Microsoft New Surface Laptop Ultra with NVIDIA GPU for Local AI - OpenAI is DEAD

A 26-minute episode from Eli the Computer Guy, uploaded on 12 October 2026, reacting to a news report that Microsoft will sell a Surface Laptop Ultra built around the same Nvidia chip as Nvidia’s small desktop AI box, for local AI work. Eli reads it as the industry admitting what he has argued for a while: AI will run mostly on the device, with the cloud called in when needed, which is just how distributed systems have always been built. Google already ships on-device models alongside cloud Gemini; Windows will route some coding requests to the local GPU and others to the cloud.

That, he argues, undercuts OpenAI’s founding bet that more data and more GPUs make better AI, delivered from the centre like a metered utility. Most jobs don’t need the biggest model, any more than his Ford Escape needs a supercar’s engine in a 25 mph zone, so small models solving specific problems win. He thinks the period when the big companies talked up AI together is over and they’re now undercutting each other, with Meta’s open releases and Microsoft’s hardware both pulling against OpenAI. If laptops are powerful enough, he asks, what does OpenAI actually sell? He separates the technology from a product and a product from a business.

He makes the latency case with a Raspberry Pi running an image model at about five minutes a picture: slow, but free, private and plenty for a small business processing photos in the background, and agents make the wait easier to ignore because no one is watching. He won’t buy the laptop himself (version-one Microsoft hardware, and Windows on ARM has gaps), and says Macs are the better ARM machines.