OpenAI recently announced a huge collaboration with Broadcom to create and deploy 10 gigawatts of specialised AI accelerators.
Broadcom will take on development and deployment beginning in late 2026.
What’s the idea? OpenAI’s ability to control its own hardware allows it to optimise models at every layer, eliminating reliance on third-party chipmakers.
Also read: EU to double AI science funding, launch €1 billion strategies
What this means for creators
The distinction between “software” and “hardware” is dissolving. If your product depends heavily on inference or unusual model behaviour, hardware-level control may enable new performance tiers.
Over time, you may see more robust APIs or infrastructure created on top of these specialised chips. This means new building pieces for your own tools.
However, in the short run, expenses and complexity will be significant. Don’t expect that everyone can jump into custom computing overnight.
What this means for entrepreneurs
This is a signal: the next AI arms race will be fought over infrastructure as much as algorithms. There is potential for success if you can build or control elements of the stack.
Partnerships with chipmakers or edge/hardware players could serve as strategic differentiators. Do not build in isolation.
Monitor your capital intensity. The hardware sector is capital-intensive. Make sure your company model (or investors) can withstand significant lead times and infrastructure bets.
