People don't have issues paying for metered electricity, heat, and water. AFAIK most people are aware of when they're wasting power/heat/water but don't go to extremes, e.g. don't run the shower constantly but don't shower with a bucket.

There is no "pay $25/month for unlimited electricity" option. And if there was, everyone would use that and all metered options would go broke overnight. Followed fairly shortly by the company offering unlimited flat-rate deals.

Which is what TFA describes, and is why there is no unlimited flat-rate deal for utilities. But that's a mature market that is not relying on growth for valuations, and isn't appealing to VCs trained on net/mobile/crypto bubbles.

I don’t know about you but where I live the electricity companies definitely offer a “flat rate” for electricity.

But those contracts are so much more expensive that virtually no one gets them.

Where do you live that you can get a flat rate for unlimited electricity? I've never heard of this (I live in Greece).

It existed when I lived in Pennsylvania, but it always had a “if you exceed this threshold consistently we will move you to a metered plan” conditional. More a way to plan your budget than truly unlimited.

Yeah, that's not unlimited. Useful data point, though. Potentially a solution for the AI companies?

Dunno if it counts because it's not unlimited usage for a flat rate, but in the Netherlands a lot of utilities like water, gas and electricity are billed at the same flat rate every month, then once a year they do a calculation and you get money back (or have to pay more if your usage was higher) based on actual usage.

I can't see any situation where true unlimited usage for a flat rate, even a wildly expensive and uneconomical one, would make any sense because as often happens you'd basically be incentivizing people to waste as much gas, electricity and water as they possibly can to "get their money's worth" or whatever, and we should be encouraging the exact opposite for those 3 things.

Same in the UK, it's mostly to spread out the seasonal cost.

Although here, if people get into debt they can be forced onto pre-payment meters where that's not an option, and the unit price is also higher. There's a lot of controversy about this.

I agree with you, that's why I can't believe it exists anywhere. What you mention is basically to reduce the costs of counting, and is still usage-based metering, just with a reasonable guess to make payment more frequent.

Power in Greece works the same way, except you (are supposed to) get the guess one month and the count the next (and pay the difference). In practice, they count less often than that.

> What you mention is basically to reduce the costs of counting, and is still usage-based metering, just with a reasonable guess to make payment more frequent.

That's not the motive where I live. There's basically 2 reasons: (1) customers (esp. those, often less well off, that resort to “money earned, money spent”) like the predictability of it; and (2) depending on the season spending changes significantly (heating, cooling), whereas salaries don't.

Usually the corrective amount is also paid out over a period of months if it exceeds X% of one month's spend.

Here in Norway, especially 10-20 years ago but still not uncommon today, you will find flats for rent with electricity included in the monthly price. Student housing especially.

We own an old 170 sq.m. semi-detached house, and the electricity bill for June was 20 euros. That's with all heating & hot water coming from electricity (heat pumps) and owning an EV that we charge at home.

That's the same for the UK but ultimately the cost is absorbed by someone, just not the tenant.

I work in heavy industrial manufacturing and we have flat rate contracts. In fact, we MUST use a minimum amount or we get billed a penalty.

In manufacturing, everything is about consistency.

Interesting. Is there an upper limit too?

People don't have stable or seasonal "token needs" or wants.

Nor do they have stable don't-give-me-any-better "answer quality" expectations from models. At least not yet.

AI model services don't have any version of the typical utility's geographical market protection or monopoly.

Worse, they have all been funded for massive loss-leader hyper growth, so to survive in the short term, they have to compete on who can raise and lose more.

Then looking into the long term, unlike the Uber's of the world, the point where a market will reach saturation, survivors finally "win", and margins can transition from negative to positive, keeps moving away.

I don't see any way in which utility bills are economically comparable to this situation.

The only "hope" the general AI service business has, is that at some point one of them pulls far enough ahead of the others that everyone else's economics collapse first.

Like a very high speed version of the centralization in advanced chip fabs. But with no profitability until the centralization happens. Brutal.

Those utilities are fairly easy to track in a way that digital services aren’t. Lights in my home consume energy at a consistent rate. Ditto for water coming out of the faucet; if I open it more, I can see it moving more water and costing more.

On the web, it’s not clear that one website is being massively more wasteful than another. From your perspective you’re engaging in the same behaviour (opening a website) but for reasons outside your control you may be spending more than intended.

I remember not having unlimited internet traffic at home and always having to worry. These days, I pay a flat monthly fee for internet access.

But leaving a light on 2x the time will equal very close to 2x the price.

Asking “what day is today” vs “create this api endpoint to adjust the inventory” will cost vastly different. And honestly I have no clue where to start to even estimate the cost unless I run the query.

