AI PCs Will Become the “New Normal” and Capture 60% of Total PC Shipments This Year as Windows 12 Launches in 2H 2025, Analysts Say

[Tsing]

This dramatic shift will be driven by the convergence of several key factors: the proliferation of Neural Processing Units (NPUs) across different price tiers, the mainstream adoption of Large Language Models (LLMs) for everyday computing tasks, and the strategic launch of Microsoft's Windows 12 in 2H 2025.

See full article...

 

[MadMummy76]

Hahahahahaha, you serious?

 

[Brian_B]

I don't really doubt that this could happen.

I mean, it doesn't mean people will use it in the manner that we think - not everyone is going to be running LLMs on their laptops on a daily basis. But like Apple, I think it's at 100% across the board in it's products for AI accelerators, and as been for many years now. It's been used for biometrics and stuff for some time now.

 

[MadMummy76]

It's not whether it can have practical uses, but that the term AI has negative connotations. I doubt many people would willingly buy an "AI PC" unless it is forced upon them.

I can see it existing like physx, nobody giving any mind to it. I don't think of my PC as a physx PC, despite it having the capability to run physx.

 

[Peter_Brosdahl]

Well, I seem to remember that at some point by the end 2025, some consumers won't have a choice. I could be wrong but I thought we had a story a few months back stating laptops/notebooks/etc would all feature AI/NPU CPUs by (edited: forgot we're in 2025) this year's end. I have a feeling the same will happen with OEM desktops as well plus a good measure of PC enthusiasts pre-builts since marketing seems to love using AI with everything.

I think that after a time personal builds will end up being the last space where we see minimal AI inclusion and even then I'm not sure for how long.

 

[Burticus]

So never upgrade the CPU again and run linux? gotcha

 

[Uvilla]

Meh, what does this mean really, Cotana et al shoved everywhere and processors that run a bit faster as you do the same thing over and over, for sure I welcome the latter, dont know about the former, though AI can be useful.
Asking AI to order groceries and turn off the lights is a lot less of an issue than companies using it to review applications and such things, that needs to be outlawed soon, but it wont, and that will be an ever increasing problem for all of us.

 

[MadMummy76]

My fear is using my data to build their models. With crap like recall running on your PC I don't believe we are fully in control anymore of our data. Just like Adobe granted themselves full unrestricted rights to your data in their EULA, MS will do the same. They don't give a frack that it is illegal, they are daring you to sue them.

Ever since Windows 10, I don't feel like I'm the customer, more like me and my data is the product.

 

[Peter_Brosdahl]

and on that note, just saw this:

 

[Brian_B]

I think it's kind of funny - our IT dept at work won't let us use Gmail or DropBox or anything because they are worried about data confidentiality and security.

But they outsourced all our email to Microsoft and use the heck out of almost everything in the 365 suite. I have no idea if they allow/disallowed CoPilot AI stuff, haven't looked into it and I don't log into the online portal except when I have to.

 

[Burticus]

I think it's kind of funny - our IT dept at work won't let us use Gmail or DropBox or anything because they are worried about data confidentiality and security.

But they outsourced all our email to Microsoft and use the heck out of almost everything in the 365 suite. I have no idea if they allow/disallowed CoPilot AI stuff, haven't looked into it and I don't log into the online portal except when I have to.

Yeah after the umpteenth ransomeware attack, our corporate IT put the smack down on ANY person email / file services. Pretty sure they are all blocked on the firewalls, but that doesn't help when so many are working from home and not always on the vpn. Stupid salespeople.

 

[Grimlakin]

Oh man don't get me started on people using AI then trying to discuss proprietary things and get AI summaries.. GUYS.. STOP IT!!

 

[Brian_B]

Yeah after the umpteenth ransomeware attack, our corporate IT put the smack down on ANY person email / file services. Pretty sure they are all blocked on the firewalls, but that doesn't help when so many are working from home and not always on the vpn. Stupid salespeople.

I guess the one good thing about outsourcing it. When it does inevitably have a problem, you can always point the finger to something outside.

