# Anthropic Joins AI Energy Alliance, Apple Launching Servers

An episode of [AI Chat: AI News on Claude, ChatGPT, Gemini, Grok, Perplexity](https://invisiblenetwork.ai/podcast/1669990482), 16 Sep 2026, 12 min.

Audio: https://content.rss.com/episodes/365073/3159487/ai-chat-podcast/2026_09_16_21_07_13_b6d90cc1-faa7-4494-9c2e-97528de89f16.mp3

Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! In this episode, we explore significant advancements in AI hardware, including Apple's return to the server market and collaboration with Nvidia. Additionally, we discuss SK hynix's plan to produce memory chips in the U.S., as well as an initiative to optimize power generation for AI data centers. Chapters 00:00 Introduction 00:51 Apple's Server Market Comeback 03:14 SK Hynix and Intel Partnership 03:57 AI Energy Management Alliance 06:02 Power Utilization Challenges 10:55 Conclusion and Recap Show Links • Get the top 80+ AI Models for $8.99 at AI Box: https://aibox.ai • How I Grow and Scale My Business with AI: https://www.skool.com/aihustle • Get the AI Chat Daily Newsletter: https://www.aichatdaily.com/newsletter

## Transcript

[0:00] Welcome to the podcast. Today, I want to talk
[0:01] about a bunch of breakthroughs, AI news and build
[0:04] outs when it comes to AI hardware. I think we've
[0:07] had a lot of talk lately about should we slow
[0:10] down AI? Should we not slow down AI? Pacing the
[0:13] frontier with Dario Amadei. But the one thing
[0:16] that I think nobody is debating is the fact that
[0:19] this AI is going to need a lot of a lot of power
[0:22] generation, a lot of data centers, a lot of chips
[0:24] and a lot of memory. And all of those things
[0:27] aren't something you could say, let's pay. the
[0:29] development of because a data center might take
[0:32] 10 years to get a proper electrical hookup to
[0:35] it. These things, this infrastructure takes a
[0:37] very long time to get built out. The manufacturing
[0:40] facilities take a very long time, the planning
[0:43] that goes into it. So today I want to talk about
[0:45] some of the areas that we're accelerating on
[0:46] and some of the big advancements we're seeing
[0:48] in the hardware side. The first one I want to
[0:51] talk about is that Apple is going to is basically
[0:53] bringing back their server market. They're coming
[0:56] back into the server market. They're going to
[0:57] do this in 2020. with their M8 Ultra, and they're
[1:01] going to use Nvidia Lynx. This is a market that
[1:03] they left. And I'll talk a little bit about that.
[1:05] I also want to talk about the fact that SK Hynix
[1:08] out of South Korea is in talks with Intel to
[1:11] make memory chips in the United States. If you
[1:14] have been following what's going on with AI,
[1:15] you know that memory is one of the massive biggest
[1:18] bottlenecks. And there are only three companies
[1:20] that are currently doing this specific type of
[1:23] memory. At the same time, we have NVIDIA, Google,
[1:26] Emerald AI, which all launching something called
[1:30] the Alliance for Grid Flexible AI Data Centers.
[1:33] the AI Energy Management Alliance. And there's
[1:36] actually 18 members. So Anthropic is part of
[1:39] it. National Grid is part of it. AES, NRG, all
[1:43] of them are coming together to create an interconnected.
[1:46] grid. There's some interesting statistics that
[1:49] go along with all of that. Now, if you're the
[1:51] kind of person that hates paying for subscriptions
[1:53] to all of those companies I just mentioned, like
[1:54] Google and Anthropic and ChatGPT and everyone
[1:57] else, I'd love for you to check out my startup,
[2:00] AIbox .ai. I give you access to over 90 of the
[2:03] top AI models. So you get Anthropic, you get
[2:05] Gemini, you get Claude, you get Grok, you get
[2:07] DeepSeek, you get Perplexa, you get all of the
[2:10] top models, image, audio, video. every top model
[2:13] and modality that you could want. There's 90
[2:15] of them. It is $8 .99 to get started and you
[2:19] use it by the token. So if you have a massive
[2:21] project, there's higher subscription tiers. But
[2:24] if you just want to test stuff out, you can chat
[2:26] with all of these different models in the same
[2:28] chat thread. And you can also connect them inside.
