When you hear of AI, it's fair to say that for many people their first idea is “ChatGpt.” Or perhaps Copilot, Google Gemini, confusion, you know, online chatbots that dominate the headlines. There is a good reason for this, but AI goes far beyond that.
Whether you're just trying to make your Copilot+ PC, edit video, transcription audio, run local LLM, or even better meet up with the next Microsoft team, there are plenty of use cases for AI that doesn't require these cloud-based tools.
So, there are five reasons why I thought about why I want to rely on online options to use local AI.
But first, the warning
Before I get into it, I need to emphasize the elephant in the room. Hardware. Without proper hardware, the unfortunate truth is that you cannot test and drive one of the latest local LLMs, for example.
The Copilot+ PC has many local features that you can use for your NPU. Not all AI needs an NPU, but the processing power must come from somewhere. Before you try to use local AI tools, make sure you are familiar with your requirements.
1. You don't have to be online
This is obvious. Local AI runs on a PC. ChatGpt and Copilot require certain web connections to work, even if the Copilot app is built into Windows 11.
Certainly, 2025's connectivity is better than ever before. You can also get Wi-Fi on a plane, but it's still not ubiquitous. Without a connection, you cannot use these tools. In contrast, you can use Openai GPT-OSS:20B LLM completely offline on your local machine. Certainly not always that fast, especially considering the GPT-5 has just been released. This model is based on the GPT-4, but can be used anytime, anywhere.
This applies to other tools such as image generation. You can perform stable diffusion offline on your PC, but you can guess online to get images from the online tools you need. Davinci Resolve Video Editor's AI tools run offline and take advantage of your local machine.
Therefore, local AI is completely portable and ultimately gives better ownership. And you are not at the mercy of the server's capacity and stability, limitations, or terms of service. It is also not mercy for these companies to change their models and lose access to older companies they may prefer. This is a current competition for many who have a switch to GPT-5.
2. Better privacy controls offline
This is an extension of the first point, but is important enough to emphasize in its own right. When you connect to online tools, you share data with large computers in the cloud. As we've seen recently, it's now reversed, but ChatGpt sessions were being cut down by Google search results under certain conditions.
If you are using online AI tools and use local to your machine, you simply can't control it. Local AI means that data will not leave the machine. This is especially important when processing sensitive or sensitive information, where security and privacy are of paramount importance.
For example, ChatGpt has Incognito mode, but data leaves the machine. Local AI takes it all offline. It is also much easier to comply with data sovereignty regulations or compliance with local data protection regulations.
However, if you end up redirecting LLM from your machine somewhere like an orama, and you end up sharing the changes you made, you need to remember. Similarly, if you enable web search on local models such as GPT-oss:20b or 120b, you'll lose a bit of total privacy.
3. Cost and environmental impact
Running a large amount of LLMS requires a huge amount of energy. It's just as true at home as you use ChatGpt, but it's easier to control both the cost and the environmental impact at home.
There's a free tier in ChatGpt, but that's not really free. Somewhere, a large server is handling the session, using huge amounts of power, and there is environmental costs. The energy use and environmental impact of AI will continue to be an ongoing problem to resolve.
In contrast, if you are running LLM locally, you control it. In an ideal world, there are houses with roofs full of solar panels and burying huge batteries. This helps to power a variety of PCs and gaming devices. I don't have it, but I could. It's just an example, but it makes the point.
The impact of cost is easy to visualize. The free layer of online AI tools is good, but you won't get the best. Why did Openai, Microsoft and Google pay for the stages that everything gives you more? The ChatGpt Pro costs $200 a month. This is $2,400 a year just to access the best tier. If you're accessing something like the Openai API, you'll pay based on usage.
In contrast, you can run LLM on your existing gaming PC. Just like me. Luckily, there is a rig that includes the RTX 5080 with 16GB of VRAM, which means that when you're not playing games you can use the same graphics card for AI with free open source LLM. If you have hardware, why not pay more to use it? !
4. Integrate LLMS and workflows
This is a Noob I code, so I'm just completely dabbing. However, using Ollama and open source LLM on my PC allows me to integrate these with Windows 11 VS code and have my own AI coding assistant. Everything is installed locally.
Also, this list has some crossovers with other points. Github Copilot has a free tier, but it's limited. You need to pay to do your best. And to use it or use the regular copilot, chatgpt or gemini, you need to be online. Running local LLM is more likely to implement it in your workflow, avoiding all this.
The same applies to non-chatbot-related AI tools. I was a Copilot+ critic that isn't yet enough to justify the hype, but the whole purpose is leveraging your PC to integrate AI into your daily workflow.
Also, local AI offers more freedom, accurately than what you use to suit your needs. For example, some LLMs are better than others. Using online chatbots uses the model they present, rather than being tweaked for a more specific purpose. However, using local LLM also gives you the ability to fine-tune the model yourself.
Ultimately, local tools allow you to build your own workflows for your specific needs.
5. education
I'm not talking about school-based education here, I'm talking about teaching myself new skills. From the comfort of your own hardware, you can learn much more about AI and how it works, and how it works for you.
ChatGpt has a “magic” about it. You can do all these amazing things by simply typing a few words into a box within your app or web browser. That's great, but there's no doubt about it, but a lot has to be said to learn more about how the underlying technology works. Required hardware and resources, building your own AI server, and fine-tuning open source LLM.
AI can do so much worse than setting up your own playground and learning more about it. Whether you're a hobbyist or an expert, using these tools locally gives you freedom to experiment. To do everything using your own data, without resorting to a single model, or without locking yourself into a single company's cloud or subscription.
Of course, there are drawbacks. Unless you have a huge hardware setup, performance is one. Run smaller LLMS easily locally and get excellent performance like Gemma, Llama, Mistral, and more. However, the biggest open source models, such as Openai's new GPT-Oss, 120B, can't work properly, even like the best gaming PCs of today.
Even the GPT-Oss: 20B slows down 20B than using ChatGPT on Openai's Mega-Servers (partly thanks to its inference functionality).
Also, you don't get all the latest and greatest models, such as the GPT-5, for use at home. There are exceptions like Llama 4. You can download this yourself, but it requires a lot of hardware to run until a small version is created. Older models also have cutoff dates for old knowledge.
But despite all this, there are many compelling reasons to rely on online alternatives to try local AI. Ultimately, if you have hardware that can do that, why not give it a try?
