AI PC Hype vs. Reality: What 2025 Holds in Store for the PC Market

Research Analyst Alek Maxfield and Senior Analyst Ben Carbonneau of TBR’s hardware research practice share insights into the performance forecasts and emerging solutions of leading semiconductor providers and OEMs.

Additionally, the pair discuss their predictions for the hardware market in 2025, including the possibility that the much-discussed AI PC refresh will be overshadowed by other drivers of enterprise spend.

Listen and learn with TBR Talks!

Submit your Key Intelligence Questions for Patrick and his guests

Connect with Patrick on LinkedIn

Connect with Ben on LinkedIn

Connect with Alek on LinkedIn

Learn more about TBR at ⁠⁠⁠⁠⁠https://tbri.com/⁠⁠⁠⁠

TBR Talks is produced by Technology Business Research, Inc.

Edited by Haley Demers

Music by Burty Sounds via Pixabay

Art by Amanda Hamilton Sy

AI PC Hype vs. Reality: What 2025 Holds in Store for the PC Market

TBR Talks Host Patrick Heffernan: Welcome to TBR Talks: Decoding Strategies and Ecosystems of the Globe’s Top Tech Firms, where we talk business model disruption in the broad technology ecosystem from management consultancies to systems integrators, hyperscalers to independent software vendors, telecom operators to network and infrastructure vendors and chip manufacturers to value added resellers. We’ll be answering some of the key intelligence questions we’ve heard from executives and business unit leaders among the leading professional IT services and telecom vendors.

I’m Patrick Heffernan, Principal Analyst, and today we’ll be talking about 2025 predictions for AI PCs with Ben Carbonneau, Senior Analyst for TBR’s IT Infrastructure practice, and Alek Maxfield, Research Analyst for TBR’s Devices practice.

Analyst’s coverage

All right, Ben, Alek, gentlemen, thank you very much for coming in today. Being on this podcast, I thought we might start with an overview of your portfolios. I’ve done this podcast now, this is, I think season three. I know my portfolio TBR really well. I know the companies that my practice covers, but I’m not entirely sure I know the companies that you guys cover. I probably know them, but I don’t know which ones fall into which practice. So, if you could maybe just walk through, who do you cover? Alek, who do you cover, Ben? And then we’ll jump into some of the questions I have about your predictions for this space going forward. 

Ben Carbonneau, TBR Senior Analyst: Sure. I can lead off this question. So, our Devices and our IT infrastructure practices are kind of organized under one broader practice. I think of it more as a hardware practice, where my coverage is mostly IT infrastructure focused, covering companies like Dell Technologies, HPE, Lenovo, Super Micro, companies like those infrastructure providers of compute for the data center, storage and networking. And then Alek also covers jointly Dell and Lenovo from a device perspective as well as some other companies. So, Alek if you want to elaborate on that.

Alek Maxfield, TBR Research Analyst: Yeah, exactly. So, our overlap is that Dell and Lenovo where they’ve got the IT infrastructure side as well as the devices side, and then the ones where I’m sort of, acting as the primary author, would be Apple, as well as HP Inc. 

Patrick: Okay. Yeah. And so, that’s interesting that you’ve got that overlap because we have something similar in other practices where we look at SAP very closely from the services side, but of course we don’t cover SAP the way that the Software practice does. So, all in all, it’s, I think that my experience of being here in TBR longer than you two, certainly, is that we’re increasingly doing more sort of cross practice stuff, breaking down a lot of the silos.

AI PC nomenclature

So, I want to get into the predictions document that you released at the end of last year, looking into 2025 and a couple of things that jumped out at me. One was you stated pretty clearly that- and what we’re going to talk about primarily are AI PC. So artificial intelligence, do you, do you call it like artificial intelligence enabled PCs or embedded PCs, or what’s sort of the right way to think about it?

