Nvidia Made $89 Billion From Data Centres. Now It Wants AI on Your Desk.
Nvidia’s data-centre revenue rose 117% while Perplexity launched a local AI agent. The real shift is hybrid AI—routing work by cost, speed, privacy and capability.
Nvidia made $89 billion from data centres in one quarter.
Then it helped Perplexity move an AI agent out of the data centre and onto a desktop.
Nvidia wins either way.
Its latest numbers show just how fast cloud AI is growing. Total quarterly revenue reached $96.2 billion, up 106% in a year. Data centres accounted for 92.5% of it.
At the same time, Perplexity launched Portable Computer. It can run an AI agent locally on Nvidia’s DGX Spark, using a 27-billion-parameter model. Local work does not count towards Perplexity’s token limits, and users can switch to more powerful cloud models when necessary.
This is not yet AI for the average laptop. DGX Spark has 128GB of memory and is specialised desktop hardware. Running locally does not mean everything stays offline either. An agent connected to Gmail, Slack or Google Drive still communicates with those cloud services.
But the direction is clear.
AI is not moving from the cloud to the PC. It is starting to run in both places.
Routine, private or time-sensitive steps can be handled locally. More demanding reasoning can still be sent to a data centre. Software can route each part of a task based on capability, cost, speed and permission.
Apple already applies this principle with Apple Intelligence. Google does it with Gemini Nano. Perplexity is now extending it to full AI agents.
The next AI race may therefore be less about finding one model that can do everything.
It may be about deciding which model should do what, where it should run and what information is allowed to leave your device.
And Nvidia sells the infrastructure on both sides.
What matters more to you: the smartest AI, or knowing where your data is processed?
#AI #Nvidia #Perplexity #AIAgents #FutureOfAI
