Exactly because of what I previously said.
Firefox comes with tiny LMs that don't take a lot of storage and use a lot computing power like regular LLMs. With smaller size comes reduced "knowledge" and capability; and because Firefox AI doesn't collect any data, it's not being trained using user activity. Every time Mozilla wants to improve AI, they have to train it more themselves and then include updated model in new Firefox update.
Now, if you compare that tiny local Firefox AI to Gemini or ChatGPT, that would be like comparing an ant to a Godzilla. Gemini and ChatGPT are using enormous models and they are constantly growing in size because they are actively being trained by users whenever they ask them something. The larger model is, the more power it requires so if you ever wondered why they need to build new data centers and buy almost all inventory of RAM, GPUs and SSDs, that's your answer.