[2:30] So if you use Clod a lot and you want Clod to
[2:32] generate images, you use our MCP connector. And
[2:34] inside of Clod, images can be generated by ChatGPT
[2:37] or video by... Google VO3 or whatever model you're
[2:40] working with. It is something that saves me so
[2:42] much time and I hope it's a super useful product.
[2:44] It is AIbox .ai. I'll leave a link in the description.
[2:47] Let's talk about what was going on with Apple
[2:49] first. So this is 15 years after they killed
[2:52] X server. If you'll remember, Apple is looking
[2:55] to get back into servers and how they'll essentially
[2:57] put this together is they will have four M8 Ultra
[3:00] chips and NVIDIA's NVLink Fusion. And all of
[3:04] this is how they're going to be building out
[3:05] their new server. basically a new server offering.
[3:09] I think what's interesting here is we've seen
[3:11] a lot of Apple is basically one of the companies
[3:13] that people have criticized heavily. They're
[3:16] like, oh my gosh, Apple's missing the AI boom.
[3:18] They're so slow. Their new Siri, the Apple intelligence,
[3:21] everything was so slow. And yet, if you look
[3:23] at the stock price of Apple, it has not declined
[3:26] over this. They have actually made some incredible...
[3:29] progress and it doesn't seem like the stock is
[3:32] declining. And one of the big reasons is because
[3:34] Apple, I don't know if it was accidentally or
[3:36] however you want to say it, but the hardware
[3:38] of Apple has been greatly used. And I think one
[3:42] of the big examples a lot of people will point
[3:43] to is when OpenClaw came out, basically everyone
[3:45] was running out to go buy a whole stack of Mac
[3:48] minis that they were running OpenClaw agents
[3:50] on. Now, I think that Mac mini craze was kind
[3:53] of short lived. It was kind of a quick moment.
[3:56] Obviously, the Mac mini is just a great. a great
[3:59] product. And so I think a lot of people would
[4:01] just use it as a as a good computer. But I think
[4:03] they Apple did see a very real, very large explosion
[4:07] in sales because of that. And they've realized,
[4:11] hey, look, I mean, you're basically a little
[4:13] server on someone's desk. Why don't we make actual
[4:16] servers for companies that are paying, you know,
[4:19] crazy amounts of money? Is this something that's
[4:20] growing up as scaling up as companies are looking
[4:22] to have their own virtual machines? They want
[4:25] their own agents in the cloud. I think Apple.
[4:27] caught the vision. They got probably a little
[4:29] taste of that. And so this is kind of the direction
[4:31] that they're moving towards. Again, this isn't
[4:33] going to happen until 2029. We know this because
[4:36] they said that this is going to be the, they're
[4:38] going to have four M8 Ultra chips. They just
[4:41] announced the M6 generation of chips in August.
[4:45] So the M8 line is at least two silicone cycles
[4:48] out. And so we know that the... announcing it
[4:51] on the M8, it's going to probably be 2029. These
[4:54] servers are also going to probably incorporate
[4:57] NVIDIA's NVLink Fusion. This is the same technology
[4:59] that NVIDIA uses to stitch multiple chips into
[5:02] one logical unit. So, you know, they have their
[5:06] chips, but they're going to probably be using
[5:07] NVIDIA to stitch them all together. This is interesting
[5:10] because... Apple is already kind of, they had
[5:13] a pretty rocky relationship with Nvidia for a
[5:15] while, but they've turned back to Nvidia and
[5:17] turned back to their chips to help power the
[5:19] revamped Siri servers. So their relationship
[5:21] has been on the rocks for a while. It looks like
[5:23] they're defying that and getting back into it.
[5:25] And this is going to be a big part of what powers
[5:27] their new server offering. We have some big news
[5:31] out of South Korea. SK Hynix is in talks with
[5:33] Intel to make memory chips in the United States.
