on their infrastructure?
AWS are quite happy to give service away for free in vast quantities, but they do it by issuing credits, not by selling below cost.
I think it’s a fairly safe bet AWS aren’t losing money on every token they sell.
> For the purposes of this post, I’ll use the figures from the 100,000 “maximum”–Claude Sonnet and Opus 4.5 both have context windows of 200,000 tokens, and I run up against them regularly–to generate pessimistic estimates. So, ~390 Wh/MTok input, ~1950 Wh/MTok output.
Expensive commercial energy would be 30¢ per kWh in the US, so the energy cost implied by these figures would be about 12¢/MTok input and 60¢/MTok output. Anthropic's API cost for Opus 4.5 is $5/MTok input and $25/MTok output, nearly two orders of magnitude higher than these figures.
The direct energy cost of inference is still covered even if you assume that Claude Max/etc plans are offering a tenfold subsidy over the API cost.
Thank you for some good intel. Thats very interesting. But, I wonder how this affects supply pricing to other customers. Not that you haven't shown the direct power costs have been borne, but the more indirect ones remain for me.
> I would be surprised if AI prices reflect their current cost to provide the service, even inference costs.
This has been covered a lot. You can find quotes from one of the companies saying that they'd be profitable if not for training costs. In other words, inference is a net positive.
You have to keep in mind that the average customer doesn't use much inference. Most customers on the $20/month plans never come close to using all of their token allowance.
Sites Reservoir isn't going to do a damned thing for municipal water systems in most of the state. You have to remember that there is not such a thing as a statewide municipal water policy. Every city or region has its own thing going on. The Sites capacity is dedicated to its investors, so depending on where you live it could be a helpful resource, or it could be irrelevant.
If you look into the actual design capacity of our municipal water systems, many of them were designed for far larger populations. The EBMUD, for example, intentionally secured 325 million gallons per day in upstream capacity because that was 10x the needs of the service area in 1929. Implicitly they assumed that the service area would grow to 4 million people, but it never did, primarily because of zoning. Today EBMUD delivers only about 120 MGD. We could more than double the service area population without water issues.
The statewide rain totals for the 2025-2026 water year so far rank 6th out of the years of the 21st century, so aren't that remarkable in context. Do you live in a place that got slapped with a peculiarly high rainfall?
California is big! That's also why there have technically been small parts of California which have been in drought for the last few years while most of the state is in good shape.
This year, Southern California is having a wet year while most of Northern California is having a relatively dry one.
We're north of Los Angeles and the area has never really handled rain well. This is also entirely anecdotal having lived here for ~35 years.
Some of the towns in our county have developments built on floodplanes. In our neighborhood, only some streets have storm drains so many of them flood. On one of the main roads numerous trees fell over damaging walls and homes.
That last set of storms that really stands out were the El Niño events in the early oughts.
I wonder if overall rainfall doesn't tell the whole story. From my experience in SF (and admittedly CA is big and people will have very different experiences) there has been an enormous amount of rainfall early in the season and then another enormous amount over the holidays, but the rest has been dry. The total may not be that much but the acute heavy storms have been pretty intense.
Well, it’s also not what I was thinking about. There were some huge floods. It may have been more than a year ago. When you get to my age, time does weird things.
Nevermind. It’s not something that’s worth any agita. It’s obvious that it’s not something that left many scars.
Same with SoCal - it's the size of NY State. CA's big and somewhat evenly populated (at least compared to similar states out east) so there's inevitably some form of environmental issue somewhere. Not to minimize these incidents ofc.
Heavy rain is usually very localized. I live in Norcal and I've seen many situations where we were getting hammered with multiple inches an hour while a few dozen miles away it wasn't raining at all, and vice versa. So even in a wet year whether your neighborhood gets slammed is a crap shoot.
The average length of ownership is a pretty warped statistic, though. It is dependent on when in the car's life cycle someone buys it. At one end of the market are new car buyers who keep them longer than average, at the other end are people who constantly buy end-of-life junkers for $500.
Something to note here is that the result of xargs -P is unlikely to be satisfactory, since all of the subprocesses are simply connected to the terminal and stomp over each other's outputs. A better choice would be something like rush or, for the Perl fans, parallel.
It is true though, in some cases. The embodied carbon of a Rivian for example is never paid back by operations. So you do have to exercise good judgement in which EV you choose. The category doesn't always win.
In my case I already own a hybrid that I only drive 2000 mi/yr and there is not yet an EV that I could buy with so little embodied carbon that it would make sense to do so. At the rate China is decarbonizing, presumably the embodied carbon of their EVs will soon be minimal, but not yet.
Thanks for linking that Rivian document later, but I think it doesn't support this claim. I'll stick with it's weird 155,000 mile lifetime for the comparison.
Rivian: 60,140 kg carbon per lifetime.
F150, at 20mpg: 78,740 kg carbon, for fuel alone.
So even ignoring the embodied carbon in an ICE vehicle, and paying comparatively high embodied CO2 cost of a new Rivian, it's better to switch immediately, (if CO2 were the sole concern, which it never is.)
An F150 is also a poor choice, so I don't think of it as a point of comparison. As an approximation, the mass of any object is related to its embodied carbon, so smaller vehicles embody less of it. Massive vehicles embody current emissions and that is worth considering.
That only works if you take it at face value that buying either of them is a rational transportation choice, which I reject. Even if I accept the people need a weird truck-shaped thing with a useless 4.5-foot bed, a far better choice on emissions grounds would be the Ford Maverick XL, which has a battery 1% as massive as the R1T's battery, yet this tiny battery cuts the per-mile GHG emissions in half. The embodied carbon payback distance of an R1T versus a Maverick XL is over 100,000 miles.
My kid races mountain bikes so I have become extremely familiar with Rivian (and Cybertruck) MTB Dad, and I think they are a joke. With only a little planning I can get three bikes and three riders in a Honda Insight, while R1T Dad needs an optional accessory to get even one bike in the bed. People choosing these things are, 99% of the time, not behaving rationally. They are buying luxury goods that they believe signal their environmental credentials.
I don't find either truck a rational choice, but the car market can stay irrational longer than the climate can stay liquid.
The cultural irrationality of the truck/car market in the US crosses all ethnicities and class lines. If we are trying to evaluate the effectiveness of EVs, I think we need to compare the Rivian to the closest fossil fuel powered vehicle, even if it's something that causes me to disrespect the people making these choices.
> The embodied carbon of a Rivian for example is never paid back by operations
Really? I could imagine it being significantly longer than an average EV, but never? Regardless of driving pattern? Got a link or can you show your math?
That sounds damned near useless for typical data analysis purposes and I would very much prefer a distributed system to a system that would take an hour to fill main memory over its tiny network port. Also, those cost $400/hr and are specifically designed for businesses where they have backed themselves into a corner of needing to run a huge SAP HANA instance. I doubt they would even sell you one before you prove you have an SAP license.
For a tiny fraction of the cost you can get numerous nodes with 600gbps ethernet ports that can fill their memory in seconds.
That is not a valid interpretation of the data. The ratio you cite, which is a pointless one, is mostly influenced by household size. SF has a relatively small household size compared to the state and nation. The vacancy rate you cite is also not a useful one that people generally understand. There were 19000 units for sale or rent during the last ACS survey, out of 418000 physical dwellings, and that's only 4.5% which is very low by historical standards.
