Post

Olix raises $312M at $3.3B valuation from Netflix’s Reed Hastings, Arm, to build Nvidia rival

Olix says it treats a data centre running AI inference like a factory that produces tokens. Each token needs hundreds of operations and different hardware.

Most companies use a single general-purpose chip for everything, but Olix’s X-1 platform distributes the model across several specialised chips, each handling a specific task. The system uses a flexible compute fabric instead of locking in a model’s design.

The chips use an optical interconnect, sending data via light rather than copper. Olix says this lowers latency and saves energy. The first chip, DX-1, is a decode accelerator for the stage when a model creates its output.

For models with 100 billion parameters, Olix claims that DX-1 can deliver over 10,000 tokens per second per user and use less power than general-purpose chips for large batches. The design is meant to scale to models with 10 trillion parameters or more.

Olix has over 140 employees in London, Bristol, Toronto, Austin, and San Francisco. The new funding will help launch DX-1 to its first customers in the second half of 2027, and support more platform development and manufacturing to increase chip production.