Except of course that they do:

"Texas freeze raises concerns about ‘ridiculous’ variable rate bills" (2021)

HOUSTON, Feb 23 (Reuters) - In Spring, Texas, about 20 miles (32 km) north of Houston, Akilah Scott-Amos is staring down a more than $11,000 electric bill for this month, a far cry from her $34 bill at this time last year.

"What am I going to do?" Scott-Amos, 43, said. She was among the millions of Texas residents who lost power during several days of bitter cold that caused the state's electrical grid, operated by the Electric Reliability Council of Texas, to break down. "I guess the option is, what, I'll pay it? I just don't feel like we should have to."

<https://www.reuters.com/business/energy/texas-freeze-raises-...>

This instance involves variable rate billing systems, where the cost per unit use can vary tremendously. There are also stories you'll find where equipment issues (broken water or gas mains, malfunctioning electrical equipment, failure to terminate per-minute-billed telephone calls, etc.). I can remember the latter featuring in some compendium of records (possibly Guiness) for a family who hadn't realised that they needed to hang up the phone between calls and got hit with a staggering (at least for the time) bill. More recently, data and roaming charges tend to be culprits:

<https://worthly.com/most-expensive/expensive-phone-bills-tim...>

In all of these cases, a significant problem is that usage is all but entirely divorced from metering, and people have no idea of what their actual usage patterns are until "invoice therapy" arrives.

(A term I'm borrowing from a friend, though I'll note that monthly feedback cycles on what are typically hourly/daily decisionmaking patterns proves a poor systems control theory match.)

Utilities are infra-as-a-service.

The point is that infrastructure is non-optional, you're not going to decide not to use it, and you rarely switch (in the case of utilities you often CAN'T switch). You set it and forget it.

AI in the context of an end-user product or tool is not that (unless you're running the AI product company, where it becomes infra for you).

It's a blender or a kitchenaid mixer. Let's say companies gave those away for free but only charged metered fees to use them. If you had to pay $4 to blend something or run your mixer (and weren't sure if the tool would get it right the first time or you'd have to re-do it 5 times) you'd use them much less. They are optional.

You'd treat the decision to bake like a financial one. Nobody asks their spouse if it's okay to take a shower.

LLMs are not essential for survival.

Yet.

Ever.

My point is that many technologies which were once thought of as luxuries eventually come necessities. Socially if not physically.

Adam Smith of all people writes of this:

By necessaries I understand not only the commodities which are indispensably necessary for the support of life, but whatever the custom of the country renders it indecent for creditable people, even of the lowest order, to be without. A linen shirt, for example, is, strictly speaking, not a necessary of life. The Greeks and Romans lived, I suppose, very comfortably though they had no linen. But in the present times, through the greater part of Europe, a creditable day-labourer would be ashamed to appear in public without a linen shirt, the want of which would be supposed to denote that disgraceful degree of poverty which, it is presumed, nobody can well fall into without extreme bad conduct. Custom, in the same manner, has rendered leather shoes a necessary of life in England. The poorest creditable person of either sex would be ashamed to appear in public without them.

<https://en.wikisource.org/wiki/The_Wealth_of_Nations/Book_V/...>

Running water, sewerage, gas service, home postal service, electricity, automobiles, telephone service, Internet service, mobile phones, universal healthcare, and many other utilities were once considered luxuries but came to be recognised as essentials.

I don't need my linen shirt either. I'll be just fine, naked in the forest.

Jokes aside, I already don't live without my LLM. Rely on it too much. Though more for comfort than survival.

Naked in the forest is underrated, until the ticks and mosquitoes arrive ;-)

Fair points, I'll add that I'm not sold that LLMs will become indispensable. There's plenty of flashy tech which hasn't.

But of the past 20--30 years' or so offerings, it seems to have a reasonably high chance.

If you bring the LLM to the forest and prompt it for how to avoid the ticks and mosquitos it will most likely tell you to put a linen shirt on.

But what you're missing is imagine you could pay a flat fee for all your utilities... and that fee is quite low. And imagine a very competitive marketplace where the switching costs are basically zero. What do you think will happen?

There are utilities which offer a uniform-monthly-billing option. PG&E in California does so as "Budget Billing":

Budget Billing is a free program that helps you easily manage your monthly energy costs . We calculate your monthly payment amount based on your average energy costs over the last 12 months and adjust your payment amount each month, so that you don't have big spikes on your bill.

While it's not a savings program, Budget Billing helps you stay in control of your bill by avoiding seasonal bill spikes.

