The desktop just became the new front line

Perplexity recommends you buy a $599 Mac Mini and leave it running around the clock as a dedicated AI computer. The same day, Google asked you to press Option+Space. The gap between free and $200 a month plus dedicated hardware is the entire strategic debate about desktop AI, compressed into a single afternoon.

Google launched a native Gemini app for Mac, built in Swift, summoned with a hotkey, free for everyone. It can see your screen, analyse files, and answer questions in context. The playbook is familiar: subsidise the product, capture the surface, monetise later. Google doesn't need Gemini on your desktop to make money today. It needs Gemini there to make sure you don't develop the habit of asking someone else.

Perplexity went the opposite direction. Its Personal Computer feature, available to Max subscribers at $200/month, is an always-on agent that accesses your local files, reads your email, manages your calendar, and sends iMessages on your behalf. The recommended setup is a dedicated Mac Mini running 24/7. Perplexity is charging a premium because it's selling agency, not answers. The agent works whether you're watching or not.

Then Mozilla released Thunderbolt, an open-source AI client that self-hosts, works with any model, and connects to internal data through MCP and Haystack. No subscription, no vendor lock-in, no data leaving your infrastructure.

Three business models, one old playbook

The pattern will feel familiar to anyone who watched the browser wars. Microsoft gave away Internet Explorer to own the platform. Netscape charged for quality. Mozilla offered freedom. That sequence played out over years. This time it happened in a day.

Mozilla is, quite literally, running the same play it ran in 2004. The pitch is identical: the incumbent wants to own your data pipeline, and we're the open-source alternative that keeps you in control. The difference is that in 2004, the browser was a window to the web. In 2026, the desktop AI client is a window to your files, your messages, and your organisational knowledge. The stakes of lock-in are proportionally higher.

I think the conventional take, that everyone is racing to the desktop, misses what's actually being contested. Google is treating AI as a utility: free, ambient, always available, like search. Perplexity is treating it as an agent: expensive, autonomous, doing work on your behalf. Mozilla is treating it as infrastructure: something you run yourself, like a database. These approaches can't all win the same user. A company that self-hosts Thunderbolt is making a fundamentally different bet about AI than someone who hands Perplexity their iMessage credentials.

For product builders, the implication is specific: if your AI feature still lives behind a browser tab and a login screen, you're competing in a market that's migrating underneath you. The desktop-native apps shipping this week have access to local files, system-level context, and persistent state that a web app cannot match. The question is no longer whether your AI should be native. It's which of these three economic models yours belongs to.

---