Alek: So, that’s something that has come up a lot over the last year since December 2023 when Intel put out their first sort of line up with the chips that were kind of creating this new AI PCusbsegment. So, AI PC is a little bit of a loose definition, I think, the loosest possible definition. And then you can specify within it, but the loose as possible definition is a PC with that neural processing unit. So that’s sort of that heterogenous processing aspect where there’s not just the CPU and the GPU, there’s an additional- the NPU specifically for those AI tasks. 

Patrick: Okay.

Alek: So, that’s kind of where the AI PC thing comes in. 

Ben: And I would only just add to that. I think you hit the nail on the head with the inclusion of the neural processing unit on those AI PC chips. So as PC silicon has evolved kind of first it was mostly CPU centric. But now at the heart of almost all PCs are what we call a system on a chip. And that integrates multiple different processing units onto one piece of silicon die and that would be a CPU, a GPU, and a NPU. So that is kind of what we’re thinking of at TBR, when we define AI PC, it’s something that features one of these AI PC SoCs. So, I think another important distinction to make is that AI workloads aren’t new, obviously, with the release of ChatGPT, you know, we hear AI more and more and marketing language. And I think, you know, a lot of the OEM is would like you to think that this is a brand new category, but really AI enabled workloads have been possible on PCs, leveraging GPUs, for gaming or professional visualization workloads from the likes of Nvidia for several, several years. And I think now a really good metric that I like to look at is, you know, while the NPU does enable some of these lighter workloads, and I think really the value proposition behind the NPU is energy efficiency. So, this handles things like video blurring and audio suppression, some of the heavier workloads like model fine tuning, those are going to be taking place on Nvidia GPUs. And a good, little number here is if you’re kind of a computer nerd like Alek and I, we follow TOPS, which is trillions of operations per second. And it’s an AI metric for performance. I think to be considered in Copilot+ PC, which a lot of these new AI pieces are. The NPU has to be able to do 40 TOPS performance alone. And that is in contrast to, I think Nvidia’s newest RTX GeForce 4090 GPU for PCs, which does 1321 TOPS. So it’s not really comparable in the sense of total workload performance, but really just kind of what the NPU is designed to do, which is handle those more pervasive background AI tasks.

Patrick: And in the broad scene of technology, it kind of makes sense that you would have, I mean, every single smaller piece of the PC or the tablet or the whatever, even the phone is always going to be a smaller amount of compute power than the mainframe, so that totally makes sense. But that is a huge difference between 40 and what’d you say 1300? 

Ben: 1300. Yeah. 

Expectation recalibration

Patrick: Yeah. So now that we know what we’re talking about with respect to AI PC’s and I like that the definition can be a little fuzzy. You stated though, in your predictions document, that the AI PC OEMs misjudged the market, thinking that new AI PCs would drive a refresh cycle. Do you think those OEMs have sort of fully recalibrated their expectations now? Do you think they see what you’re seeing, which is that slow ramp up to adoption that’s dependent on other ecosystems? Or are they still do you think they’re still betting on that hard charging refresh cycle? Where are we right now.

Alek: Yeah, I think they’ve definitely tempered their expectations over time because I think when we were first seeing AI PC kind of emerging as that new product category, the proliferation of the AI PC was sort of at the forefront of what was going to be driving the next refresh cycle. And they’re definitely still mentioning it, and they definitely see it as one of the factors. But I think right now it’s sort of the least of three important factors, which are, huge aging install base where those commercial PCs are going to have to be replaced regardless of anything else. The end of Windows 10 support, which is going to be- it’s currently set for October 2025. And then the last aspect is the kind of leaps that they’re making in the technology in AI PC and how they’re sort of creating experiences to drive adoption of the AI PC. So, it’s definitely still one of the factors, but I think it’s sort of been- they pulled back a little bit from it being THE factor.

Business model changes?

Patrick: So, it’s obsolescence and obsolescence are driving the first two. And then really it’s the new stuff is going to latch on to that. Yeah, that makes sense. And so, when I think about them having to recalibrate does that- how much- does it change their business model at all in terms of how they’re investing. Or is it sort of we were going to make those investments in bringing out this new technology anyway, investments both in the technology itself, but then also in like the go to market and the, you know, the sort of selling it. Has that changed at all or are they sort of continuing down that same path?

