What Laptop Do You Need for AI? Real Requirements for Online Tools and Local AI

You do not automatically need an expensive laptop to use artificial intelligence.

Many online AI tools perform the demanding processing on remote servers. Your laptop mainly runs the browser, displays the results and handles your files.

Running an AI model directly on your laptop is different. Your own hardware must provide the memory, processing power and compatible software required by that model.

Before buying an “AI laptop”, answer one question:

Will you use AI online, run models locally, or do both?

That answer matters more than an AI sticker, a processor’s marketing name or the presence of a dedicated graphics card.

This guide explains the basic requirements, the specifications worth paying for and when a refurbished laptop can be a practical choice.

Requirements checked: 5 November 2026, against the official documentation listed under Sources.

Quick answer: what specification should you look for?

For everyday online AI alongside study, office work and browsing, a laptop with 16GB RAM, an SSD and a supported operating system is a sensible buying target.

An existing 8GB laptop may already be adequate for lighter use. Try your intended tools before replacing it.

For local AI, there is no single specification that fits every model. Choose the application and model first, then check memory, hardware compatibility and performance.

The recommendations below are practical buying guidance. They are not universal software minimums or promises that every model will run.

1. Understand the difference between cloud AI and local AI

Cloud AI: the processing happens elsewhere

With a cloud-based AI service, your request is sent to the provider’s systems, where the model runs.

Your laptop needs to operate the interface comfortably, upload or download files and run your other applications.

For example, Google’s Gemini web-app documentation identifies supported browsers and account requirements. It does not require customers to own a dedicated AI graphics card simply to use the web service.

A suitable internet connection, a current browser and a supported operating system are the starting point. Particular features may also depend on the account, subscription, region or device.

Source: Google Gemini Apps Help: requirements to use the Gemini web app [1]

Local AI: your laptop performs the processing

Local software such as Ollama or LM Studio can run downloaded models on your computer.

Your laptop must store the model, load it into available memory and generate the response.

A small model answering a short question is a different workload from a larger model processing lengthy documents. Hardware suitability must be judged against the actual task.

Hybrid AI: both approaches are involved

Some applications combine local processing with cloud services.

Do not assume that installing an application means all its processing happens offline. Check which features use local models and which contact external services.

2. Is 8GB RAM enough to use AI?

For light online use, it can be.

If you mainly use an AI assistant in a browser, write documents and keep only a few applications open, an 8GB laptop may meet your needs.

The limitation often comes from the rest of your work. Browser tabs, video meetings, spreadsheets and background applications all use memory.

If the laptop becomes slow during ordinary multitasking, adding an online AI tab may make that limitation more noticeable.

For local AI, 8GB is restrictive. Some small models may run, but model choice, context length and available free memory become important.

LM Studio recommends at least 16GB RAM. Its documentation notes that 8GB Apple-silicon Macs may still work with smaller models and modest context sizes.

Source: LM Studio documentation: system requirements [2]

3. Should you choose 16GB or 32GB RAM?

16GB is a balanced starting point for online AI, office work, study and initial local-model experimentation.

32GB provides more working room for demanding multitasking, development environments and suitable local models.

However, capacity is only part of the decision. Check whether your software uses system RAM, dedicated GPU memory or unified memory.

A Windows laptop with 32GB RAM and a graphics card with 4GB VRAM does not have one interchangeable 36GB graphics-memory pool.

Also check whether memory can be upgraded. Some laptops have replaceable modules; others have soldered memory that cannot be increased after purchase.

4. Do you need a dedicated graphics card?

For ordinary online AI, usually not

When the model runs on the provider’s servers, a stronger laptop graphics card does not directly accelerate that server’s response.

A dedicated GPU may still be useful for other tasks, such as editing video, rendering 3D scenes or running local AI tools.

Buy it because your workload benefits from it, rather than because every AI task supposedly requires one.

For local AI, compatible graphics can help

A supported GPU can accelerate local inference. The important details include:

• Exact GPU model.
• Available dedicated memory.
• Supported drivers.
• The AI application’s acceleration backend.
• The model and settings.
• Laptop power and cooling limits.

Ollama publishes its supported-hardware information. Check that list for the actual GPU and operating system before buying.

Source: Ollama documentation: hardware support [4]

5. What is VRAM?

VRAM is memory associated with a dedicated graphics processor.

Local AI workloads may use it for model weights and other working data. If the workload does not fit, the software may use system memory, reduce performance or fail to run.