But that only works for so long, until everyone else starts pointing the finger at the guy that made the decision to outsource in the first place.

 

[Grimlakin]

I guess the one good thing about outsourcing it. When it does inevitably have a problem, you can always point the finger to something outside.

But that only works for so long, until everyone else starts pointing the finger at the guy that made the decision to outsource in the first place.

Yea after a huge incident where i work right after we outsourced most of our it. Our CIO 'retired'.

 

[Uvilla]

Yea after a huge incident where i work right after we outsourced most of our it. Our CIO retired.

When they outsourced, did they then opened the spigot of money? Im betting they did.

 

[Grimlakin]

When they outsourced, did they then opened the spigot of money? Im betting they did.

Rumors say yes.

 

[Marees]

MLVC: Multi-platform Learned Video Codec for Real-World Deployment​

Tanel Pärnamaa, Martin Lumiste, Ardi Loot, Evgenii Indenbom, Andrei Znobishchev, Ando Saabas

Neural video codecs have surpassed classical codecs in coding efficiency but remain impractical for deployment due to cross-platform incompatibility and high computational cost. Existing quantization-based solutions fail to produce deterministic results across diverse hardware platforms, leading to catastrophic decoding failures.

We introduce MLVC, a hardware-robust neural video codec designed for practical cross-platform inference. The key idea is to explicitly transmit scale parameters through the hyperprior, which guarantees entropy coding consistency across devices without requiring bit-exact arithmetic. While this increases bitrate overhead, we recover most of the coding efficiency through architectural improvements (gated memory, ReGLU activation), a long-term reference recovery mechanism, and domain-specific perceptual training.

On the VCD video conferencing benchmark, MLVC achieves >70% BD-rate (MOS) improvement over hardware HEVC, the strongest deployable baseline, while reaching subjective quality competitive with DCVC-RT, which cannot operate across diverse platforms.

Both the encoder and decoder run at 100 FPS on average on commodity NPUs from Apple, Intel, and Qualcomm.

MLVC is the first neural video codec to combine competitive compression performance, real-time speed, and cross-platform robustness across diverse consumer devices, making it suitable for widespread deployment. Code will be released.

MLVC: Multi-platform Learned Video Codec for Real-World Deployment​

GitHub - microsoft/mlvc: MLVC: Multi-platform Learned Video Codec for Real-World Deployment​

Announcing the Open-Source Release of ML Video Codec (MLVC) | Microsoft Community Hub​

Linux and Open Source Blog
Naba_Kumar Microsoft
Jul 24, 2026

MLVC is the product iteration of DCVC (Deep Contextual Video Compression) family of NVC (Neural Video Codec), open sourced by Microsoft Research since 2021, with improved compression efficiency, real-time performance on commodity Neural Processing Units (NPUs), and cross-platform support. We recently published this work in the paper MLVC: Multi-platform Learned Video Codec for Real-World Deployment. We are releasing the source code because we believe the next generation of video coding will be built openly, and we want the broader community — researchers, video codec engineers, platform vendors, product teams, as well as general developer community — to build it with us.

For example, for 360p video at 30 fps, where H.264 requires 1 Mbps, MLVC requires roughly 122 kbps for equivalent quality — about one-eight the bitrate under real-time conditions.


Beyond video compression efficiency, MLVC also offers:

• NPU-first design. MLVC is built to run almost entirely on the AI accelerators already shipping in modern devices — Apple Neural Engine, Qualcomm and Intel NPUs — at no more than 50% NPU utilization, leaving NPU headroom for the rest of the system.
• Real-time execution at the targeted operating points. Demonstrated 540p at 30 fps on Apple, Intel, and Qualcomm hardware.
• A scaling-law trajectory. Empirically, MLVC's coding efficiency improves with increased model capacity and additional training compute.
• Content-adaptive behavior out of the box, without hand-tuned Rate Distortion Optimization heuristics.

