Micron Exits Consumer Memory to Focus on AI Chips | Tech Field Day News Rundown: December 10, 2025

Nvidia is gonna sell chips to China. Reactive hackers. IBM is buying confluent.

Intel's not selling their networking, bu after all IBM says, the data center boom is a bust. CloudFlare is gonna break the internet again, and we're gonna take a closer look at Micron because they're moving on from consumer Ram. And this week's episode of the Tech Field Day rundown.

Hello everyone. Welcome to the Tech Field Day rundown for December the 10th. And you know, something else that's important that's measured off in tenths, that's the Dewey Decimal system.

If you don't know what I'm talking about, you should watch UHF Con in the librarian will teach you all about the do we decimal system. Uh, but luckily when it comes to things that are decimals and measuring and stuff like that, I have a co-host who knows all about that because he measures everything in science units. Al, it's good to see you again, Always a pleasure to be here and particularly on National Lagger Day.

Uh, it is of course summer here in New Zealand, and it has just been rather hot. Uh, and so a nice cool lagger is always a great way to finish a hot day. Well, good luck with that because you know what else we've got that's hot is some news from this week.

I know you, you think to yourself, will the, the years winding down, how much more excitement can you pack into the last three weeks of the year? The answer, of course, is quite a bit. We're gonna start off with probably one of the bigger stories.

The Trump administration has reversed an earlier export limit and will now allow Nvidia to sell its H 200 AI chips to approved customers in China, but they are gonna keep the most advanced models from being exported. They're still banned. In exchange, the US government is going to take a hefty 25% cut of the chip.

Sales officials say that the move balances national security concerns with economic interests, but those pesky critics warn that it could boost China's AI capabilities despite ongoing concerns. Al, do you think that it's a good idea that Nvidia should be able to sell its old busted H two hundreds to China? Well, it seems that there's confusion here that national security being absolutely vital and staying ahead of China near Pier enemy, uh, at least potentially enemy, uh, is, is absolutely vital.

Yet, if you give the corporate treasury or the government treasury a little bit more money, it's no longer a concern. Um, seems a little weird. We've covered China's progress on AI previously and have particularly seen that, uh, Chinese chip foundries are building their own AI chips and although they're not quite up to the standard of what NVIDIA's producing progress is gonna be fast there.

So I'm not sure whether this is simply a case of the horses already bolted and allowing China to receive some of the older, uh, GPUs from Nvidia is no longer so much of a risk because they have their own organically created GPUs that maybe are, are gonna be at least as good. So we're, we're not necessarily letting them get any further ahead than they would've been. Of course, this doesn't cover the latest, uh, Blackwell and the, the upcoming Veru chips.

Uh, those are still not gonna be allowed to export to China. And there's also some interesting language around export to approved buyers within China. Uh, we know that the US government is somewhat skeptical of the technology companies in China that are aligned to the Chinese Communist Party and the Chinese military.

And so we wonder just how many approved buyers there will be in China for this new technology. Of course, we've also covered in the past the ways that, uh, non-approved buyers in China and also other sanctioned states have been acquiring this kind of, uh, advanced technology despite all of the best sanctions. So I'm not sure that walls around the export of these products are working very well.

So maybe just taking some money as these, uh, illegal exports or maybe, uh, restrictive exports are happening, make them legal and take some money. That's an argument we've seen before around all kinds of, uh, regulation. There's a critical vulnerability in React server, really common web framework, and it lets, uh, attackers run malicious code using a single unauthenticated HTDP request, the, uh, floors tagged as CVE 20 25 5 5 180 2.

And it comes from the unsafe deserialization and react server components and effect a whole lot of the Java frameworks like next JS and, and vitae, uh, pass on Redwood, uh, exploits a highly reliable, that's the last thing we want from next point. Uh, and proof of concept code is public making it very easy to get on board. Many apps are at risk given if, uh, even if they don't use React directly, because they use components that themselves have react in them.

Admins and developers are urged to update react, uh, related dependencies and maybe, uh, implement some further security controls to avoid this remote code execution. Uh, this seems pretty horrific to me, Tom. Uh, are you aware of React in your environments?