Alek: I think, that was something that was actually really interesting. I think I kind of get what you’re getting at where, I think Dell was one of the companies that was really interesting to look at last quarter because, just recently in January, they had sort of a refresh of their whole portfolio where it’s AI PC focused, but they were conspicuous silent on AI PCs in 3Q24. Where they were kind of, I think they were modifying their sort of go to market timing a little bit because they see the refresh cycle and AI PC adoption as being pushed forward. They were kind of trying to capitalize a little bit later, kind of seeing the writing on the wall in terms of that sort of cycle being pushed out.

Co-opetition

Patrick: All right, that makes sense. And that actually explains why in reading the predictions document, you highlighted not Dell, but actually HP and Lenovo more. And one of the questions I had on that was you talked about new offerings and you said these companies I’m going to quote here, “will continue to position these agents as complements to Microsoft’s Copilot+ rather than replacements. As they will shy away from both attempting to compete with Copilot+ and other cloud-based offerings, and from alienating Microsoft, a vital partner when it comes to AI.” So, when I read that, it strikes me as such a clear example of co-opetition vendors having to compete and cooperate at the same time. Microsoft has done that its entire life of Microsoft. That’s who they are, but HP and Lenovo may be not as well versed at or well-practiced at that co-opetition. So, are they going to be able to manage? Because doing that requires managing your alliances, your ecosystem, your go to market, all those challenges around how you position yourself in co-opetition. So, when you think about HP and Lenovo, are those companies ready for that kind of environment? Are they going to be able to succeed in that kind of environment? 

Alek: I think so. I think we kind of see it as sort of an extension of what they’ve done already with the Windows side of things, where we have the PC OEMs and they obviously have to work in tandem with Windows, where, for example, there will be things like Windows Defender, is the Windows proprietary security features.

Patrick: Right. 

Alek: And then, you know, the OEMs have their own security sets that sort of work alongside that. So, I think it’s going to be kind of an expansion of that.

Patrick: Right, that’s interesting. So, and that’s an area where they have been doing that forever and they’re not new to the game of doing cybersecurity, they’ve been for a long time embedding and enabling cybersecurity alongside Microsoft. But at the same time they’re competing with them. So, Dell in particular, I think of as having a really robust set of cybersecurity offerings. 

Alek: Yeah. Exactly. 

Advisory offerings

Patrick: Okay. All right. Yeah. That makes a ton of sense. So, you also mentioned in the predictions, I’m going to quote again, “OEMs will always focus on the development and delivery of advisory services that help organizations identify how they can maximize their ROI when purchasing AI PCs.” I have to admit, I read the whole thing. That’s the first thing that jumped out at me. I know this is the third question and the second time I’m quoting, but that word advisory for me just sets off alarms because maximizing ROI when buying AI PCs, that can mean so many things. It can mean business model changes, efficiency through organizational changes, things that consultancies typically do. It can also mean staying very much within the lanes of what is the value you’re going to get, or how can you optimize the value in this particular purchase or in this particular set of purchases? Some OEMs are very good at staying in that lane. Other OEMs, in my experience, have a tendency to want to wander into the business side of things and into the business consulting side of things. So, when you think about the companies you cover and this very much goes to both of you, do you anticipate the advisory side of their offerings expanding as they expand how much they’re selling AI PC’s, how much their portfolio expands? Does the advisory part expand with it? And I’m highly biased when I ask that question because I think it shouldn’t. I think the companies that do well are the ones that stay in their own lane. But if there’s the opportunity there and they would be leaving money on the table for a consultancy, maybe they’re not going to do that. So where do you come down on how this is going to all play out? 

Alek: I think really kind of your last point was sort of what my thinking was on it, where they don’t have any interest in leaving the money on the table in these situations. And also, there’s been this huge push, you know, over the past several quarters, and it’s just going to continue over the next, you know, several years where companies are really focused on increasing their services businesses. You know, just because hardware is a tough market.

Patrick: It is.

