Blog
AI subscriptions are the electric bill. Your PC is the solar panel
The trouble is that it all comes at a cost. You’ll pay for a cloud AI subscription, with a limited number of tokens which typically expire at the most inopportune time. But no matter what, cloud AI will always keep charging you. It’s the exact reason so many people invest in solar panels for their home. If my local utility company is going to keep charging me money (and raising their rates) why not invest in my own power generation and get it it for “free?”
We’re entering an era where your local PCs can become a personal AI “solar panel” of sorts, driven by AI goliaths like Nvidia and Microsoft — but not only by those titans.
Welcome to PCWorld’s newsletter, Smart Mode! The focus of this newsletter is productivity: from software to hardware to peripherals, all with the idea to get what you need to get done, efficiently.
I’m your host, Mark Hachman, covering the productivity tips, tools, and trends that you need to know about, both from within PCWorld and beyond. If you want the latest issue in your inbox each week, just sign up right here.
It’s a cost thing
Nvidia, naturally, has always been in favor of powerful, local GPUs. Nvidia’s Personal AI Router (PAIR) is an intriguing concept, combining every GPU-equipped device (AMD or Nvidia) on your network into a shared pool of AI computing power. Nvidia said it has tested RTX PAIR on up to 18 devices on a local network, though there is a catch: PAIR assumes a hardwired network is in place, not the mesh routers throughput homes. That means that you’d have to build in a hardwired switch of some kind.
Nvidia
I actually think a smarter implementation is Exo, a project that’s been on my to-do list. Exo basically (supposedly) just finds and assimilates both CPU and GPU resources and storage into an AI cluster, daisy-chaining them together via Thunderbolt connections. For me, with lots of Thunderbolt cables, this seems like a no-brainer. It’s also much more forgiving for homes with existing hardware of all stripes… though hardwiring a Thunderbolt cluster likely requires much tighter quarters than an Ethernet cluster built around your house, which RTX PAIR offers.
The problem with both solutions, of course, is power: you’re going to be pulling a lot of juice and generating a lot of heat to make either solution work. Wouldn’t it be nice to take advantage of the cloud and local AI?
That’s where Microsoft is now, with what it’s calling “unmetered intelligence.” In a presentation at IFA in Berlin last week, Microsoft’s corporate vice president of Windows + Devices, Mark Linton, argued that PC makers and owners should just “offload to the PC” their AI workloads.