This is why a graphics card’s memory capacity matters alongside its processing capability.

An application’s baseline recommendation is not a guarantee for every model. LM Studio recommends at least 4GB dedicated VRAM for Windows, but that does not mean a 4GB card can comfortably run every available model.

A seller should identify the GPU and its VRAM separately. “Dedicated graphics” is not enough information for a local-AI purchase.

6. Why model download size is not the same as memory required

A model file’s size is only one part of its running requirements.

Additional memory can be needed for:

• The conversation or document context.
• Runtime buffers.
• Image or audio components.
• Multiple simultaneous requests.
• The operating system and other applications.

Quantisation reduces the precision used to store model weights and can reduce memory needs. The trade-off depends on the model and quantisation method.

Longer context and concurrent requests can increase memory use. Ollama documents these relationships in its FAQ.

A 5GB model download should therefore not be assumed to need only 5GB of free memory while running.

Source: Ollama documentation: FAQ [5]

7. How much SSD storage is useful?

For online AI, your storage needs mainly come from the operating system, applications and files you keep.

For local AI, downloaded models and generated media can consume substantial space.

A practical guide is:

• 256GB: suitable for lighter online work and limited local files.
• 512GB: a useful starting point for general work and local-model experiments.
• 1TB or more: worth considering for several models, development projects or media libraries.

Allow free space for updates, temporary files and new downloads.

An SSD improves loading and everyday responsiveness. It does not replace the RAM or VRAM needed during model execution.

8. What processor do you need?

There is no universal rule that AI requires a Core i7, Ryzen 7 or a particular premium chip.

For online tools, the processor must handle your browser and accompanying applications comfortably.

For local AI, performance depends on the processor architecture, memory bandwidth, supported instructions and whether work can be accelerated by a GPU.

Check the exact model and generation. “Core i7” describes a broad family spanning many years.

Software compatibility also matters. For example, LM Studio’s Windows requirements specify AVX2 support for x64 processors. Other applications may have different requirements.

Do not assume that a newer product name or a higher family number guarantees better performance for your chosen model.

9. Do you need an NPU or a Copilot+ PC?

An NPU is a processor designed to accelerate certain AI operations efficiently.

It can be useful when the application supports it. It does not automatically accelerate every AI program.

Microsoft’s Copilot+ PC category has specific requirements, including a 40+ TOPS-class NPU, 16GB RAM and 256GB SSD/UFS storage, alongside the applicable Windows requirements. Individual features may have additional conditions.

Source: Microsoft: Windows 11 specifications and Copilot+ PC requirements [6]

Keep these distinctions clear:

• Access to Microsoft Copilot does not prove a laptop is a Copilot+ PC.
• A Copilot keyboard key does not prove eligibility.
• An older Core i7 is not automatically a Copilot+ processor.
• An NPU is not compulsory for every online or local AI task.

Ask which exact features a seller means when describing a device as “AI-ready”.

10. Are MacBooks suitable for AI?

Compatible Apple-silicon MacBooks can be useful for online tools and supported local-AI applications.

Their unified-memory design allows supported processing components to share memory. However, macOS and other applications still need part of that capacity.

Do not assume all advertised memory will be available to the model.

Check:

• Exact chip.
• Unified-memory capacity.
• Supported macOS version.
• Application compatibility.
• Model format.
• Expected context size.
• Storage capacity.

LM Studio currently supports Apple-silicon Macs and states that Intel Macs are not supported by its current Mac application. Other runtimes have their own compatibility rules.

An Intel Mac’s inability to run one local application does not mean it cannot access browser-based AI services.

You can compare the model year, chip and memory of each MacBook in our Apple range.

11. What do AI image and video tools require?

Using an online image-generation service is different from generating images locally.

With an online service, the provider supplies the generation hardware. Your laptop still needs to manage uploads, downloads and any editing you perform afterwards.

Local image and video workloads vary according to:

• Model architecture.
• Output resolution.
• Batch size.
• Precision.
• Additional control models.
• Upscaling.
• Frame count and video duration.
• Memory-saving settings.

ComfyUI supports several hardware routes, including CPU execution, but its documentation notes that CPU execution is slower. Support varies by platform and installation method.

Source: ComfyUI documentation: system requirements [7]

There is no reliable universal claim that one VRAM capacity runs every image or video model.

Choose the model and workflow first, then check its official requirements and relevant tests.

12. Can a refurbished laptop be a good choice?

Yes, particularly for online AI alongside everyday productivity.