Alek: PC hardware is a really tough market. So, they’re definitely going to be trying to I think, increase their own services capabilities to sort of push- 

Patrick: Services is a tough market too, 

Alek: That’s true. That’s true. 

Patrick: It’s dependent so much on people and permission and hardware companies traditionally have gotten permission through the things that people actually- the physical things that you’re buying, and people has not been the top priority for hardware companies.

Ben: Sure. Yeah. So from my perspective, I see kind of more of that advisory piece as it relates to these AI deals and engagements more being kind of like, if I think of it and I’m the OEM, I think of it more in my cost of sales rather than something that I’m explicitly offering to gain revenue. I think it’s just a way- I think especially for PCs like Alek was alluding to, obviously a commoditized market. You know, PC OWMs are hoping that maybe it won’t be the case of the AI PC. I think it will continue to be a commoditized market. So, they need to offer these services to bolster their margins and whether they’re really gaining any extra revenue dollars from these advisory services, I think is less to the point. And more of the point is just kind of increasing the depth of the engagement with the customer to sell those things that they’re good at, which are like support services. 

Patrick: Right? 

Ben: You know, not those advisory pieces, but I think almost and I kind of tie it back to, you know, what HP and Lenovo are offering with HP Companion, and Lenovo AI Now, those two AI agents. It’s really kind of having these complementary adjacent offerings in their portfolio to be able to try to differentiate themselves from the other competitor. When really, you know, who knows? Do they really think these are going to be big consumption drivers? I don’t think that they’re going to be super differentiated like advisory services. Again, you know, these are hardware OEMs. They’re not skilled in the people business. Right.

Patrick: Right.

Ben: So, I think they understand that. But by having these offerings and being able to talk about them in a discussion, I think it just helps their sales motion with a deeper engagement, whether the services wind up getting delivered, or the AI agents wind up actually being used over Copilot. 

Patrick: Yeah. I’m glad you said that, because in the predictions document, you talk about how it’s going to be very hard to differentiate. And services might be an area where they can do that. And I read that and I was skeptical for two reasons. One, differentiation is almost always impossible. It really is sort of a holy grail and sometimes a holy grail that is too much effort and is wasted trying to reach it. But then I was skeptical that services could actually be where a hardware OEM could come and do that. But I think I see something here where they’re not doing it for the revenue so much as they’re doing for the retention. They want to hold on to those clients. They want people to get locked into buying their equipment, their stuff. And actually, that is something that services companies, that’s the most important metric of their own, you know, how they’re evaluating their own success is client retention. And so, if they can latch onto that as part of what their services do for them, then they actually will- the differentiation will come simply through the “we hold on to our clients.” It’s going to be an interesting space to watch. 

Changes from GenAI

So, I got to ask, you know, the sort of this time of year- all the questions around generative AI and thinking about what’s coming. And I’m curious, and this is a not an easy question, I understand, but put yourself in December of this year, December of 2025, what are you going to look back and say, this is the development that I expected to happen, and I’m glad it did, or this is what came out of nowhere and I’m surprised by it. Or this is the way that we’re talking about generative AI now that was different from where we were at the beginning of this year in January. Like, what’s the biggest change? I guess in a long way of saying and giving you time to think, what’s the biggest change that’s going to come because of generative AI in your particular space? Or maybe more broadly across the ecosystem?

Ben: Personally, for me you know, with respect to devices specifically, I think it’ll be changes in the adoption of some of these AI agents. So, I know that, for instance, Copilot+, one of the main features that Microsoft is touting is recall, and the ongoing joke kind of in the devices space is that recall keeps being recalled. So, I think it went out just a few months ago to some of the Windows Insiders. I think it’ll be interesting to see adoption of kind of some of these more almost seemingly invasive technologies like Microsoft Recall and how that will be received by the public. And then how that might drive adoption. I think another thing, with respect to AI PCs is we always hear that refresh, you know, that third factor, like Alek was saying, AI PCs being a driver of refresh is really dependent on killer apps being developed. 

Patrick: Yeah. 

