the compute divide
/kəm-PYOOT di-VYDE/·noun
The widening gap between those with access to serious computing power — and everyone else.
Definition
The unequal distribution of AI-scale computation among nations, institutions and companies: the GPU-rich and the GPU-poor.
The individual version: gated access to frontier AI tools — identity checks, credit walls, vetted previews — sorting users into compute-haves and have-nots.
Origin
First recorded in AI-policy discussion in the mid-2020s on the pattern of "the digital divide," gaining currency in 2026 as access to top models tightened through pricing, verification and allocation.
Usage
- The digital divide was about getting online. The compute divide is about what you're allowed to run once you're there.
- Scholarships used to buy books; now they buy tokens. That's the compute divide.
- Export controls turned the compute divide into foreign policy.
Frequently asked
How is the compute divide different from the digital divide?
The digital divide concerned connectivity. The compute divide concerns capability: access to the processing power and frontier models that increasingly determine productivity.
Who sits on which side?
A handful of firms and states own most frontier compute; universities, startups, and the Global South largely rent slivers of it.
Why did the term gain traction in 2026?
Access tightened visibly — ID verification, vetted previews, usage credits — making the divide legible to ordinary users, not just labs.
Can the compute divide narrow?
Cheaper inference, open-weight models, and public compute programs push one way; export controls and capex concentration push the other.