Surface Laptop Ultra Tests Offline AI on Battery (Full Transcript)

A hands-on test shows local document AI, image creation and 1080p video generation, with nearly identical speeds on battery and AC power.
Download Transcript (DOCX)
Speakers
add Add new speaker

[00:00:00] Speaker 1: This laptop can run AI without an internet connection, even generate videos. But how well does it work on battery and who is it actually for? Microsoft just released the Surface Laptop Ultra. It's a premium laptop built for demanding AI and creative work. We'll try three things. We'll work with a document offline, generate an image and turn that image into a video. Then we'll generate that same video on battery and compare how long it takes. Microsoft provided this laptop for me to review and I get to keep it. Let's check it out. When you ask Chachapiti a question, servers in a data center do the work and send the answer back. With this laptop, we'll see if we could do some of that same work right here without sending it off to a data center. Now, why would you wanna do that? You can work without internet once everything is downloaded. You can use tools that keep your files on your computer and there are free tools that you could run without a monthly subscription. But your laptop has to be able to handle all of that work. Answering a question is one thing. Creating a video takes a lot more power. Is this laptop up to the task? It has a powerful NVIDIA GPU and 64 gigabytes of memory shared between the processor and graphics. That gives you room to run larger AI models and tackle tasks like generating images and videos directly on the laptop. Let's see how well it actually works. Let's start by chatting about a document. Here, I'm using LM Studio with Google's Gemma 4 E4B model. I've already downloaded everything onto this computer. Now, I'll turn off Wi-Fi so we could test this without an internet connection. I have a holiday cookie box launch plan for the Kevin Cookie Company. It includes things like the budget, who's responsible for what, and deadlines. Let's attach the PDF by dragging it into the chat. Then, I'll press Windows and H to dictate my prompt. Summarize the launch plan in three bullets. Then, make a table of action items, owners, and deadlines. Use only the document. If a deadline isn't specified, say so. Now, in the bottom right-hand corner, let's run it. That came back pretty quickly. Here, we have the three main points and we also have a table showing who needs to do what by when. Now, let's actually spot check one of them just to make sure it's accurate. Here, we have Maya and she's on point for finalizing recipes and the ingredient list. And here, it correctly shows that her deadline is October 22nd. And right down here, we have Jordan, who's on point for coming up with a local delivery plan. The original document had no deadline and here, it looks like it did not make one up. It just says that there's no deadline. I like that. Now, let's ask a follow-up question. Right down here, I'm going to ask it how much revenue would selling all 200 boxes generate and can you calculate net profit? Let's run that. There, it took about 15 or so seconds and at $24 per box, at 200 boxes, it looks like about $4,800 in revenue. But the document doesn't actually specify the labor and other operating costs. So, there isn't enough information to calculate a net profit and here, it calls that out. This is what I like about running AI locally. Even without internet, I can still get help with the document and I didn't have to upload it anywhere. Of course, you still need to check your answers, but I can keep working even when I'm disconnected. Now, you don't necessarily need a laptop this powerful to just summarize a document. So, let's give it something even more demanding. Now, let's create an image using Comfy UI with the Z-Image Turbo model running right on this laptop. If you're new to Comfy UI, I've linked my getting started tutorial in the description. I want to make a storefront for the Kevin Cookie company. Then, we'll use that image as a starting point for a video. Now, right here, I've already typed in my prompt. Feel free to pause if you want to read it. Now that I have my prompt, in the top right-hand corner, I'll click on Run. And right over here, it's now generated an image of the Kevin Cookie Company storefront. Now, one thing I like is if we look at the sign here, it's spelled Kevin Cookie Company correctly and that looks like a nice cookie store. For me, this could be useful for trying out a store design or maybe creating a visual for a presentation. I'm doing all of this locally. Once everything's downloaded, I can keep experimenting and I don't need to send each request to an online service. Getting this set up only took a few minutes and now let's take this storefront and let's turn it into a video. Now, let's bring our cookie shop to life. I'm using LTX 2.3 image to video template in Comfy UI. Now, right here, I've loaded the storefront image that we just created and to the right of that, I've described how I want it to move. There's my prompt. Now, right down here, I'm creating a five-second video that's 1920 by 1080 pixels. And right down here, I've set it to 25 frames per second. For this first run, the laptop is plugged in to its included charger and I have Windows set to best performance. Now, in the top right-hand corner, let's click on Run. You can hear the fans ramping up while it generates the video. The image came back really quickly. The video is taking longer, but again, we're asking it to generate five seconds of motion which is about 125 frames. And right over here, we have the result. Let's watch it. For me, this could be useful for turning a still concept into a short promotional clip. And all of that processing happened right here on the laptop. But what if I wanna do this away from my desk? Let's try unplugging the charger and we'll try this same job on the battery. Microsoft says this laptop is designed to maintain its performance on battery. Let's try this same video again. I'm using the same model. Here, we have the same image. We also have the same prompt, the resolution, the clip length, and the generation settings. And Windows is still set to best performance. Over here, I'll unplug the power and right down below, we can see that the battery is currently at 99%. Right up here, let's click on Run. And there it is. It completed the same exact video. And if I go down to the battery meter down here, we went from 99 to about 96%. Plugged in, the video took about two minutes and 31 seconds to generate. On battery, it took two minutes and 34 seconds, just three seconds longer. So in this test, unplugging the laptop barely changed how long I had to wait. I could still generate the video away from an outlet. This short test doesn't tell us how long it will last throughout a full workday, but for this particular job, it used a few percentages of the battery. So who is the Surface Laptop Ultra actually for? For me, the biggest benefit is having AI available wherever I'm working. I've come to rely on AI. When I have internet, I can use online services. When I'm disconnected though, I can use the tools I've downloaded onto my laptop. I won't have all the same models or features, but I can still get work done. We've worked with a document offline, generated an image, and created a video, even on battery. You don't need this much power to summarize a document, but it becomes a lot more useful for generating images and videos, demanding creative work in apps like DaVinci Resolve or development that needs NVIDIA's CUDA support. Before buying this laptop, check your software. This runs Windows on ARM. So some programs use a compatibility layer and certain plugins or accessories may not work. These results that we saw in this video are from my 64 gigabyte model and also the settings that I've shown. Now, my biggest hesitation with this laptop is the price. It starts at $2,600. This is an expensive laptop. If you mostly use email and office and online AI, it's probably overkill. But if you need demanding AI or creative tools on your own machine, and you also wanna be able to take that capability with you, this could make sense. Thanks for watching and I'll see you in the next video.