MLVC is more than just a research concept for Microsoft. We are currently rolling it out in Microsoft Teams, where it is being validated on real peer-to-peer video calls with active telemetry and A/B testing. The integration runs alongside fallback to conventional video codecs for hardware or reliability constraints — the kind of mixed-environment deployment that real products need. The scaling and reliability insights from this rollout are shaping the codec and its roadmap.

 

[MadMummy76]

This is so stupid on so many levels. First off, making up the details with AI is not equivalent quality. It is a fake image that we know can and will deviate from reality due to hallucinations.

Second, I'm pretty sure bandwith is far cheaper than the computation required to generate video with AI at 30 fps.

Welcome to the fake new future.

 

[Grimlakin]

This is so stupid on so many levels. First off, making up the details with AI is not equivalent quality. It is a fake image that we know can and will deviate from reality due to hallucinations.

Second, I'm pretty sure bandwith is far cheaper than the computation required to generate video with AI at 30 fps.

Welcome to the fake new future.

But the computation to process the video is happening on the NPU and GPU... you know what this means? When you go to task manager you'll see stuff happening on your NPU!! FOR THE FIRST TIME IN THREE YEARS since they have been included in CPU's!! Isn't that grand?!

Sorry... I'm on board... it's workload for the sake of workload. Sure it saves on bandwidth and that will be important to not interrupt grandmama and little timmothy having their iCall video call for grandmama's last birthday when it has to fail over to satellite and cost corporations money for satellite bandwidth.

It's BS is what it is... Maybe 2-3 years from now it will make more sense. But really the NPU's should have come this year... to serve a purpose not released 3 freaking years ago in laptops. Heck I don't think current desktop CPU's have them at all. Only the AI processors and stuff with integrated GPU will work this.

Sigh...

Like what happens on that call when say... I'm on the call with my GPU, My boss is on the call with this GPU, we're doing video codec transmission. Is it whatever codec to the host servers... then IT's splitting the stream to two streams to feed back to members in the meeting based on available processing hardware, effectively just adding MORE bandwidth use as you are probably delivering ALL streams to ALL members and letting their client decode the one it wants to.

I suspect it's the last example and Microsoft is just doubling down on bandwidth consumption for both codec's.

 

[garkot9]

For example, for 360p video at 30 fps, where H.264 requires 1 Mbps, MLVC requires roughly 122 kbps for equivalent quality — about one-eight the bitrate under real-time conditions.


Beyond video compression efficiency, MLVC also offers:

• NPU-first design. MLVC is built to run almost entirely on the AI accelerators already shipping in modern devices — Apple Neural Engine, Qualcomm and Intel NPUs — at no more than 50% NPU utilization, leaving NPU headroom for the rest of the system.
• Real-time execution at the targeted operating points. Demonstrated 540p at 30 fps on Apple, Intel, and Qualcomm hardware.
• A scaling-law trajectory. Empirically, MLVC's coding efficiency improves with increased model capacity and additional training compute.
• Content-adaptive behavior out of the box, without hand-tuned Rate Distortion Optimization heuristics.

MLVC is more than just a research concept for Microsoft. We are currently rolling it out in Microsoft Teams, where it is being validated on real peer-to-peer video calls with active telemetry and A/B testing. It’s great that the workload is handled by the built-in neural processor, rather than heating up the main CPU or the graphics card. Plus, there’s finally a genuinely useful application for all those trendy AI-focused units that are being packed into every laptop these days. Constantly reading through technical specs like this can make your brain start to overheat, so sometimes you just need a break. To unwind, I’ve been looking at an article on https://setapp.com/how-to/best-iphone-data-recovery-apps, which offers plenty of helpful advice on file security. And the fact that Microsoft is already testing this technology directly within Teams is fantastic it’ll immediately show just how smoothly the codec performs across different devices. The integration runs alongside fallback to conventional video codecs for hardware or reliability constraints — the kind of mixed-environment deployment that real products need. The scaling and reliability insights from this rollout are shaping the codec and its roadmap.

Do you think they’ll be able to adapt this algorithm for high resolution for standard TV series streaming in the near future?