<https://www.pge.com/en/account/billing-and-assistance/financ...>

That’s not the same at all. That’s a personalized plan based on your usage but smoothed out over the year for predictable bills, in contrast with Claude/openai’s uniform $20/$200/mo plans. You can’t use as much electricity as you want in that PGE plan. You still pay for what you use.

If everyone’s usage was above the set rate they would include rate limits which is literally a usage limit, which they have proven to do.

So no in fact you do get X tokens for 20 dollars and off your usage is too high they would absolutely put a stop to that. It wouldn’t even surprise me if they have per customer rate limits.

Where is the distinction here, it isn’t an unlimited plan, the limit just isn’t communicated.

That's the default in Germany due to yearly meter reading.

It's not at all flat rate, though, as you'll settle the balance after the year like a tax filing (refund or back pay).

PG&E's programme also works with a re-balancing of payments to account for over- or under-utilisation. My understanding of recent changes to the system are that those adjustments themselves are also spread out over a number of months, again to increase overall predictability and avoid surprise billing.

Ahhh, hmmm, fair. The German approach to surprises is to take the tariff sheet that basically boils down to a flat connection fee + price-per-kWh (the meter counts in kWh), sometimes split over multiple counters for tariff's that distinguish between prime time and off-time consumption, and often coming in two or three variants which are chosen between based on what the counter says (they tend to bump the connection fee and drop the price-per-kWh above a yearly kWh threshold). Then the consumer goes to look at the meter whenever the consumer wants, calculates via simple delta (subtract baseline value from current counter value) how much has already been used in this measuring period, and compares to for example the previous year around the same time. Or if usage is sufficiently even across a year, one does some simple linear extrapolation ("x days have passed since the start of this metering period, y_start was the start counter value, y_now is the current counter value, z days long is the current metering period"->"z * (y_now - y_start) / x").

I assume the utility support would explain this way of calculating/predicting/tracking if one asks nicely.

The reasons for not smoothing out further than monthly payment intervals in a yearly measurement interval is AFAIK due to regulatory restrictions on credit/debt, and because this is more due to a metering process efficciency/overhead limitation than because they inherently want to offer these fixed/steady payments: you don't want to send out a meter reader more than about once a year, and doing it that often also conveniently spreads out seasonal variations letting the bill predictions(/estimates used for calculating the monthly payments that are due) work much nicer and easier to understand than some summer/winter split tracking.

Note this information is all about the default/baseline tariffs, which the local electrical energy seller (which is a distinct entity from the local electrical power/grid provider, who is the one that has the natural monopoly) can't refuse due to credit score or the like and which can't have any remotely long term/notice periods. Many tariffs are quite similar in style, though, due to the inherent limitations of yearly-read non-smart energy consumption counters.

Around now seems to be where finally you as a customer can demand upgrade to a smart meter which notably (from my understanding) gains the ability to live-report energy flux for 15 minute quantization/intervals, matching the European super grid's electricity market. A nice side-effect closely related to that market quantization is that these meters allow selling PV production to other units/properties nearby (at least where the connection/current path doesn't actually go through grid operator owned wiring, e.g. units in an apartment building) without having to pay grid fees on that energy sale despite the energy going through the meters instead of being sold through a behind-the-meter bypass feed/line. That "virtual" metering of the hyper-local PV energy sale makes it just much easier in terms of infrastructure, compared to some actual real-time load-balancing setup that somehow spreads an apartment building's monolithic PV array's power across multiple units's behind-the-meter breakerboxes taking care to not wastefully oversupply any one as backfeed would at best be compensated at about a third of the all-fees-included retail rate (it's about the same as the grid fee part that's payable by non-local buyers of electricity, both around 10ct/kWh) instead of 1:1 compensating the approximately 30ct/kWh buy-rate of some local apartment's meter. Also, less active infrastructure at a scale that doesn't afford on-site 24/7 staffing for monitoring and customer-side-uptime-enhancing redundancy features is always good. I'm counting a mere meter that only meters and doesn't steer/control electricity as passive infrastructure from the power flow's POV. Especially if it's built to fail in a way where power still flows but the metering/counting is broken, at least where there's more than just yearly reporting from the meter (and thus non-metering would be caught quickly).

Electricity uptime of this grade is something with perhaps underappreciated benefits: pretty much all but critical systems and those that run non-crash-safe software will not require a UPS, as power outages at least in urban areas (except those where local distribution systems drown in climate change flash flooding) happen once every few years, either due to larger-scale blackouts (entire city/metro/state) or from non-redundant local infrastructure dying of old age.

Imagine all your home appliances ran on magic instead of electricity, and you paid the magic company for each unit of magic used, except you had no idea how much magic each appliance needed to operate.