Nvidia Paid $6B to Make AI Free — Here’s How to Run It ($0)

Nvidia, reports say, spent around $6 billion to acquire the AI startup Poolside — and the goal isn’t a closed product you rent by the token. According to reporting (WSJ, 2026), the aim is to build open-weight Nemotron models to rival open releases like DeepSeek and Kimi. “Open weights” is the phrase that matters, and it’s why a multi-billion-dollar bet lands directly on your laptop, for free. Let me show you what that actually means and how to run it today.

What “open weights” actually gives you

A normal AI product — the kind you chat with through a website — lives on someone else’s servers. You send text in, you get text back, and you pay per request. The model itself is a black box you never touch.

Open weights flip that. The company publishes the trained model file — the “weights” — and you download it and run it yourself, on your own machine. No API key. No per-request bill. No sending your prompts to anyone’s server. That’s the whole point of Nvidia’s reported Nemotron strategy: put capable models in everyone’s hands instead of behind a paywall. And you don’t have to wait for the finished mega-model to feel it — you can run an open Nemotron model right now.

Run it in three commands

The easiest on-ramp is Ollama — a free tool that downloads and runs open-weight models locally with almost no setup.

1. Install Ollama from ollama.com (macOS, Windows, and Linux — one installer).

2. Pull the model:

ollama pull nemotron-mini

This downloads the weights once to your machine. After that, it’s yours — offline, forever, no account.

3. Run it:

ollama run nemotron-mini "Explain what a REST API is in two sentences."

The model answers on your laptop. Turn off your Wi-Fi and it still works. Ask it a hundred more questions and the bill is exactly $0.

The honest limits

I’d be lying if I told you a free laptop model matches the frontier chatbots. Here’s the real picture:

• nemotron-mini is small. It’s roughly a 4-billion-parameter model — great for quick tasks: summarizing, drafting, explaining a concept, simple code. It is not frontier-level reasoning, and it will trip on hard, multi-step problems. Use it for what it’s good at.
• The big one isn’t out yet. The headline trillion-parameter Nemotron that’s meant to rival the top open models is, per the reporting, still in development. What you can run today is the small, already-released family — a real taste of the strategy, not the finished flagship.
• Open weights ≠ frontier for free. Downloading and running a model yourself is genuinely powerful and genuinely free. It just isn’t magic — small local models are small local models.

Why this matters for you

Here’s the shift worth internalizing: the most valuable companies in tech are now spending billions to make capable AI free to download and run. That means the barrier to learning AI has basically collapsed. You don’t need a research budget or an API bill to experiment — you need a laptop and one ollama pull.

Open weights = free AI on your laptop, today. Run nemotron-mini, poke at it, break it, learn where it’s strong and where it isn’t. That instinct — knowing what a model can and can’t do — is the exact skill the next decade of jobs will pay for.