Ben: So, you know, there’s a lot of investment going toward ISVs, whether that’s from some of the silicon vendors or from the PC OEMs and developing applications that leverage the NPU. But personally, it’s my take that I don’t think there will be any definitive killer applications for the AI PC. And I think this just ties back to that 40+ TOPS versus 1321 TOPS thing we’re talking about at the top.

Patrick: Right, right.

Ben: So, I think the NPU’s value is really in extending battery life as the chip can be more efficient for certain workloads than the GPU. But enabling a net new killer application, I personally don’t see that happening. 

Patrick: Okay.

Ben: So, in December, I’ll be eager to see if I’m eating my words, or if I was-

Patrick: It’s sometimes a harder prediction to make that something’s not going to happen. So. Yeah, Alek anything from you on this one?

Alek: Maybe not this December, but next December. I think something that is interesting to me is like, how much of the increasing AI PC adoption is going to be because people are going out of their way to buy AI PCs, because they want the AI PC specific features, or how much of it is just going to be the consequence of, “well, I need to refresh my PC anyway because I’ve had it for four years and all the PCs in this price bracket or whatever that I’m looking for, just have an NPU” regardless of whether you really want it or need it or not. And how much- because the whole thing is, you know, we’re going to have however many percent of the total PC shipments that go out are going to be AI PCs by this time. And it’s like, maybe this is something that you can’t really know, and they’re not really- the OEMs aren’t going to say this, but I think that there is sort of that dynamic that I’ll be watching where it’s, you know, how much is it really that they’re sort of driving this great interest in the AI PC and how much of it is a consequence of they’re just stuffing their portfolios with AI PCs?

Patrick: It’s just baked in. Right, right. Yeah. 

Energy considerations

How about one last question. Energy. You touched on briefly in the piece and then, you know, I’ve got an interest in it. Because there is that concern, that huge concern in our broader generative AI predictions piece, we talked about energy quite a bit and the concern that the demand for energy around generative AI. And with an AI PC, is it alleviated because you’re only at 40 TOPS. You’re not at 1300 TOPS? But I really honestly don’t know. So, what is the long-term play around energy with respect to AI PCs? Do they take more energy? Do they spin faster, therefore they require more. 

Ben: To me, I think, that runs in parallel with kind of what I see in the infrastructure market. If I just think of the PC as an edge device. I think, you know, what we’ve seen over the past several years, there’s more and more data being collected and processed by organizations, and more of more of that is happening at the edge. I think there’s always that, you know, go to the cloud or repatriate from the cloud. That’s always going to be, kind of a cyclical thing based on costs. But I think that as more AI processing happens at the edge, there’s definitely, a possibility that it does save that energy with less being sent up to the cloud, processed in the cloud, and being sent back to back to the core data center. 

Patrick: Right

Ben: So, less data transmissions and more AI processing at the edge, I think does have- you know, is possibly something that will conserve energy in the long run. 

Final thoughts

Patrick: Yeah. Well, I hope you’re right, because I think that’s one of my biggest concerns. So excellent. Gentlemen, this was so much fun. Really enjoyed it. And, we’ll have another chat if not in December- or if not before that, certainly in December. So we can find out whether we were right or wrong. Thank you guys.

Alek: Awesome. Thank you.

Patrick: Next week, I’ll be speaking with TBR Principal Analyst Boz Hristov and Senior Analyst Kelly Lesiczka about the potential labor pyramid collapse. Don’t forget to send us your key intelligence questions on business strategy, ecosystems, and management consulting through the form in the show notes below. Visit tbri.com to learn how we help tech companies large and small answer these questions with the research, data, and analysis my guests bring to this conversation every week. 

TBR Talks: Decoding Strategies and Ecosystems of the Globe’s Top Tech Firms

Join TBR Principal Analyst Patrick Heffernan weekly for conversations on disruptions in the broader technology ecosystem and answers to key intelligence questions TBR analysts hear from executives and business unit leaders among top IT professional services firms, IT vendors, and telecom vendors and operators.

“TBR Talks” is available on all major podcast platforms. Subscribe today!