Uh, no, I'm not and mostly because I think 5 5 1 8 2 so personal, why don't we give it some cool name like React to Shell? Yeah, that's, that's what they're calling it. This is another example of a little problem that I like to call.

Well, how could they figure that part out? So you, you mentioned that there was a de serialization problem. Basically, the client server connection is allowing a specifically, um, crafted HTTP packet to, to hijack the execution logic because it's not secured and it's not looking for incorrect data structures.

Again, I go back to how, how, how did you figure this out? I know I just threw a whole bunch of stuff at your server until it broke, and that's what they did. 0, you've got a problem and you need to upgrade sooner rather than later.

10 out of 10 is not something to mess around with. And I, I've noticed that we've seen that a lot over the last year. 8 10 out of 10 because they are so easy to use.

This is not one of those things where, you know, it's, it's got a high score, but it's so difficult to like, you know, gain access to the data center and put a CD in the drive to hack it. No, no, no. This is pretty easy to pull off and it's almost 100% effective, effective on the affected versions.

You've got to upgrade and don't just assume, oh, well, I don't use React. Figure out what software you have does use React. You mentioned next Js Wu, uh, Redwood, S-A-S-D-K.

There's a whole list that is a huge mess that you're gonna have to sort out. You've got to go upgrade these things because if this is as easy as it looks in the proof of concept, you're probably gonna get owned before you know what's going on. So don't, don't walk, run.

And I know you're probably on change freeze December, but security patches are, uh, probably an exception to that rule. IBM is set to buy data streaming company confluent for $11 billion in cash. They're aiming to strengthen their AI and cloud services with Confluence real-time data technology.

This is the biggest deal that IBM has done in years, and they're expecting to close it sometime in 2026. The intention is to improve how businesses feed data into AI systems and analytics tools. This one hit in the morning, uh, this morning, and it was kind of interesting to see how it went from, well, there's a rumor that this could possibly happen to within a couple of hours.

It absolutely was gonna be happening. Al do you think that IBM making this big acquisition is a good move for them? I think it is.

I think one of the things we've seen with IBM is that they've recognized that getting into new markets, that transforming IBM to be relevant in the cloud and AI world is not just a matter of big science projects that they've done in the background, but it's also about finding innovation that's happened outside of IBM and acquiring that, that innovation. Uh, the last huge deal they did, of course, was acquiring Red Hat in 2019, which was very much a recognition that, uh, Linux systems and large scale deployment of Linux systems is vital to having any kind of cloud estate and that IBM needed some presence in that cloud estate. We're highlighting also that last year, uh, IBM acquired HashiCorp, who, uh, have a whole bunch of tools for helping developers and infrastructure professionals work at cloud scale on things that aren't necessarily restricted to, to just public cloud can also be on premises.

So we're seeing this as part of the, uh, approach that IBM has shifted to from let's build everything in-house, let's lead by being the most expert and, uh, understanding by building our own things to let's also acquire things from outside that are gonna be valuable to build our portfolio up. And I think it's a really good thing. Uh, I was interested that $11 billion was the price for Confluent and Confluent is, uh, essentially a managed services tool manage management tool for dealing with Apache.

Kafka itself is an open source, uh, and free software tool. Uh, it is probably one of the most widely deployed tools for dealing with large volumes of streaming data. And this is, as I play into ai, this is about data that's coming into your organization and large volumes, but with what I characterize as low value per unit of data and using AI to translate that into actually higher, uh, value data in, in smaller volumes.

So getting, handling large volumes of data, feeding it into your AI pipeline absolutely is part of how people are, how businesses are going to get value out of AI long term. Dealing with that huge volume of, uh, business data and logistics data and even operational data and trying to make sure that we can efficiently operate at large scale. This is absolutely where IBM should be playing.

So really glad to see this acquisition. Of course, it's great news for Confluent. Uh, the offer was at a roughly 34% premium on the stock price that Confluent had when this was announced.

Of course, confluent is now trading up 30%, so it's matching up. Uh, the market loves this as well because IBM stock prices is also going up despite the fact that they're about to spend 11 billion. Uh, yeah, seems like a really good move for IBM and should bolster the deployment of IBM software and solutions into cloud hungry customers, and of course, ai because we can't have a story without AI in it, Intel has decided maybe not to sell or shut down its networking and Edge Group, the NEX group after all, saying that the unit's important for integrating its chips, software and systems across AI data centers and Edge computing, uh, talks with Ericsson about taking a minority stake in, in next have ended, and the business unit will remain a part of Intel as the company reinforces under its new leadership.