[5:36] The South Korean memory giant would use Intel's
[5:39] US fabs for the first time. This is probably
[5:41] a big shift in where HBM is going for AI accelerators
[5:46] and where they're going to get built. The reason
[5:48] why this isn't interesting to me is because there's
[5:49] only three major companies that are building
[5:51] this specific type of memory. And by the way,
[5:54] SK Hynix is what's making the memory that powers
[5:56] a lot of NVIDIA's tech. So this is a major company.
[5:59] We have them, we have Samsung, and then we have
[6:02] one other company out of Idaho that is building
[6:05] these specific memory chips. And so this is really
[6:08] kind of a global bottleneck. But I think they're
[6:10] seeing that most of the demand is coming inside
[6:13] of the United States. Obviously, China is going
[6:15] to have a lot of demand, but there's problems,
[6:17] could be sanctions, or I'm not sure what South
[6:19] Korea is. situation is there. I think they see
[6:22] the demand coming out of the United States and
[6:23] they want to move closer to it. They're also
[6:25] making this, the interesting part is the relationship
[6:29] they have with Intel right now. Intel right now
[6:31] and Intel Foundry has basically been searching
[6:34] for an anchor customer to justify a massive US
[6:36] manufacturing build -out. They've been talking
[6:38] about this for a long time. We know that Intel
[6:40] stock was sliding and kind of, I don't want to
[6:43] say cratering, but Intel has struggled for a
[6:45] long time. It really should be one of the big
[6:47] companies we're talking about all the time. And
[6:49] we had the old CEO leave. We brought in a new
[6:51] CEO and Intel's been on kind of a generational
[6:53] run for the last little bit. They made a lot
[6:55] of partnerships with the United States government.
[6:57] And so it feels like Intel is kind of getting
[6:59] its mojo back and they're making a lot of moves.
[7:01] They recently announced a massive build out,
[7:04] US manufacturing build out. And with that, they
[7:07] need. customers. And so it looks like SK Hynix
[7:09] might be one of the big customers that's going
[7:12] to start building in their US fabrication facilities.
[7:15] This is all according to some reports by routers.
[7:17] This particular deal would be the first time
[7:20] SK Hynix produces memory in the United States.
[7:23] Traditionally, it's all been done in South Korea.
[7:25] SK Hynix is also basically the dominant supplier
[7:28] of high bandwidth memory to NVIDIA's AI accelerators.
[7:31] So it's... you know, just kind of speeding up
[7:34] the supply chain on a lot of that. We also have
[7:37] NVIDIA, Google and Emerald AI, which are launching
[7:40] a alliance for what they're calling grid flexible
[7:43] AI data centers. This is called the AI Energy
[7:46] Management Alliance. There's about 18 different
[7:48] companies that are inside of this. What's interesting
[7:50] with this, it was just launched this month and
[7:53] founders Emerald AI, Google and NVIDIA are all
[7:57] estimating that flexible data centers could unlock
[8:01] a hundred gigawatts in the existing US grid.
[8:04] So what they're saying is that essentially the
[8:06] United States. the US power grid is only running
[8:09] at about 50 % utilization on average. And I mean,
[8:12] you can imagine there's peak hours. And sometimes
[8:14] when that happens in California or Texas, you
[8:16] hear about like blackouts or brownouts or, you
[8:19] know, and there's a big storm and there are parts
[8:21] of the grid go down, like there's all sorts of
[8:22] issues. But generally speaking, on the average
[8:25] day, we're sitting at about 50 % utilization
[8:27] is probably like peak hours in the evening time
[8:30] when everyone gets home and is charging their
[8:32] electric cars or And, you know, in the middle
[8:34] of the day, when people are at work, it's lower
[8:36] in the middle of the night, it's off, right?
[8:38] So there's ups and downs in this, but it's on
[8:40] average about 50 % utilization. Their thesis
[8:43] on this, there's actually something called brattle
[8:45] group, and they're estimating that. each 10 %
[8:48] gain in grid utilization is going to lower electrical
[8:51] rates by about 3 .4%. So they're saying, look,
[8:54] we have the capacity to generate double the electricity.
