> “The biggest issue is a belief that AI should be how we solve everything, while ignoring the resources that it costs. This culture is omnipresent across tech.”
Well said. It reminds me of the peak of crypto hype, but worse and more pervasive. There's this attitude that no matter what the problem is, the solution MUST be LLMs.
No matter what the problem is, the solution will involve some form of machine learning, inference, or both, with massively-parallel processing that is probably (and unfortunately) centralized to some extent. Hence the need for data centers.
They don't need to be built in the middle of downtown freaking Seattle, though.
> No matter what the problem is, the solution will involve some form of machine learning, inference, or both, with massively-parallel processing that is probably (and unfortunately) centralized to some extent.
I think the point people are making is that this claim is not self-evident, and there's a remarkable lack of justification for it whenever it gets asserted. If you're convinced that this tool can literally solve every problem we have in society, it would help to explain why you're so confident about that. So far all I've heard ever is "exponential growth", which is not particularly convincing when a high school precalculus class gives you enough knowledge to be able to understand that there are curves that look a lot like exponentials before suddenly hitting diminishing returns.
Here in Reno the city council imposed a similar moratorium, buckling under a deluge of NIMBY pressure, citing talking points that are obvious bullshit to anyone who's actually set foot in any of the dozen or so existing local datacenters. All that it accomplishes is a guarantee that future projects and their tax revenue will move to the next county over — and then we'll be wondering why we're still stuck with crumbling infrastructure woefully undersized for our population and an economy dominated by a dying tourism industry.
If I was more tin-foil-hat inclined, I'd hypothesize that this wave of anti-datacenter activism is an astroturfing campaign pushed by the CCP to make sure the US deliberately refuses to compete with China in the technology sector. Or it's an astroturfing campaign by incumbent tech companies to block competition via regulatory capture and grandfathering. Or it's an astroturfing campaign by the agricultural sector (and/or companies like Nestlé) to deflect attention from their multiple-orders-of-magnitude greater water consumption. The reality's probably a lot less exciting, though: just a bunch of people who mean well and are rightfully opposed to Big Tech capitalism, but have been misled (probably by some or all of the above) into throwing out the babies with their bathwater.
That's easy enough to fix by charging datacenters at a higher rate than residential customers. Most electrical utility companies already have separate residential/industrial/commercial rates, specifically to prevent large-scale consumers from spiking small-scale consumers' prices.
Here in Nevada, NV Energy's in the process of getting state PUC approval for datacenter-specific “large-load electrical service agreements” specifically to ensure datacenters foot the bill for the infrastructure and generation buildouts needed to support them. Hopefully it goes through, since that seems to me like the exact right way to go about it.
> “The biggest issue is a belief that AI should be how we solve everything, while ignoring the resources that it costs. This culture is omnipresent across tech.”
Well said. It reminds me of the peak of crypto hype, but worse and more pervasive. There's this attitude that no matter what the problem is, the solution MUST be LLMs.
No matter what the problem is, the solution will involve some form of machine learning, inference, or both, with massively-parallel processing that is probably (and unfortunately) centralized to some extent. Hence the need for data centers.
They don't need to be built in the middle of downtown freaking Seattle, though.
> No matter what the problem is, the solution will involve some form of machine learning, inference, or both, with massively-parallel processing that is probably (and unfortunately) centralized to some extent.
I think the point people are making is that this claim is not self-evident, and there's a remarkable lack of justification for it whenever it gets asserted. If you're convinced that this tool can literally solve every problem we have in society, it would help to explain why you're so confident about that. So far all I've heard ever is "exponential growth", which is not particularly convincing when a high school precalculus class gives you enough knowledge to be able to understand that there are curves that look a lot like exponentials before suddenly hitting diminishing returns.
Here in Reno the city council imposed a similar moratorium, buckling under a deluge of NIMBY pressure, citing talking points that are obvious bullshit to anyone who's actually set foot in any of the dozen or so existing local datacenters. All that it accomplishes is a guarantee that future projects and their tax revenue will move to the next county over — and then we'll be wondering why we're still stuck with crumbling infrastructure woefully undersized for our population and an economy dominated by a dying tourism industry.
If I was more tin-foil-hat inclined, I'd hypothesize that this wave of anti-datacenter activism is an astroturfing campaign pushed by the CCP to make sure the US deliberately refuses to compete with China in the technology sector. Or it's an astroturfing campaign by incumbent tech companies to block competition via regulatory capture and grandfathering. Or it's an astroturfing campaign by the agricultural sector (and/or companies like Nestlé) to deflect attention from their multiple-orders-of-magnitude greater water consumption. The reality's probably a lot less exciting, though: just a bunch of people who mean well and are rightfully opposed to Big Tech capitalism, but have been misled (probably by some or all of the above) into throwing out the babies with their bathwater.
Could this headline be rewritten, "Employees of X ask Government to Stop Competition?"
They asked them last week. It's going to vote today.
Build them in Tukwila or Auburn, then.
Why do they need to be built in populated areas at all?
Shorter employee commutes? Shorter last-mile shipping distances? Lower latency to/from local customers? Closer proximity to points of intersection of fiber backbones? Closer proximity to existing electrical/water/sewer infrastructure?
Shorter employee commutes? Shorter last-mile shipping distances? Lower latency to/from local customers? Closer proximity to points of intersection of fiber backbones? Closer proximity to existing electrical/water/sewer infrastructure?
But we're being sold a vision of putting them in low-earth orbit. That means, among other things:
- They don't need a lot of employees to babysit the hardware, or in fact any at all
- They don't need water. Radiative cooling is evidently just fine by itself, even without convection or conduction
- They don't need any networking infrastructure that satellite IP links can't provide
- They don't need anything but localized photovoltaic power
Every argument for putting data centers in space applies equally to putting them literally anywhere on Earth.
Cynically, so that tech companies can get their power usage subsidized by the local population through their higher energy bills.
That's easy enough to fix by charging datacenters at a higher rate than residential customers. Most electrical utility companies already have separate residential/industrial/commercial rates, specifically to prevent large-scale consumers from spiking small-scale consumers' prices.
Here in Nevada, NV Energy's in the process of getting state PUC approval for datacenter-specific “large-load electrical service agreements” specifically to ensure datacenters foot the bill for the infrastructure and generation buildouts needed to support them. Hopefully it goes through, since that seems to me like the exact right way to go about it.