Perplexity and Nvidia Want Your Home Office to Cost More Than a Car
Perplexity just dropped a 'local' version of its AI agent, and it’s a masterclass in hardware privilege. While the original relied on the cloud, this new 'Portable Computer' demands a literal supercomputer on your desk just to run a chatbot without latency.
The new Portable Computer by Perplexity effectively ditches the cloud to run entirely on the user's hardware. While the previous Computer agent used a clever OpenClaw approach to offload heavy lifting to remote servers, this iteration is a total isolationist, demanding serious iron to keep your digital butler local.
Forget that $900 Mac Mini; that setup won't cut it here. The system is currently locked to heavy-duty gear like the Nvidia DGX Spark, a machine that costs roughly five times as much as a standard consumer desktop. Linux power users with Nvidia RTX cards might get a pass, provided their rig doesn't melt under the pressure of running a 27B parameter model.
Speaking of models, the choice is currently restricted to Alibaba's Qwen 3.8 27B or a custom-trained PPLX 27B, with Nvidia's Nemotron 3.5 Lightning expected to join the party soon. Despite being trapped inside your own hardware, the agent still manages to poke its head into your Slack, Google Drive, Gmail, and GitHub, acting as a direct bridge between your sensitive data and your local GPU's thermal throttling limit.
The audacity of branding a multi-thousand-dollar piece of data center equipment as a 'Portable Computer' is the kind of marketing gymnastics usually reserved for luxury watchmakers. If Nvidia succeeds in its plan to value Perplexity at $30 billion, the cost of keeping your AI agent local might eventually rival the cost of a down payment on a house, proving that privacy is truly the most expensive luxury item on the market.
Source: VentureBeat
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.