A refurbished laptop with sufficient memory, an SSD and a supported operating system may provide what you need without requiring a new premium device.

For local AI, verify the individual specification. A “workstation” badge does not guarantee enough GPU memory, and a “gaming laptop” label does not confirm compatibility with every AI runtime. If you are considering a workstation laptop, check the exact GPU and its VRAM on the individual listing.

Before buying, check:

• Full CPU model and generation.
• Installed RAM and upgrade options.
• Exact GPU and dedicated VRAM.
• SSD capacity.
• Operating-system support.
• Battery condition.
• Cooling and fan condition.
• Correct charger.
• Cosmetic condition.
• Item-specific warranty and returns.

A refurbished device should be assessed against the same workload requirements as a new one. Our guide to Grade A and Grade B refurbished laptops explains what cosmetic grades do and do not tell you, and our warranty policy explains how warranty claims work.

13. Can local AI work without internet?

Some local applications can operate offline after their required models and runtimes have been downloaded.

LM Studio documents offline operation for local model chat and document processing. Downloads, model discovery and updates require connectivity.

Source: LM Studio documentation: offline operation [3]

However, web search, cloud models, external tools and online storage can introduce network activity.

Check the complete workflow before assuming that information stays on the device. Local processing alone does not secure files against other users, malware or an exposed local server.

14. Using an AI model is different from training one

Running an existing model is called inference.

Fine-tuning adapts an existing model. Training a large model from scratch is a substantially more demanding task.

A laptop that runs a small chatbot should not be advertised as capable of training large AI models.

Students and developers can use ordinary laptops for programming, smaller experiments and connecting to remote compute.

For substantial training, investigate the actual framework, model and compute requirements before buying hardware.

15. Remember subscriptions, software and running costs

Laptop hardware is only one part of the cost.

Online AI services may involve subscriptions or usage charges. API-based tools can have separate billing from consumer subscriptions.

Local models may avoid some service charges, but still involve electricity, storage, setup time and possible hardware upgrades. Model licences may also impose conditions on use.

A powerful laptop does not automatically include paid AI software or unlimited access to a provider’s services.

Compare the total cost of your intended workflow.

16. A buying checklist you can actually use

Before purchasing, answer these questions:

1. Which AI tools or models will I use?
2. Do they run online, locally or both?
3. Is my operating system supported?
4. What does the software officially require?
5. How much system RAM will my other applications need?
6. Does the model need dedicated VRAM or supported unified memory?
7. Is the exact GPU supported?
8. How much storage will downloads and project files use?
9. Can memory or storage be upgraded?
10. Will I need paid software, subscriptions or API credits?
11. Can the seller demonstrate the intended workload?
12. What warranty, battery cover and returns apply?

For most people using online AI for writing, study, research and office work, a supported laptop with 16GB RAM and an SSD is a sensible starting point.

For local AI, buy against a named workload rather than an “AI-ready” label.

Frequently asked questions

Do I need a new laptop to use AI?

Not necessarily. Try your intended online tools on your current supported laptop first.

Is 8GB RAM enough?

It can be enough for light online use. It provides less room for multitasking and restricts local-model choices.

Is 16GB enough for local AI?

It can be sufficient for some smaller models and settings. It is not a guarantee for every model or application.

Do I need a gaming laptop?

No. Gaming hardware may help with supported local workloads, but ordinary online AI does not require a gaming laptop.

Will a better graphics card make online AI respond faster?

It will not directly accelerate a model running on a provider’s servers. Your internet connection, browser and overall system responsiveness can still affect the experience.

Can I use AI without an NPU?

Yes. Many online tools and local runtimes do not require an NPU. Specific built-in features may require one.

Is more RAM always better than a better GPU?

Neither is universally more important. The model must fit in the relevant memory, and the processing hardware must suit the runtime.

Can a refurbished laptop run AI?

Yes, provided its actual specification and supported software match the workload.

Is local AI always private?

Not automatically. Check whether the application uses cloud models, external tools, online services or shared storage, and secure the laptop itself.

Sources

Official documentation checked on 5 November 2026. Requirements can change, so check the current version before buying.

1. Google Gemini Apps Help: requirements to use the Gemini web app
2. LM Studio documentation: system requirements
3. LM Studio documentation: offline operation
4. Ollama documentation: hardware support
5. Ollama documentation: FAQ
6. Microsoft: Windows 11 specifications and Copilot+ PC requirements
7. ComfyUI documentation: system requirements

Find a laptop for the way you will use AI