Olix Raises $312M for AI Inference Chip Development

London-based deep-tech firm Olix has successfully closed a substantial Series B funding round, raising $312 million at a valuation of $3.3 billion. This significant capital injection comes as the company accelerates its development of specialized hardware for AI inference, a critical and rapidly expanding segment of the artificial intelligence market. The funding round saw participation from a high-caliber group of investors, underscoring the immense potential seen in Olix's novel approach to AI computation.

Among the key backers are industry giants like chip designer Arm, the high-frequency trading firm Hudson River Trading, and influential angel investor Reed Hastings, co-founder of Netflix. The round was reportedly led by New York-based deep-tech investor Fundomo, which also participated in Olix's previous funding. Existing investors, including Hummingbird Ventures, Crane, Plural, Creandum, Phoenix Court, and Transition, all increased their commitments, demonstrating strong confidence in the company's trajectory.

Adding strategic weight to the round, the UK government's Sovereign AI fund also contributed. This state-backed investment highlights the UK's focus on fostering domestic capabilities in advanced semiconductor technology, positioning Olix as a key player in this national strategic initiative. The government views advancements in chip manufacturing as a crucial lever for economic and technological sovereignty.

Olix is challenging the conventional wisdom in AI hardware design by proposing a new architecture for inference. Instead of relying on universal chips for all computational tasks, Olix's approach, exemplified by its DX-1 decoder accelerator, distributes AI models across multiple specialized chips. This "roll-out" strategy allows each chip to handle specific parts of a model, optimizing performance and efficiency. The system utilizes an optical interconnect for high-speed data transfer between chips and a deterministic compiler for workload distribution across racks, aiming to overcome the limitations of current inference hardware.

The company claims its architecture can deliver over 10,000 tokens per second per user for models in the 100 billion parameter class, with superior output per watt compared to traditional universal chips. Notably, the DX-1 is designed to operate using on-chip SRAM and avoids reliance on advanced packaging and High-Bandwidth Memory (HBM), components currently facing significant supply chain constraints. This strategic design choice could provide Olix with a competitive advantage in a market grappling with component shortages.

The influx of capital will be instrumental in bringing the DX-1 chip to market by the second half of 2027. The funding will also support the expansion of Olix's team, including the strategic appointments of Nick McKeown, a renowned expert in software-defined networking and a Marconi Prize recipient, to its board of directors, and Matt Briers, former CFO of Wise, as its new Chief Financial Officer. These additions bring deep technical and financial expertise to Olix as it navigates the competitive AI chip sector, which has seen substantial investment, with companies like Cerebras and d-Matrix also securing significant funding.