Reed Hastings and Arm back Olix in $312M AI inference round

The investment signals a strategic shift toward specialized hardware as general-purpose GPUs face scaling and supply chain bottlenecks. By bypassing high-bandwidth memory and utilizing optical interconnects, Olix addresses the two primary constraints of modern AI data centers: physical interconnect latency and component shortages. For the streaming and video industry, such architectural breakthroughs are critical for reducing the high operational costs of real-time AI video processing and automated video intelligence generation at scale. Industry observers should track the scheduled 2027 rollout to see if Olix can match Nvidia's incumbent software ecosystem performance.

The funding of Olix comes amid a massive capital rotation into optical interconnect technology. Market projections from Tickeron and Mordor Intelligence in early 2026 suggest the optical interconnect sector will grow from $16 billion in 2024 to over $33 billion by 2031. This growth is driven by the physical limitations of copper, which has reached a ceiling in power density and signal degradation at the 1.6T speeds required for next-generation AI clusters. Per Tickeron (August 2026), 2026 marks the first year of mass production for 1.6T optical modules, shifting photonics from a peripheral research interest to a core infrastructure requirement.

Reed Hastings’ participation reflects his deepening involvement in the AI sector following his transition to Executive Chairman at Netflix. While he has historically focused on software, Hastings has recently diversified into AI infrastructure and safety. In March 2025, he donated $50 million to Bowdoin College to launch the Hastings Initiative for AI and Humanity, and in March 2026, he joined a $50 million investment round for UA1, a fund specializing in AI-driven defense technologies, per Forbes Ukraine. His backing of Olix suggests a move toward the foundational hardware layer as the cost of multi-agent AI token costs becomes a primary concern for content platforms.

Olix joins a competitive field of European and American startups challenging Nvidia's dominance. In early 2026, Dutch startup Axelera AI raised $250 million, while London-based Fractile secured $220 million for its own inference-specific architecture, according to Sifted (August 2026). These startups are increasingly focusing on the "decode" phase of inference, which is more memory-bound than compute-bound. By optimizing for this specific stage, companies like Olix and Groq—which also utilizes SRAM-heavy designs—aim to offer significantly lower Total Cost of Ownership (TCO) for hyperscalers managing trillion-parameter models.