ai AI Insights
Arow Summary
The review tests Microsoft's Surface Laptop Ultra, a premium Windows on ARM laptop with an NVIDIA GPU and 64GB of shared memory, for local AI work without an internet connection. Using LM Studio and Google's Gemma model, it summarizes and analyzes a PDF accurately offline. With ComfyUI, it generates a storefront image and turns it into a five-second 1080p video. Video generation took about 2 minutes 31 seconds plugged in and 2 minutes 34 seconds on battery, dropping battery charge from 99% to 96%. The reviewer finds it compelling for mobile local AI, image/video generation, CUDA-dependent development, and demanding creative applications, but notes its $2,600 starting price and Windows on ARM compatibility concerns.
Arow Title
Surface Laptop Ultra: Offline AI and Battery Test
Arow Keywords
Surface Laptop Ultra Remove
local AI Remove
offline AI Remove
Windows on ARM Remove
NVIDIA GPU Remove
LM Studio Remove
Gemma Remove
ComfyUI Remove
image generation Remove
video generation Remove
battery performance Remove
CUDA Remove
creative work Remove
Arow Key Takeaways
  • The laptop can run downloaded AI tools locally, enabling document analysis without an internet connection or uploading files to a cloud service.
  • In the offline PDF test, the model summarized the document, extracted action items, and avoided inventing a missing deadline.
  • ComfyUI generated a storefront image locally, then produced a five-second 1920×1080, 25 fps image-to-video clip.
  • The video generation benchmark was nearly unchanged on battery: 2:31 plugged in versus 2:34 unplugged, with battery falling from 99% to 96%.
  • At a $2,600 starting price, it is aimed at users who need portable local AI, demanding creative workloads, or NVIDIA CUDA support—not routine office and web-based AI use.
  • Buyers should verify app, plugin, and accessory support because the device runs Windows on ARM.
Arow Sentiments
Neutral: The review is balanced: it highlights strong local AI and near-identical plugged-in versus battery video-generation performance, while emphasizing the high starting price and possible Windows on ARM software, plugin, and accessory compatibility limitations.
Arow Enter your query
{{ secondsToHumanTime(time) }}
Back
Forward
{{ Math.round(speed * 100) / 100 }}x
{{ secondsToHumanTime(duration) }}
close
New speaker
Add speaker
close
Edit speaker
Save changes
close
Share Transcript