OpenAI wants 3000W to run a high-end multi-GPU server to serve 1000 users paying $20/month each. They've got $20,000/month of income from the server, if they can power it.
Grandma wants 3000W to run her central AC, so she can live comfortably in Phoenix, Arizona. She's got $2,300/month of pension income.
If you try to raise the per-kWh price of electricity to reduce demand, OpenAI won't be the ones who get priced out. It'll be Grandma.
Not the point you meant to make, but, maybe Grandma shouldn't live in a place that requires constant use of an air conditioner to be survivable.
Americans keep moving to places that are fairly inhospitable to humans - too hot, too little water, etc. It'd be nice to see migration patterns shift away from the currently popular deserts and hurricane-prone coasts.
The places with the nicest climates to build in also don’t want more building to happen and do everything in their power to prevent it, so people are migrating to those places, but people are also getting priced out of those places and moving to cheaper locales in the desert and in hurricane-prone areas.
Do they? Here in the UK it's been a huge problem for years now that residential rates have been locked to basically ~25p/kWh, but businesses have been paying 3 or 4 times that, it bankrupted so many places when their monthly bills literally went from like 2k a month to 8k a month. It's the reason why charging a car at home is so much cheaper than at public charging stations, they just cannot get the same rates since they are classed as businesses.
Commercial and industrial rates are lower than residential rates in the US and elsewhere. The bigger users are apparently better at negotiating rates while the residential customers take what they are given.
Which leads to the hilarious edge case of saving money by charging your car at a Tesla Supercharger instead of at home, because they're paying that much less for electricity than you.
In NY at least, Tesla super chargers charge rates above residential rates, so that does not actually happen unless you are grandfathered into unlimited free supercharging.
I left the UK in 2018 so this may be a strange question: I remember news about a massive but very brief energy price hike around when Russia began their "special operation" and everyone said "right, no more Russian gas", so is that hike still present?
I'd search, but the first few results get me geo-blocked because I'm not in the UK.
>>and everyone said "right, no more Russian gas", so is that hike still present?
Absolutely. Shell and BP still make record profits while home price cap has gone down a little bit but generally everyone is still paying through the nose for their electricity.
At least here industrial users have to pay spot power prices so they immediately see large rises in cost if their demand/power factor messes with the grid.
Grandma's AC doesn't run 24×7. Most utility districts have some sort of subsidy or rate limit program for low income customers.
Ultimately the only way to supply the constant power consumption demanded by data centers and keep consumer prices reasonable will be to build more fission power plants. This approach has several serious cost and safety issues but it's the only thing we're certain will work.
Except in this example she lives in Phoenix, so yes sometimes it absolutely does. At times the overnight low temp in the summer doesn't go below 90 degrees.
> but it's the only thing we're certain will work.
Nothing, not even nuclear, has 100% coverage with 100% reliability. And anything less than 100% can be handled with statistics. The goal is 99.99%, because that's the grid reliability so there is no significant benefit for going over that. 99.99% reliability with a combo of wind, solar and batteries is cheaper than nuclear.
The cheapest is 95% solar/wind/batteries and 5% natural gas.
We're using natural gas peaking so far, but this is actually still not that cheap.
I am of the attitude that even with no additional fossil fuel infrastructure constructed, we would be fine if we made an effort.
A mix for new generation of something like 70% solar/wind/batteries, 10% better grid interties, 10% expanded peaking hydro, and 10% a better system of load modulation, works just fine.
Load following with a market rate for things like water heaters, car chargers, and various industrial processes is something that's been implemented in other countries half-seriously, but only treated as an epithet or a scam here. To actually work, you can't just "Charge people a market rate" and expect them to fiddle with the breaker box, you have to actually integrate that feed into a decision-making algorithm run by home automation, which can make distinctions like "We need the car to be 90% full by 7AM, but we can adjust between charging at 1KW and charging at 6KW to minimize expected billing based on a minute to minute rate schedule. The water heater can range between 130F and 160F as required due to a mixing valve. The average dew point in the HVAC system needs to stay between 40F and 60F on a three-hourly basis, but can be modulated in ten minute increments according to demand." These sorts of interfaces need to be simple enough that Grandma can establish them conversationally.
Grid interties allow Los Angeles to sell Boston sunlight and Chicago to sell Houston wind.
Peaking hydro is just pumped storage minus the new pump and the new dam. An existing hydroelectric dam just gets refitted for twice as many generators and starts running on a 50% duty cycle, while allowing more variation in the water level.
Ive heard people claim that but the only example of decarbonizing to actually ever happen is France with nuclear and that was not particularly expensive (and happened 30 years ago)
- the carbon intensity of pretty much every country's grid has dropped dramatically in the last 25 years. First by replacing coal with natgas, then by adding renewables.
- nuclear cost France $2/W in 1982 dollars ($6.50 today). solar+batteries cost California $4/W in 2023 dollars.
- the late 70s was a lot more than 30 years ago. (I was born in the 70's, I wish I was still in my 30's).
You have quite a few wrong statements in your reply! The french nuclear fleet rolled out mostly in the 1980’s not the “late 70’s” but that’s unimportant. The main issue with your statement the “solar+batteries” doesnt include enough batteries to have reliable power. The amount of batteries installed in. California still depends on fossil fuels to meet demand on cloudy and windless weeks so its not deeply decarbonizing the way nuclear does
At $50/kwh of batteries + $50/kwh of battery install labor...
And at $150/kw of panels, $100/kw of inverters & copper, $100/kw of mounting, $100/kw of install labor, and $100/kw of real estate...
And at a 1W : 1kwh/yr (11%) duty cycle (my area should be 50% over that, but let's use conservative assumptions for low-maintenance vertical mounts)
And a level of conservativeness around power supply that reserves 60 hours of battery bank...
...
I can run a steady 1kw load (a GPU, let's say) for ~9kw of panels ($4950) + 60kwh of batteries ($6000), or $10,000. Capital cost of $10/watt.
Vogtle 3 and Vogtle 4 are going to end up averaging about 2.1GW for an estimated construction cost of perhaps $34 billion by the time everything is complete, or a capital cost of about $16/watt.
There is reason to expect the nuclear construction industry to drop in costs as it ramps up in size, but solar is already cost competitive here and it does not require the government to take over various costs like insurance, nor require the same political capital, approval process, or long financing period before initial payoff.
With that little battery you’re going to need to keep some natural gas plants running to be able to survive a few cloudy days in a row without a blackout.
Or just buy electricity from one state to the north or south where they didn't catch the same storm as you.
Or start running a nearby hydroelectric dam at 300% of average flow rate.
Or stop smelting aluminum in your attached aluminum smelter. Or equivalently, your bitcoin mine.
Batteries and panels continue to rapidly drop in price.
As your PV+battery system gets bigger in relation to your load, chance of a failure mode drops much faster than linearly; Not only are you looking at the far end of a bell curve from battery capacity alone, you're also generating more PV on the cloudy days.
When you take into account that nuclear takes 10 to 20 years to deploy, you could incrementally deploy much more solar and battery energy in that 20 years than the nuke plant could generate by the time it's assembled, its construction is completed, and put online.
OpenAI wants 3000W to run a high-end multi-GPU server to serve 1000 users paying $20/month each. They've got $20,000/month of income from the server, if they can power it.
Grandma wants 3000W to run her central AC, so she can live comfortably in Phoenix, Arizona. She's got $2,300/month of pension income.
If you try to raise the per-kWh price of electricity to reduce demand, OpenAI won't be the ones who get priced out. It'll be Grandma.