Uh, the reversal comes as Intel's finances improve, and so it's requirement to sell off assets comes away of investments by Nvidia, the US government and SoftBank, uh, gives Intel a bit more space, a bit more breathing room to rebuild its strategy and focus on the core things that make Intel unique. Uh, seems like lots of up and up for Intel in the last couple of months, Tom. It has, and that all comes thanks to money.

It's amazing how that fixes almost all problems, right? This very much comes down to Intel looking to offload a business unit because they needed cash because they were not doing so well in the first half of 2025, and then round roundabout August or so, suddenly some people are interested in them. We've seen big investments from US government, from SoftBank, from Nvidia, from a lot of companies pouring billions of dollars into Intel, which has allowed them to renegotiate some deals, which has allowed them to say, maybe we need these pieces after all.

I get it. Originally the idea was let's get rid of anything that's not a chip making business and toss it out to see who will buy it. In this case, Ericsson really wanted this networking business unit because we need the cash to bring all these foundries and factories online to be able to continue business.

But again, with an infusion of a few billion dollars into your capital fund, turns out you can do a lot and not have to sell off everything that's bolted down. I, I applaud Intel for sticking around here because one of the things that this provides is not just networking. These components are manufactured and integrated into a lot of things.

We've talked, you know, over the year about how Intel has been forced to cancel products, has been forced to lay off thousands of workers. I think that the, the Tide has finally turned for Intel. I mean, I saw a report yesterday that there's a good possibility they might actually end up shipping one of those gaming gps that they've been trying to put out for a while that could give them bolster in the market to open up a new line of business that will then allow them to take in some money, maybe get those foundries online so that companies like Apple will start buying chips from them again.

And once they're back on firm footing, that really will give them an opportunity to shine. The question is, will it be enough to topple the rest of the companies in the market who are screaming ahead, selling things that are not boring, old gaming GPUs and edge networking gear, because we all know that everybody's hot for AI right now, and that's not an area where Intel's really focusing. I think what they're doing is they're providing an alternative in the country to manufacture those chips, to manufacture those devices that would potentially avoid tariffs and other things like that.

I'm glad Intel's holding onto it. I hope Aons not too disappointed that they're not gonna be able to buy it out. But look, folks, it's a good possibility that if this, if all of this goes sideways, you may get to pick it up cheaper than you were hoping.

We're gonna go back to IBM because CEO Arvan Krishna argues that the data center industry's massive AI plans really don't make economic sense. He estimates that it would take around $8 trillion to build the capacity that companies are currently promising. He says the odds of reaching artificial general intelligence soon are about 1%.

Data centers are gonna need to be rebuilt every five years, and the revenue that they generate can't cover the debt that you need to fund them. Krishna's warning echoes a growing consensus that the current AI build out might be a little bit over hyped and financially unrealistic Goals, uh, are there unless major technological or economic changes occur. Al this isn't the first time that we've heard somebody throw cold water on the current AI planning, but it's the first time that we've heard the CEO of a large company involved in the whole market tossing what it bounced to a pretty big pale of cold water on it.

Do you think Arvin Krishna is onto something here? Well, I think it, it lines up with what I've been seeing that a lot of the AI hype seems to be talking about a continuous exponential growth forever. And anybody who's looked at exponential growth knows that it can't go on forever.

Uh, sooner or later you end up with, uh, needing more resources than exist to support that, that growth and with the exponential growth that happens really fast. So this idea of, uh, needing $8 trillion to build out all of the data centers that are being committed to that doesn't actually seem unreasonable. The math kind of adds up that, that's the forecast.

So where is the, well presumably two to three times that much, 16 to $24 trillion worth of value coming from, right? Because if you're spending $8 trillion to build out the infrastructure, you'd be better be getting a multiple of that back in terms of value to business. And that's where I see the big challenge at the moment.

So simply going by this, we're gonna exponentially increase the amount of resources we use to build ever bigger and more complex things that just doesn't stack up for long. Uh, it it absolutely works at the beginning. What's likely to happen along the way?

