The absolutely wild thing is that this is being reported as a
good thing.
This prediction isnt coming from Fortune, its coming from OpenAI themselves, who have shared it with the media. OpenAI want people to think this.
The idea that their product is going to require more than 30% of the entire US's energy supply, more than entire countries, is delicious to shareholders. That's a lot of money changing hands, a lot of infrastructure that will need to be built. So much growth. It is a level of power usage that implies their product has become critical to humanity, instead of just being an extremely inefficient bullshit generator.
...Its also a lie. Yes, OpenAI is consuming a silly amount of energy, but this is just more OpenAI marketing being repeated as fact. I called it an "idea" for a reason, OpenAI would love you to believe this is going to happen because the moment we stop shoveling money into them, the bubble is going to pop and we might have another global financial crash.
“It’s pretty amazing,” Chien said. “A year and a half ago they were talking about five gigawatts. Now they’ve upped the ante to 10, 15, even 17. There’s an ongoing escalation.”
Just another 17 gigawatts, just another trillion dollars, trust us it's gonna work. They just keep saying stupider and stupider numbers to fuel their own stock price, and the media blindly reports it like its guaranteed. The numbers left the realm of reality some time ago.
In fairness to Fortune, they do at least consult one person who is like "what? That seems impossible" but the headline still reports it as fact. They also note that the centers would cost about $850 billion, and that NVIDIA's pledged funding to build them totals... $100 billion. And the media still acts like this is 100% going to happen, because Sam Altman said so, and the guy selling AI is definitely a trustworthy source on AI.
For an outlet called Fortune, they really aren't very concerned by how Sam's wildest dreams are basically a trillion dollars short, and will be missing roughly 100% of the power he claims they'll need.
Ed Zitron's blog has looked at the deal in more detail, and its even more fucked. That $100M investment from NVIDIA isn't even upfront - OpenAI only get it after they build about $325 million of datacenter. Hey, wait a minute-
Wait, its okay, because OpenAi are also going to get Oracle to build $400M of datacenters! No, there's no actual progress on this, and Oracle is currently trying to fundraise $15 billion dollars, aka bugger all. Where is either of them going to get that money? Who the hell knows, I'm probably mixing some figures here, but Sam sure as hell doesn't know either, especially when he claims they're going to build an additional gigawatt of compute EVERY WEEK.
Over here in reality, it takes about two and a half years, and also YOU ARE A TRILLION DOLLARS SHORT, WHY ARE WE JUST REPEATING ANYTHING YOU SAY-
OpenAI's current revenue is about six (6) billion dollars.
Revenue, not profit, they only ever report revenue because they burn it all in a massive pit and lose billions each year. They can totally fund all of this if they just magically grow to several hundred times their current size,
and also figure out how to make pure profit from a business that currently makes negative profit on billions of dollars of investment.
so yeah, AI energy usage is horrendous and might screw any slim chance we had of mitigating climate change. But OpenAI want the situation to be orders of magnitude worse than that, because stock price go up. We are in the bad timeline.
I wonder if there's a single author who wrote about AI ending humanity without the AI even existing, simply because we all lose our minds and destablise the global economy trying to give one guy the money he says he needs. I don't know if there's a solution to this, but it is terrifying that this is what all our systems are building towards. We could be ending world hunger at a fraction of this price.