AI hardware startup Olix Computing has raised $312 million in a Series C round, tripling its valuation to $3.3 billion. Backed by Arm and Netflix co-founder Reed Hastings, the London-based company is developing the DX-1 chip for AI inference. The chip uses on-chip SRAM instead of HBM and a novel 'slow and wide' optical interconnect, aiming to process over 10,000 tokens per second. Olix plans to ship its hardware in the first half of 2027.
Olix Computing funding round
- ▪The funding round included contributions from Arm Holding plc and Netflix Inc. co-founder Reed Hastings
- ▪Following its latest funding round, Olix Computing is now valued at $3.3 billion, triple its valuation from a previous round in February
- ▪AI hardware startup Olix Computing Ltd. has raised $312 million in a Series C funding round
DX-1 chip architecture
- ▪This on-chip SRAM approach is designed to store the large "KV cache" generated during AI inference without needing slower external memory
- ▪The DX-1 chip holds a language model in fast on-chip SRAM memory, forgoing off-chip HBM memory
- ▪London-based Olix is developing a chip called the DX-1, specifically optimized for the "decode stage" of AI inference workloads
Optical interconnect technology
- ▪The optical interconnect transmits data as light and uses a "slow and wide" design with a large number of slower channels
- ▪The use of NRZ encoding removes the need for costly digital signal processors to correct data errors
- ▪Olix's DX-1 chips will be part of a data center appliance called the X-1, where they are linked by an optical interconnect
- ▪Olix's interconnects use a reliable encoding method called NRZ, which is less prone to errors than the faster PAM4 technology
Performance specifications
- ▪Olix's DX-1 chip enables large language models with 100 billion parameters to process over 10,000 tokens per second
- ▪Multiple X-1 racks can be linked into clusters to run models with up to 10 trillion parameters
Manufacturing timeline
- ▪Olix Computing hopes to start shipping its chips in the first half of 2027
- ▪The company will use its latest funding to enhance its custom silicon platform and accelerate manufacturing efforts
Story comments
Loading comments…