Posted by jatins 1 day ago
Think about the world before the 2017 "Attention Is All You Need" paper.
Did anyone predict that paper, what preceded and followed it?
Nope.
Same case now.
Maybe the word "prediction" is the problem; "guessing" would be better.
OK, what the heck. I'll make a prediction too:
You better buy SpaceX stock now. The way things are going in the US, the only way we build AI data centers at scale will be in space. Politicians have turned data centers into punching bags to be used to gain votes. We can't build power plants and people are being led to believe all kinds of things. Regardless of which, if any, are true or not, the rate of construction of AI data centers in the US is and will be seriously constrained by realities on the ground.
Hence my prediction: It's all going to space.
Even if you believe they can get cost to launch 1KG down to 100 bucks... (current falcon heavy is $1550) you still have the problem with all those solar panels.
1GW of space AI requires... 1 GW of solar to power it.
Current manufacturers of space solar is 1-2 MW &.. space solar aint cheap. $100 per watt. They estimate each spacex will be 250 kW. So.. $25M just for the solar for each satellite. Yeah you can use less efficient / cheaper solar, but then your weight goes up.
Also... Ground AI datacenters haven't nailed down how quickly GPUs depreciate. Accountants claim 6 years while CRWV says it's longer and most realists say it's shorter. It's based on how quickly GPUs will improve. Which is QUICKLY.
In space you'll have to depreciate GPU, all supporting hardware including the satellite, electronics, solar panels, and launch costs. On the ground it's just the GPU & you can even sell the old one down.
ground based AI needs power. $FRVO is an interesting ticker if you believe in their technology. Just announced a deal with google yesterday.
One correction: SpaceX makes their own solar arrays. The cost is nowhere near what you quoted. Source: I worked for SpaceX for a few years.
Also, you can generate a lot more power in space per unit area due to not having the limitations created by our atmosphere and weather. It's continuous generation at 30 to 40 percent higher per-unit-area radiation. Generation on earth is roughly an inverted parabola (on a good day without clouds and weather) that, at best, if you integrate the area under the power curve, delivers 66% of the total equivalent energy say, a nuclear reactor, could deliver during the same period. So, once you increase radiation from about 1,000 W/m2 (impossible to achieve on earth due to weather and other realities) to 1,400 W/m2, 24/7, no weather, etc. The scenario quickly becomes vastly different.
Geothermal is very interesting, thanks for the ticker, I'll look into it.
All that said, I think the US (and Europe, if they care to survive) needs to have, as a top national priority, a massive program for the construction of nuclear power plants.
A few years ago, I ran through an analysis of power generation needs to convert our entire fleet of vehicles to electric power. That required at least doubling our power generation capacity. This was the equivalent of having to build 1,200 nuclear power plants, each producing 1 GW 24/7.
This is impossible to achieve with solar, or wind, or the combination.
So, even if we decide not to care one bit about AI data centers, we need to double our power generation capacity (and infrastructure to deliver it).
Add AI data centers to that equation and the number could easily climb another 600 nuclear power plants.
Here's the problem that a naïve conversation with an LLM will not uncover: Humanity and Politics.
Could we embark on such a massive energy-generation project? Yes, absolutely.
Is it realistic? Nope. Not even close.
In other words, we can do it but it is impossible...which sounds like an oxymoron until you consider that we have lots of examples of projects that are absolutely plausible that, once they contact political and government reality, quickly become impossible. The best-worst example I always grab is the California high speed train disaster.
He's created a huge following (and is presumably making a lot of money) from pushing a hardcore AI-skeptic narrative, and I can't blame him for seeing that opportunity and running with it. We're ultimately all responsible for recognising these people and weighting their advise as necessary.
Additionally, from a public reputational perspective making bad predictions simply doesn't matter. In finance we're all aware of perma-bears who will predict the sky is about to fall, and when it doesn't just argue that the disaster is still coming, but is taking longer than expected, or that some unforeseeable thing happened which has compounded the risk but has for now kicked the can down the road.
So ultimately, it will be very hard to say Zitron is wrong unless he starts time-boxing his predictions, which I don't believe Zitron has done for obvious reasons.
That said I don't listen to him much at all. I've tried to listen to him a few times, but it's become evident very quickly that he doesn't understand the technical details enough to be making the predictions he's making and seems way too emotionally invested in the arguments he's pushing without good reason. I have strong personal filters for low-quality sources like Zitron – if someone raises enough flags I avoid them like the plague.
And I say this as someone who started as a doomer, and is increasingly a pessimistic pragmatist (“LLMs have value as tools, but not nearly as much monetary value as the main players believe they do”).
I get it though: for those of us who grew up alongside the net and tech sector, who loudly decried M$ greed for ME/Vista/8/11 but celebrated them at XP/7/10, who remembered when Google’s “Don’t Be Evil” was spoken with serious reverence, the current era of tech feels toxic and nauseating. Current AI is a prime target for that discontent, as are the companies whose motives very clearly aren’t societal progress so much as reality authoring and authoritarianism. In that vein, Zitron is magnetic because his entire position is “you’re right to be mad and they’re all going to die from hubris without you having to actually do anything”, which itself panders to the human desire for personally preferential outcomes sans individual effort.
Properly picked apart though, and he has as much substance to offer as the ardent boosters: a handful of “trust me bros” with a smattering of distractions to wind you up, but never actually address your concerns or questions.