Seriously?! OpenAI and Broadcom Are Doing…Hardware.

Right, so the cloud-flare geniuses at OpenAI finally realized that relying on NVIDIA for everything isn’t a brilliant long-term strategy. Shocking, I know. They’ve partnered with Broadcom to build custom AI chips – because apparently writing software wasn’t enough of a headache. Broadcom will design and manufacture these things, specifically tailored for OpenAI’s models (mostly GPT, naturally).

The deal involves Broadcom investing $1.5 billion in OpenAI *now*, with more cash potentially flowing later. It’s all about scaling up their AI infrastructure, reducing dependency on the aforementioned GPU overlords and…wait for it… “long-term compute supply.” Like they didn’t see this coming? Idiots.

They’re talking about “next-generation” chips by 2026. Yeah, sure. I’ll believe it when I see it actually *work* and isn’t just another power-hungry silicon brick. And of course, they’re being vague on the specifics – because transparency is for suckers.

Basically, OpenAI wants to control more of its own destiny (and profit margins), and Broadcom wants a piece of the AI pie. It’s business, I guess. A frustratingly predictable piece of business at that. Don’t expect miracles though, this will probably just create another bottleneck somewhere else.

Source: https://techcrunch.com/2025/10/14/openai-and-broadcom-partner-on-ai-hardware/

Speaking of bottlenecks, I once had to debug a routing issue caused by someone using a potato as a network switch. A *potato*. Seriously. People are astonishingly incompetent. This OpenAI thing? Probably won’t be that bad, but I’m not holding my breath.