Amazon Bets $150 Billion on Data Centers Required for AI Boom (yahoo.com) 26
Amazon plans to spend almost $150 billion in the coming 15 years on data centers, giving the cloud-computing giant the firepower to handle an expected explosion in demand for artificial intelligence applications and other digital services. From a report: The spending spree is a show of force as the company looks to maintain its grip on the cloud services market, where it holds about twice the share of No. 2 player Microsoft. Sales growth at Amazon Web Services slowed to a record low last year as business customers cut costs and delayed modernization projects. Now spending is starting to pick up again, and Amazon is keen to secure land and electricity for its power-hungry facilities.
"We're expanding capacity quite significantly," said Kevin Miller, an AWS vice president who oversees the company's data centers. "I think that just gives us the ability to get closer to customers." Over the past two years, according to a Bloomberg tally, Amazon has committed to spending $148 billion to build and operate data centers around the world. The company plans to expand existing server farm hubs in northern Virginia and Oregon as well as push into new precincts, including Mississippi, Saudi Arabia and Malaysia.
"We're expanding capacity quite significantly," said Kevin Miller, an AWS vice president who oversees the company's data centers. "I think that just gives us the ability to get closer to customers." Over the past two years, according to a Bloomberg tally, Amazon has committed to spending $148 billion to build and operate data centers around the world. The company plans to expand existing server farm hubs in northern Virginia and Oregon as well as push into new precincts, including Mississippi, Saudi Arabia and Malaysia.
Re: (Score:3)
I know plenty of people who are dabbling with LLMs for their work. I use it occasionally to generate regular expressions for scripts that I write. You're woefully uninformed to be commenting on this story.
Re: (Score:2)
So far these new LLMs have no applications but entertainment. The companies that want to sell them are promoting them for all sorts of other stuff, but they don't live up to the hype. These are bad investments.
It's only a bad investment if they can't foist the cost off on some unsuspecting C-suite that's slobbering on a salesman's bile-spew about AI. And sadly, there are plenty of C-suites willing to gobble up any old shit so long as its on the front-side of a hype wave.
Re:This is not a moneymaker (Score:5, Interesting)
I doubt that it's a bad investment. If they can't use it to run LLMs, they'll find another use. These are datacenters, they don't just have one potential use. If the AI doesn't fly, they can use them to run AWS or something.
Re: (Score:2)
I doubt that it's a bad investment. If they can't use it to run LLMs, they'll find another use.
the building and networking yes, but im not so sure about tensor cores being reusable... can anyone confirm?
Re: (Score:2)
They can always use them to generate bespoke AI porn and put millions of amateur OnlyFans out of business. /s
Re: (Score:2)
I worked in the AWS SOC for five years, and when I was there they were building data centers literally as fast as they could pour concrete and could barely keep up with the growth in demand. Andy Jassy, the current Amazon CEO, shepherded AWS from the days of "Let's try this, hold my beer" to "We're making so much money that Bezos has to pay dividends because he can't spend it all". He's almost certainly keeping close tabs on the demand and the need to keep customers happy with their offerings.
Fraction (Score:2)
Data center space is getting very pricey (Score:3)
Re: (Score:2)
Oncer people realize that, no, the singularity is NOT right around the corner, a far more likely outcome is that whole thing will level off. People figure out the productive niches for LLMs (there will be many) and train the programs to fill those niches. Once the programs are actually trained,
Re: (Score:2)
I'm not denying Altman's accomplishments
well i am
Re: (Score:2)
Once the programs are actually trained, they won't be consuming massive amounts of computational power like they are currently.
and you heard it here first folks, the end of training ai models is soon to end!
Let The Beast Rise (Score:1)
So all of our small businesses were sucked up by Amazon and now the profits from that event is funding the Beast. Welcome O Great Satan.
Re: (Score:2)
2/3 of everything sold on Amazon is presented by third party companies, and the vast majority of them are small and medium businesses. Thousands of businesses which would have folded up during the pandemic only survived on their Amazon sales.
Twenty years ago a small clinic, small manufacturer, accounting firm, etc. needed a DB server, an email server, a file/print server, and likely one or more standalone box for some app or process and half a dozen desktop machines. That wasn't enough to have an on-staff
It's easy to finance $150B moonshots... (Score:3, Insightful)
...when you earn massive profit and don't pay tax, (or at least have control over the piddly percentage you choose to pay). Just reinvest in your own future growth and world domination, risk of financial loss be damned. Of course you still need to pay for your lobbyists as a matter of doing business.
Vote for Joe Biden [archive.is] to collect tax where the money is.
Virginia/Amazon should put a Gen4 reactor near IAD (Score:3)
"For a few months in 2022, Dominion Energy Inc., which powers Virginia’s data center alley, couldn’t keep up, pausing connections to facilities that were otherwise ready to come online. The utility expects demand to nearly double over the next 15 years, with the growth driven primarily by data centers."
Re: (Score:2)
Just need to factor in a 100 megawatt fusion plant or three for every new data center.
https://www.datacenterdynamics... [datacenterdynamics.com]
Don't worry about Jeff (Score:2)
If the bet should turn out badly, we'll have to bail him out somehow.
You know what's gonna be funny? (Score:2)
Maybe not related, but .... (Score:4, Interesting)
I noticed in just the last week or two? My Echo Dots at home are suddenly MUCH more responsive than they've been for quite some time. As soon as I give a command like, "Alexa, turn off the living room light.", I get a near instant response of "OK" as the light is turned off.
It was getting progressively more sluggish up until now, and I assumed it was a sign of Amazon cutting resources for the Alexa-enabled smart devices. (They laid off a big chunk of staff supporting them, etc.)
I'm wondering if Amazon is throwing more resources at it again, now, with the idea it's needed for AI tie-ins?
Re: (Score:2)
I've noticed the same (plus it was having trouble with Spanish), I think that more likely what's happened is that the installed base of Echo devices had expanded to the point that they were straining the resources dedicated to them so they slowed down. The way that Amazon delivers resources to internal customers is that each is charged for the resources they use, when I worked in Amazon Physical Security we were charged X-amount every month for the CPU time, memory, bandwidth and storage that every NVR, ev
resouirces (Score:2)
In my lifetime batteries have become 10x (at least) more powerful, lights much brighter, cpus faster, magnets stronger and so many other limits of materials science have been breeched. It's great to see the new possibilities with these new resources, but I think the previous limitations gave rise to better solutions.
I'd prefer to see a 10x increase in LLM training performance come from an order of magnitude improvement in the training algorithms than from so much more server capacity coming online. Adding a
Re: (Score:3)
In your lifetime, algorithms have also become 1000s of times faster and more efficient. It just isn't as noticeable as tracking TFLOPS/watt improvements. This research paper [arxiv.org] claims the impact of algorithmic improvements in computer vision was roughly the same as the impact in improved hardware efficiency over the past decade or so. I've seen similar claims in other papers related to chess programs.
It's very likely you will see a 10x increase in LLM training performance from algorithmic improvements over the
It's a gold rush... (Score:2)
...and Amazon finds itself in a perfect position to sell pickaxes and shovels. Can't blame them.