SAN FRANCISCO: RadixArk has come out of stealth with one of the biggest seed rounds of 2026, a $100 million raise aimed at making frontier AI infrastructure more accessible.
Accel and Spark Capital co-led the investment. A string of strategic names joined the round, including NVIDIA Ventures, AMD Ventures, MediaTek, and Walden Catalyst. Salience Capital, A&E Investments, HOF Capital, and LDV Partners also came on board.
The sheer size of the round speaks to how hungry investors are for startups operating at the compute layer of AI. RadixArk is going after a specific problem, the infrastructure bottleneck that keeps most enterprises from fully adopting frontier models at scale.
The startup builds end-to-end hardware and software for training and serving large AI models. The platform aims to cut costs for companies deploying advanced AI workloads. RadixArk wants to lower barriers traditionally held by hyperscaler cloud providers.
Strategic participation from NVIDIA, AMD, and MediaTek signals deep silicon alignment. The chip giants rarely co-invest in single rounds, making the configuration notable. Each likely views RadixArk as a meaningful customer for future generations.
Accel’s leadership reflects sustained venture conviction in AI infrastructure plays. The firm has backed multiple billion-dollar compute startups during the current cycle. Spark Capital similarly concentrates capital on foundational AI layers.
“The next phase of AI requires infrastructure built for inference economics,” a RadixArk co-founder says.
The funding lands amid massive infrastructure spending across the AI sector. Hyperscalers commit hundreds of billions to AI data centers annually. Specialized startups bet on niches that big cloud providers cannot serve efficiently.
DeepInfra recently raised $107 million for similar inference-focused infrastructure. CoreWeave continues its rapid scale-up across enterprise GPU markets.
RadixArk plans to disclose technical specifications and customer details over coming months. The company has not announced general availability timing.
Previous
