OLIX announced on August 3, 2026 that it had raised 312 million dollars in a Series B round at a 3.3 billion dollar valuation, and appointed Professor Nick McKeown to its board of directors. New investors in the round were Fundomo, Arm and Hudson River Trading. Existing backers Hummingbird Ventures, Crane, Plural, Creandum, Phoenix Court and Transition all increased their commitments, and Netflix co-founder Reed Hastings participated as an angel investor. The company also named Matt Briers, previously chief financial officer of Wise for nine years, as its CFO.
The money funds DX-1, an accelerator aimed specifically at the decode stage of large language model inference rather than at training. The architectural choice that distinguishes it is memory: DX-1 holds a model in fast on-chip SRAM instead of high-bandwidth memory, which means the design uses no advanced packaging and no HBM at all. OLIX says the part delivers over 10,000 tokens per second per user on 100 billion parameter models, at higher output token throughput per watt than general purpose chips, and that the architecture scales to models of 10 trillion parameters and above.
That memory decision is the commercial thesis. HBM and advanced packaging are the two hard bottlenecks in AI accelerator supply, both concentrated in a handful of suppliers with capacity contracted years ahead. A decode accelerator that needs neither sidesteps the queue, at the cost of fitting the working model into on-chip SRAM and stitching many chips together with photonic interconnect and deterministic rack-scale scheduling.
OLIX targets delivery to first customers in the second half of 2027, which places any revenue well past the current GPU cycle and after several competing inference-specific parts are due to ship. For an infrastructure buyer the entry is worth tracking not for near-term procurement but as a read on whether inference silicon is genuinely diverging from training silicon. The presence of Arm and Hudson River Trading on the cap table is the more immediate signal: an IP licensor and a latency-sensitive trading firm both see enough in a non-HBM decode architecture to fund it at a multi-billion dollar valuation three years before shipment.