Well, people are gonna run outta money, people building some of these infrastructures. Now, it's not gonna be the really big companies, the, the really large companies building up these infrastructures have have deep pockets. They don't run outta money fast.

It'll be the smaller, uh, I'm expecting to, to see some of the neo clouds struggle. Uh, maybe their exits will be to sell their, the capacity they've built to these larger companies that are promising to deliver these huge amounts of capacity. But I'm expecting to see some, some changes in there changes.

What I'm really hoping for is a fundamental technological change. That means we don't have to continuously go with this exponential growth building a model that is 10 or a hundred times as big as the last model that we built in doing this every 18 months. That's what's driving this exponential kind of thing.

My other concern about it is that these models are supposed to be built on human created content. And I think deep seek was the first model that we saw where the, the content that was being used to build the AI was generated by ai, by ai. And that seems like a sort of, uh, robber, a snake eating its own tail situation where you eventually run out of anything that's valuable.

You, you end up with a completely garbage model because it's being fed so much AI generated content, and we know AI generated content is currently not nearly as good as human generated content. So there's some, some interesting things about how this market's gonna play out. I continue to keep my fingers crossed for some technological shift that will get us away from exponential growth in order to get not exponential improvements.

There is definitely a challenge with the size of data centers, the amount of power those data centers are gonna require. And then that financial challenge of replacing them every, well Ivan said five years, maybe it's as quick as three years because that's the pace at which GPUs become outdated and need to be replaced to be cost effective to continue to use. Uh, no, whenever somebody tells you there isn't a bubble, it can't possibly be a bubble, uh, it's quite often is a bubble and maybe we'll see this, uh, come back down again.

My hope is we don't see a huge loss of business value. We see new technology that allows us to deliver more business value without more cost, often a requirement that we're looking for in business. Hey, you remember that?

React to Shell vulnerability, kindly titled CVE 25 20 25 5 5 1 8 2. Uh, well, it's being actively exploited by China linked groups and it's attacking all kinds of tools. I think this is going to be, uh, our next, um, log for JS Turkey hunt or, uh, Whack-a-Mole trying to find all of the places, uh, cloud CloudFlare forced a global outage to block these attacks because of course, you inspects the attacks and say we're having those.

Um, and AWS is warning organizations to patch immediately. Experts say rapid AI assisted weaponization of vulnerabilities is becoming the new norm. Just when you thought security stories could avoid this trap of ai, Tom, here it is for you again.

Yeah, you're probably thinking to yourself, didn't Tom just talk about this? Yeah, but I'm not talking about the vulnerability. I'm talking about the response to it because what happened, CloudFlare shut down to block those incoming connections because one of the things that you see a lot when you have one of these brand new zero days is you have a bunch of people that are trying to make as much hay about it as possible as quickly as they can.

And that's exactly what happened here. Everybody rushed out and tried to use it as an attack. Well, the downside of having CloudFlare as your, as your, uh, proxy layer, if you wanna call it that, is that when CloudFlare goes down like it did a few weeks ago, it can knock the whole internet offline.

The good news about having CloudFlare as your proxy layer is that when they detect spikes in traffic that are related to certain things like this, they can do something about it. Now, you probably are thinking to yourself, well, CloudFlare didn't go down. You're right.

It didn't go down for long because all they had to do was drop all of the current connections and then institute rules that prevented react cer sessions from flying all over the place using nofor packets. Packets pretty easy, huh? The AWS component of this story is even more interesting because, you know what I would do if I was AWS to limit my liability for all of these things, if I detect that you're running a vulnerable version of React, I deny incoming connections to your cloud until you patch.

That is the new area that we are gonna live in. We are going to be at the mercy of our providers because I don't think AWS or Google or Microsoft or Oracle or IBM or anybody who runs a cloud wants to have that much extra exploitation traffic running across their network. And so if they have an opportunity to create some momentary pain for their customers, for the good of them, then take your castor oil and be done with it.

Now, I, I don't know if, if Amazon's gonna go quite that far but far, I wouldn't doubt it. Maybe the next time this rolls around or sometime in the future that they're willing to say, Nope, enough is enough and we're gonna make you fix it this time. We'll have to see what happens.

