Olix will ship its first photonic chips in 2027 for AI inference workloads as enterprises worldwide move beyond general-purpose GPUs. The startup joins companies across three continents raising capital for custom silicon—China's Nio subsidiary GeniTech closed a $330M Series A in February 2026 for autonomous driving chips while European and US startups pursue similar specialization strategies.
Photonic chips use light instead of electricity to transmit data, cutting power consumption in data centers. Early systems achieved 10x energy efficiency improvements for matrix multiplications, the core neural network operation. Hyperscale operators from Amazon Web Services to Alibaba Cloud are expanding AI capacity, creating demand for chips that reduce electricity costs now reaching $200-300 per kilowatt-hour in Singapore and parts of Europe.
Language Processing Units offer another path, using SRAM-centric designs to accelerate transformer models. These architectures store model weights in on-chip memory, eliminating memory bandwidth constraints. SRAM-based chips deliver 50-100x lower latency than GPUs for inference—critical for real-time applications from autonomous vehicles in Germany to financial trading systems in Hong Kong.
Advanced packaging enables chipmakers to combine photonic components with traditional CMOS logic. TSMC's CoWoS platform in Taiwan and Intel's EMIB technology support these designs, creating integration paths that emerged only in 2023. HPE is deploying these accelerators in enterprise systems across North America and Europe, stacking memory dies on processing units to reduce data movement costs.
Inference represents 80% of production AI compute costs globally, creating a $50B+ market for specialized processors by 2028. Foundation model providers from OpenAI ($840B valuation) to Anthropic ($380B) need real-time deployment infrastructure. Custom silicon targeting specific architectures captures this demand more efficiently than general-purpose alternatives, particularly as electricity costs and data center capacity constraints tighten across developed markets.
GeniTech's funding validates autonomous driving semiconductors as distinct from datacenter AI, with different requirements across regions—Chinese regulations demand local processing while European safety standards impose redundancy requirements absent in US markets.
