Raja Koduri’s Oxmiq Labs raises $35M to license custom AI silicon

Oxmiq Labs is positioning itself as the 'Arm of AI,' shifting the market away from capital-intensive, multi-year chip programs toward a flexible licensing model. By collapsing the GPU, CPU, and tensor components into a single IP block, the company addresses the high cost of custom silicon that currently limits AI hardware innovation to hyperscalers and top-tier chipmakers. For the streaming and media ecosystem, this lowering of design barriers could accelerate the arrival of highly specialized, low-power inference hardware for edge-based video processing and agentic AI workloads. Watch for the delivery of Oxmiq’s first production-ready IP batch to determine if its software-first approach can truly erode Nvidia’s dominance in the custom accelerator market.

The rise of custom silicon is reshaping the semiconductor landscape as hyperscalers move to reduce their reliance on merchant GPUs. Per Yahoo Finance in June 2026, demand for application-specific integrated circuits (ASICs) is accelerating, driven by the need for power efficiency in specialized AI workloads that general-purpose hardware cannot match. This trend has fueled a record year for semiconductor investment; according to Crunchbase data from January 2026, U.S. semiconductor startup funding reached $6.2 billion in 2025, an 85% year-over-year increase. High-profile rounds for companies like Etched, which raised $800 million for transformer-specific chips, and Cerebras, which secured $1.1 billion, underscore the scale of capital competing with incumbent leaders.

Simultaneously, the technical focus has shifted toward chiplet-based architectures and advanced packaging to manage costs. Per Silicon Analysts in June 2026, memory components now represent nearly 50% of total manufacturing costs for leading AI accelerators like the Nvidia B200, making Oxmiq’s focus on 'near-memory compute' and software portability increasingly relevant. The addition of Jim Keller to Oxmiq’s board further validates this modular approach. Keller, who leads Tenstorrent and previously designed architectures for Apple and Tesla, has been a vocal proponent of open-source hardware standards. His involvement, alongside technical veterans from Intel and AMD, suggests a strategic push to create a cross-industry infrastructure that circumvents the proprietary software moats currently defining the hardware market.