I watched my Mac Mini stutter through a heavy prompt and felt the tug of a hidden bill. You’ve been trained to expect AIs to live in the cloud and quietly charge you for every extra call. I realized then that moving the model to my machine could change that balance of power.
My Mac Mini hummed as I tested a long prompt. Perplexity launches a local AI model that runs on your GPU instead of the cloud
I’ll be blunt: Perplexity has been quiet lately—save for its cameo as Joe Rogan’s fact checker—but it just made a bold, pragmatic move. You and I have wrestled with two problems for years: subscription token chases and handing our data to remote servers. Portable Computer tries to sidestep both by putting the model on your hardware.
Computer was Perplexity’s always-on agent platform that mostly ran its thinking in the cloud while sitting on devices like a Mac Mini. Portable Computer inverts that setup: the model runs locally. That means fewer surprise bills and more control over what leaves your machine.
Can I run Portable Computer on a Mac?
Short answer: not yet. Perplexity’s initial roll-out targets Linux boxes with Nvidia RTX GPUs and Nvidia’s DGX Spark desktop supercomputer. A Mac Mini with an M4 chip costs roughly $900 (€830), but Perplexity’s local-first approach currently favors heavier iron—expect to pay about 5x that for a DGX Spark, roughly $4,500 (€4,140).
At a datacenter demo, an Nvidia DGX Spark sat like a small refrigerator. What that hardware choice means for you
Perplexity made a trade: local models demand horsepower. That’s why the company ships Portable Computer with models it can run on high-end Nvidia GPUs. Initially you’ll get Alibaba’s Qwen 3.8 27B or Perplexity’s post-trained PPLX 27B, with Nvidia’s Nemotron 3.5 Lightning coming soon.
That narrower model set is the cost of doing everything on-device. Gone are the 19-model smorgasbords that Computer once advertised, but you gain latency, offline capability, and the relief of no token-led monthly surprises—Portable Computer won’t bill you per call.
Which GPUs are supported by Portable Computer?
Right now it’s Nvidia-first: RTX-class GPUs on Linux and the DGX Spark. If your rig has comparable Nvidia silicon, you’ll be in the game; if you’re on Apple silicon or older consumer cards, you’ll have to wait or look to cloud-based agents.
A Slack channel lit up with a chart pulled from a laptop. How Portable Computer mixes local security with app access
Perplexity demoed Portable Computer sending a data analysis to Slack and connecting to Google Drive, Gmail, and GitHub while keeping most computation local. That hybrid approach means the model acts like a locked safe for your compute but will ask before it opens the door to the cloud.
That “ask before pinging” line matters. Perplexity says Portable Computer will request permission when it needs to offload work—so you get a heads-up before your agent uses remote hardware or services.
Is Portable Computer fully offline?
Not absolutely. It defaults to local processing but can contact cloud resources with your permission. Think of it as local-first with a leash on cloud calls: you decide when the leash loosens.
I’ve been around enough product demos to smell partnership plays. Nvidia’s fingerprints are obvious—rumors say the chipmaker is considering an investment that would peg Perplexity at about $30 billion (€27.6 billion). That explains why Portable Computer is engineered to run beautifully on Nvidia silicon: investors like to see compatibility before writing checks.
Perplexity’s strategy flips the script on the agent market. You sacrifice model variety and accept a higher upfront hardware bill, and in return you get privacy, predictable costs, and speed. It’s a trade that will appeal to people and teams who want the model near their data and far from recurring token drain.
I’ll warn you, though: moving AI onto your GPU changes the operational checklist. You’ll be maintaining drivers, monitoring thermals, and thinking about backups in a way cloud users rarely do. If that sounds like work, you might prefer cloud agents; if it sounds like control, Portable Computer could be a tidy fit.
Perplexity just made local AI practical at scale for people with serious Nvidia silicon—so the question becomes: will you keep your AI in the sky or bring it home to run on your own silicon?