But, you know, hope springs eternal. Right? We've got a story that we wanted to take a closer look at because quite honestly, I haven't gotten off of my soapbox yet.

Micron is leaving the consumer memory market that would be their crucial brand. Now, you're probably thinking to yourself, but Tom k crucial is one of the biggest memory sellers out there. Why are they doing that?

Oh, it's because they've decided to focus on high bandwidth memory for AI data centers. The company is gonna continue to sell consumer products until February of next year, 2026, probably because that's the stock that they have built up. And you may recall in the news over the last couple weeks that you've heard that there is a global RAM shortage.

High bandwidth memory sales are growing super fast, and that means that AI focused memory is more profitable than consumer products, hence Micron deciding that now is the time to exit stage. Right. Al, I'm gonna let you start off on this while I put an extra level on my soapbox.

Do you think that this is a good move for Micron? Good for micron? Undoubtedly they're doing this because they get more money per gigabyte for, well, more money per wafer producing high bandwidth memory than they do producing consumer ram.

Uh, bear in mind that the same foundry that makes the this ram also makes SSDs. Mm-hmm. Now, crucial SSDs also gonna disappear to make more high bandwidth memory.

Uh, it's absolutely good for, for, uh, micron, but for consumers. Yeah. I'm already hearing stories even in, in the Discord servers that I'm on that are completely unrelated to, to computers.

And I, hey, I'm hearing stories of people basically making decisions not to buy new computers 'cause RAM costs so much and they simply can't afford to buy enough RAM to make the upgrade worthwhile. Uh, this is something that we're seeing from essentially that massive demand for AI resources. It's taking all of the oxygen out of the industry and it's now affecting consumers as well as professional organizations.

Uh, when can we see an end to this, this tightness that we're seeing in, in the RAM market? Well, there need to be new foundries for building ram, or there needs to be a decrease in the demand for high bandwidth memory. Um, I don't see new foundries coming online fast.

They take quite a while to build. So Foundry capacity is definitely a, a limited resource. Uh, hopefully there's some of those foundries being built in the US that will come online not in the next couple of years.

Uh, so yeah, this, this, uh, shortage of server ram and then the flow-on will be SSDs is, is gonna be with us. It's going to have some impact on us as long as AI is soaking up all of the high bandwidth memory resources that are available at a high value. Um, Tom, have you got that second box on your soapbox?

So, uh, you can take a whole other tack on this. Yeah. This is the turning into a CrossFit soapbox right now.

Uh, I think this is actually a terrible decision, Byron, and they're not gonna know it for another two quarters, so that might as well be in the next century as far as they're concerned. Um, I think about it like this, uh, there's this influencer culture out there, right? Um, obviously we're influencers in the tech space and, and that's why we have this, uh, podcast.

But I want to think about the people who do like the influencer thing, where like they, they order a bunch of like cheap clothes from Amazon and then they try 'em on, right? You've seen these halls, right? Whether it's Temu or Sheen or Amazon or, or whomever.

And a lot of times what happens is when the influencer starts out, they're buying the affordable stuff, right? You know, this is a $3 top, these are $5 shoes and whatever, and they get popular because everybody wants to see the cool new looks or whatever. And then they start getting sponsored and then they start getting brand deals.

And the next thing you know, they're not opening $5 shoes, they're opening $500 bags, they're opening thousand dollars coats because the people who sent that to them want to show off for this big audience. Well, what happens when that audience suddenly thinks, well, these people are not speaking to me anymore. They're chasing the big dollars, they're chasing the clout.

I'm not gonna follow them anymore. This is actually a crisis that's going on in the influencer culture right now because of that, a very thing they've gotten too successful. They've, they've tried to get too much money from everybody, and in doing so, they've wrecked the audience that got them where they are.

Do you see the parallels in this story at all? I know you do Al I mean, I'm talking to the people at, uh, at Micron right now. Um, do you, do you see the parallels in that story?

Let me give you a hint. I need you to, to close the email chat that you have with your stakeholders for a minute. Your shareholders, they don't matter because in three months you won't matter because once you've sold out of all of the ram that you have in stock for your consumer markets, yes, nobody's building new PCs right now because they can't get ahold of the parts because there's no ram, there's no video cards.

