FWIW, there isn’t much correlation between where tech company headquarters are and where tech company datacentres are. There is value in placing them close to users, but they don’t need to be particularly close to employees (except for the small fraction of employees that do physical maintenance in the datacentres).
I don’t know of many super-current datacenter operations which have any resident personnel, except maybe physical security. Operations and monitoring are remote, and maintenance and repair can be dispatched from miles away (perhaps even states away).
I see downstream that there’s a lot of scenarois, more or less realistic.
Re a HX in a river, there’s gotta be a really funky river if the HX is gunked up just by the water flowing through it. I guess you 'mericans hafta have really funky rivers compared to what we have here west of the pond if that’s a real concern.
It’s not just “gunk”.
YouTube video breaking down all of the problems of putting any AI datacenters in space, much less 100 Gigawatts worth of solar panels to power them, which is what Elmo proposed putting into space every year - which is, by the way, 25% of the entire US power grid’s capacity.
I think you’d be surprised at how gunked up stuff like this can get even in clear waters. It’s a matter of time. They may be fine for a year or two but crud will collect and needs to be dealt with which is expensive.
Why? If you can launch a satellite you can shoot one down; it’s basically the same technology.
I can.
No protesters, maybe.
Oh, I don’t think this is such a big risk. The proposed systems would still be distributed, it wouldn’t be one orbital structure the size of an Imperial Star Destroyer. AIUI, the base AI model (weights) is large but pretty static, and its profile for each user represents a smaller, separate bundle of data. In real-life (terrestrial ) cloud, there are already mechanisms for background data transfer for redundancy across regions. Current Starlink satellites can exchange data amongst themselves. So I don’t think your interaction/history with the system would necessarily be lost if the hardware failed on a single satellite.
Reference the wiki section: limitations of anti-satellite weapons.. You’d need a liquid fueled rocket for medium earth orbit or higher, and shooting down a satellite has never been attempted at that altitude or with that technology. You’re trying to hit something smaller than a car moving at 9000 mph from 12,000+ miles away. It’s a tall order.
It’s possible in theory, but gnoitall said “for now” which seems very much to be the case.
Shooting them down is actually much easier, since you just need to get to that altitude; you don’t need to actually get into orbit.
Well, I don’t know about super-current, but the datacenter I worked at around a decade ago had resident personnel the last time I checked about five years ago, and still offers colocated services. Part of the colocated services is on-demand access and on-demand remote eyes and hands. It’s almost impossible to offer the latter without on-site staff.
Maintenance and repair can be dispatched from miles away, but it’s much faster when they’re on-site. Also, maintenance is much cheaper when you don’t have to pay for your electrician or A/C tech to travel and stay in a hotel.
I’m not talking about intercepting a satellite with a suborbital rocket, which is tricky in terms of timing (to say the least), nor am I talking about taking down small military satellites with defensive systems of their own. I’m talking about knocking out a big, dumb space station, slowly and safely, with an orbiting kill vehicle. We’ve been doing orbital rendezvous since 1962 - it’s hardly new technology. The only difference is that one of the vehicles will explode (or employ some sort of non-nuclear EMP).
… and the damage need only be relatively minor.
On the ground, light bomb damage gets repaired.
If light bomb damage stops a datacenter operating, the fact that a person with a spare cable or two could fix it becomes irrelevant if you can’t get anyone there because it’s in orbit.
Many would disagree with this assessment of Musk’s non-idiocy as he’s promoted many crazy ideas, both technical and social. According to his various predictions in the past, Tesla would have had full self-driving (FSD) many years ago such that today no one would actually need to know how to drive a car, and he’d have a colony on Mars by now. Though no one would doubt his financial cunning, Musk is not entirely sane.
The reality is that Musk is a trillionaire on paper because of the IPO of SpaceX, and virtually the whole of SpaceX revenue comes from government contracts. And no government is going to promote data centers in space. The problems are so extensive that the idea is basically comical.
Putting such a thing in orbit would create huge problems in all the areas so critically important for large data centers: the need for reliable ultra-high-speed bandwidth, the need for huge amounts of power, the need for massive cooling capacity, and the need for rapid access for maintenance. It’s just not gonna happen.
To be fair, the other day my car (HW3, older FSD 12.6) drove me across town to Krispy Kreme, and then back home, no intervention ended on either trip. (…but then tried to turn into the neighbour’s driveway…)
Another issue is the square-cube law. The bigger the data center, the more heat it needs to dissipate. OTOH, the heat to dissipate is the proportinal to solar panel size. The other issue is that the coolant, if like the ISS, needs to be pumped - a constantly moving part that makes it less reliable, more prone to breeakdown, not to mention the risk of coolant leak, which is much more an issue in orbit where repairs are difficult - plus it would need a supply of maneuvering fuel… Also, unless you get to a fairly high orbit, the satellite will experience regular eclipses as it goes behind the earth, meaning a decent-sized battery is required - another item that adds a lifespan and weight to the system. If the orbit is too high, outside the shield of the earth’s magnetic field, more subject to radiation interfering with electronics. And there will still be brownouts in the Earth’s penumbra from time to time.
Definitely some on-site staff beyond just security (and the data centers I work with have no colocated services, just our own maintenance). There are power systems to monitor and people to install new machines in the racks if nothing else.
That being said, if you picked a big tech employee at random, they very likely do not have physical access to the data center, nor do they live near one.
Oh, totally true. I don’t have access to my current company’s data centers. Even though I could arrange it if I had a purpose, I wouldn’t be allowed to bring even a phone in with me. Corporate owned data centers are generally really tightly controlled.
Exactly, just put a lot of ball bearings at the same orbit and watch the fireworks.
ETA: not exactly the same orbit because then they’ll have the same speed and not make much of a dent, but a suitable intercepting orbit.
To also be fair, in matters of safety, “works most of the time, more or less” is not anywhere near good enough.
If I had a Tesla with FSD, I too would consider it a fun feature, but that isn’t the point. As with many technical challenges, getting to 80% success may be relatively easy, it’s the remaining 20% that takes most of the time and effort. Teslas under FSD with an inattentive driver have been involved in some serious crashes. It will be a very long time before any jurisdiction waives the requirement for a driver’s license for the owner of any FSD-enabled vehicle. By “very long time” I mean “maybe never”. This is yet another one of Musk’s ketamine-enabled pipe dreams.