[8:56] If we went from 50 % to 60%, everyone's electrical
[9:00] costs would drop 3 .4 % because there'd be more
[9:03] abundance in electricity. And so it would just,
[9:06] I mean, I think a lot of it gets down to like
[9:08] at peak usage times, you're not going to be paying
[9:11] peak prices. And so if there's just more electricity,
[9:15] there's more surplus. The peak, I guess, would
[9:17] be smaller. And so you'd save money. So that's
[9:20] their estimation. NVIDIA Digital Reality and
[9:23] Emerald AI are going to switch on. They're about
[9:25] to switch on about 100 megawatt power flexible
[9:28] AI factory in Virginia later this year. So there's
[9:30] obviously something they're looking at. Getting
[9:32] power for these data centers and for a lot of
[9:34] these projects is an absolute nightmare. It's
[9:36] very tricky. Apparently, US data centers currently
[9:39] wait up to 10 years or more for a grid connection
[9:42] under some of the existing interconnection processes.
[9:45] And this is why you had Elon Musk when he stood
[9:47] up his Memphis facility. He literally went and
[9:50] just, you know, purchased every... a portable
[9:53] diesel generator he could get and shipped it
[9:56] over to the facility. And instead of plugging
[9:58] into the grid, just powered the whole thing off
[10:00] of diesel generators, you know, much to the chagrin
[10:02] and a whole bunch of lawsuits about like pollution
[10:04] and that kind of stuff. But this is how he got
[10:06] the data center up and spinning. And lest you
[10:08] think it's kind of random, I mean, that data
[10:10] center in particular, he's building much bigger
[10:12] projects. But that particular data center is
[10:15] making a billion dollars a month, about 900 million,
[10:18] a billion dollars a month, roughly. renting their
[10:21] extra space, their extra compute to Anthropic.
[10:24] So Anthropic's paying them a billion dollars
[10:26] a month. The only reason they can make that much
[10:28] money, though, is because they kind of threw
[10:29] this together in a sort of not traditional way.
[10:34] And so a lot of these data centers aren't looking
[10:36] to go and rent every portable generator. I think
[10:39] SpaceX, or I guess it's SpaceX now, it was XAI
[10:42] at the time when they built this thing. I'm sure
[10:44] they've kind of... They built in some more sustainable,
[10:46] I'm hoping they've built in some more sustainable
[10:48] power generation methods. But at the end of the
[10:51] day, these things are very hard, very tricky.
[10:53] So the goal of this new alliance is to figure
[10:55] out how to plug into the grid and just increase
[10:58] the grid's utilization. The costs should come
[11:01] down for everyone, but they are also going to
[11:02] be able to plug in a lot of these big data center
[11:04] projects. It's a problem we're going to have
[11:05] to wrestle with. China is absolutely... out accelerating
[11:08] us to an insane degree when it comes to power
[11:10] generation. And if we want to power all of the
[11:13] projects that are currently under construction,
[11:15] we need to solve this. So this appears to be
[11:17] the solution that is being looked at by many
[11:20] people. Guys, thank you so much for tuning into
[11:22] the podcast today. If you enjoyed this episode,
[11:24] make sure to go check out AI box .ai. Leave a
[11:27] rating and review on the show wherever you get
[11:29] your podcasts. It helps the show a ton if you
[11:31] wouldn't mind. And make sure to go check out
[11:34] AIbox .ai if you want to get all of the different
[11:36] AI models plugged into ChatGPT or Cloud or wherever
[11:39] you do your work, cursor, et cetera. You can
[11:42] get 90 different AI models plugged into that
[11:44] so you can do audio, video, image generation
[11:46] all inside of the same workspace. Thanks so much
[11:50] for tuning into the podcast and I'll catch you
[11:51] guys all in the next episode.

Shows are listed from Apple Podcasts' charts and directory, and Podcast Index where it is used. Every show and episode belongs to its publisher and plays from the publisher's own feed; this site hosts none of them.