Well, that, that's actually not true. There're starting to become a lot of video cards that are still in stock. They're just too damned expensive.

What's next? Like Al mentioned, maybe the hard drives are next. Who knows?

Maybe monitors are gonna go out, I don't know. But when nobody's building new PCs because they can't afford it, because they can't find the parts, that means that nobody is going to be putting money into your company anymore. Yeah, yeah.

You're gonna be living high off the fat of the land while you can build more HBM and sell it to Nvidia a MD. Qualcomm got news for you guys. Nvidia may be trying to buy as much HBM as they can right now, but do you know what happens when there are only one or two companies in a market?

They can dictate the price. Could you imagine what would happen if Nvidia pulled a Lee a Coca and walked up and said, I will buy Ram for you for a hundred dollars a stick. And you're like, no, HBM costs $500 a stick, and Nvidia says, no, it costs a hundred dollars a stick because that's all I'm going to buy it for.

In case you're curious, that's how Leh a Coka ended a, uh, UAW strike. He walked in, he said, I have a whole bunch of jobs for people that wanna earn this much money, but I don't have any jobs for people that wanna earn what you're asking. Nvidia can control your market.

Well, who are you gonna fall back on? Oh, that's right. You can't sell the consumers anymore 'cause you killed off that brand because you turn all of your printing presses into business focused printing presses.

Now the stakeholders and the shareholders are gonna tell you, this is a brilliant move because there's more margin in Ram right up until there's not. What's the old quote about going bankrupt? It happens slowly and then all at once.

Well, what's gonna happen to the demand for HBM? Well, it's gonna grow slowly and then fall off all at once when nobody has a market for these things anymore. And when nobody suddenly wants to have AI accelerators because they can't figure out what to do with them, then what are you gonna do?

Oh, hey, we're bringing the crucial brand back in three or four months when we get the things moved back over to making consumer Ram again. I hope there's still a market for PCs. I hope the Steam machine and the PlayStation five haven't wrecked everybody's desire to buy a machine now, because I promise you, the people who are doing work who are writing all that code for ai, they're buying laptops that already have RAM integrated into it.

They're not buying the Ram individually. You know what's gonna happen? All the people who are building these PCs to play games are gonna move on to do something else.

And then Crucial is not gonna have a market to go back into because the two or three companies that are left speaking to those hobbyists are gonna be selling it for whatever they can. And your big market, your high margin market is gonna go poof there. I've jumped onto my soapbox.

Well, it's always good to have you on your soap soapbox for a little while. What's also always good is having Tick Field Day events to look forward to. We're gonna take a little break from events over the holiday season.

We're gonna return at the end of January. I'm going to be hosting AI infrastructure Field day four in, uh, Silicon Valley, January 28th to 30th. It looking to be a pretty packed schedule as we've seen on the rundown.

AI gets everybody to the, uh, to the uh, uh, and then of course I'll be back for Cloud Field day 25 in March 11th and 12th. And there seems to be a month in between and, and it's February. And we have a couple of things that might just turn up on the Tech Field Day website in February.

Keep your eyes open for those. There might be, uh, familiar things coming back again. I think, uh, there should be some really fun stuff in February.

We'll, of course have a schedule of events right through 2026, uh, for us to have lots of fun at as, as the year progresses on lots of, uh, security events. I know there's more AI events to come. Uh, so thank you very much for joining us for this episode of the Tech Field Day Rundown.

It's awesome to have you with us this week. Rundown the news of this week. Of course, you can catch the new episodes every Wednesday as our YouTube video or in your favorite podcast application.

Make sure to give us a like and, uh, nice review as you are following us there. The rundown is also streamed on Textron TV because of course we are part of the Futurum Group, and you'll find Tom and myself as well as Stephen Foskett on various of the Futurum Group programs. We will be back next week on Wednesday to talk about the IT News of the week.

Well, in fact, we'll be back on Wednesday to talk about the news of the year. That was Tom and I will wrap up the year with a review of the biggest things in the year. That was until then for myself and for Tom Hollingsworth.

From all of us here at the Tech Field Aid team, we're wishing you and yours a great day and a great week. We'll talk to you next week.