On 2026-08-04, @cursor_ai posted that MoK now powers training across tens of thousands of GPUs at Cursor.
MoK now powers training across tens of thousands of GPUs at Cursor. In production, it raised end-to-end training throughput by 1.41x over our previous DeepEP-based stack.
What was announced
- MoK is now used for training across tens of thousands of GPUs.
- It delivered a 1.41x improvement in end-to-end training throughput compared with the prior DeepEP-based stack.
Context
The announcement concerns internal training infrastructure at Cursor, an AI coding tool company. No connection to Grok or xAI products is stated.
Limits of this report
This report contains only the claims in the source post. No details on MoK architecture, prior benchmarks, or future plans are provided.