Why AI Computers Cost THOUSANDS (It's Not What You Think!)

Description

Chapters

Transcript

0:00 If you're sitting there thinking like,
0:02 wait, $4,000 for a computer or $10,000
0:05 for a computer in an era where we can
0:07 get a decent computer for 1,200 bucks,
0:09 why would I do that? Again, because
0:14 AI tokens
0:16 um are very valuable things and you can
0:19 you can think of it less like I'm buying
0:21 a machine and more like I'm I'm buying a
0:25 potential employee
0:27 that I can train up to do stuff for me
0:30 that works 24/7.
0:32 The implication of that
0:36 is that
0:38 as AI capabilities get better and more
0:40 people discover them and AI starts to
0:43 take over work and more people start
0:45 using it at work, more people start
0:47 using it at home, more people start
0:48 using it everywhere
0:50 the
0:51 demand for for AI inference, the the
0:54 ability to use these things is going to
0:56 go up dramatically. Now,
0:59 Apple and and
1:01 AMD and Nvidia and all of these
1:03 companies that are going to make
1:05 specialized hardware to to to push
1:08 inference to what they call the edge
1:12 so
1:13 to your local computer is going to help
1:15 with that, but
1:17 the hunger
1:19 for for training AI and for using AI is
1:23 going to skyrocket. So, I don't see data
1:25 centers going away anytime soon. What I
1:27 do see is data centers
1:32 on the ocean.
1:33 So, I saw a proposal today uh of a
1:36 company that is doing floating data
1:38 centers where the wave action generates
1:41 the electricity and the ocean water
1:43 cools cools the things. Um
1:47 Elon Musk and a couple of other
1:48 companies are talking about putting data
1:50 centers in space. Um there's likely
1:52 going to be data centers in the middle
1:54 of the desert with massive solar farms.
1:56 Um there's likely going to be data
1:58 centers next to hydroelectric dams or
2:01 substations where there's spare
2:03 electricity. There's like there's all
2:06 sorts of plans going on.
2:08 Um I don't see them going away at all.
2:11 Why do we need all these data centers?
2:13 Well,
2:14 AI
2:16 So, if you think
2:18 think steam engines
2:21 and then think electrical grid
2:24 and then think Interstate Highway
2:27 and then think internet
2:30 and then think AI. Um computers,
2:33 internet
2:34 um
2:36 AI is going to be infrastructure-level
2:41 um
2:42 capacity for every country. So, if you
2:44 don't have enough compute to do all the
2:47 stuff, to run all the robots, and run
2:49 all the cars, and
2:50 you know, the fleet of robo taxis that
2:53 Tesla's going to put out and everyone's
2:54 going to put out um
2:57 your country will not be able to
2:58 compete. So, having access to
3:01 high-powered inference
3:03 will literally be power.
3:06 Like like electrical power.
3:08 Right? If you're going to live without
3:09 AI compute, you're not going to be able
3:11 to do as much stuff. So,
3:13 um yes, local inference will help. I
3:15 think
3:17 most
3:19 inference for individuals will migrate
3:21 out to the edge.
3:23 Like we will have phones within the next
3:25 3 years that are so powerful and then
3:28 models that are so good that are that
3:30 are quantized for phones that everyone's
3:33 individual device like this will will
3:35 likely be doing things as powerful as
3:38 what I'm doing on a 64 gig MacBook Pro
3:40 right now. But the high-end compute
3:42 stuff, the I think the the appetite for
3:45 that is insatiable.
3:47 Watch the full replay at
3:48 community.thesalon.ai.