French artificial intelligence startup ZML has launched ZML/LLMD, a free inference server designed to optimize the execution of large language models across a diverse range of hardware accelerators. Endorsed by Turing Award winner Yann LeCun and headquartered in Paris, the company aims to dismantle vendor lock-in and streamline AI inference deployment across Nvidia, AMD, Google Tensor Processing Units, Apple Metal, and Intel Arc architectures. As generative AI integration accelerates across enterprise and consumer sectors, inference processing has emerged as a critical bottleneck, often hindered by fragmented software stacks and proprietary hardware ecosystems. ZML/LLMD addresses these inefficiencies by enabling open-source models to run at peak performance regardless of underlying silicon. Founder Steeve Morin emphasized that the platform is engineered to deliver measurable speed and energy efficiency gains, offering organizations a hybrid approach to infrastructure that can lower operational costs and reduce dependency on single-vendor solutions. The startup operates in a rapidly expanding segment widely described as an inference gold rush. Competitors such as Baseten, Inferact, and RadixArk are similarly capitalizing on enterprise demand for optimized model serving. However, ZML differentiates itself through a lean twenty-person engineering team and a strategic focus on hardware-software co-design. The company recently secured twenty million dollars in venture funding, backed by investors including 20VC, Kima Ventures, LocalGlobe, and Puzzle Ventures. Notable backers include founders and industry leaders such as Solomon Hykes, Clement Delangue, and Julien Chaumond. ZML/LLMD enters the market as a complimentary offering, a deliberate strategy to capture deployment data and refine product-market fit before transitioning to a commercial model. Morin stated that monetization will be phased in based on utilization patterns and enterprise demand, prioritizing long-term adoption over immediate revenue. The initial release follows an earlier public ML framework from 2024, with the company planning further architectural updates and expanded chip compatibility in subsequent quarters. Beyond immediate commercial implications, the platform carries strategic significance for Europe’s emerging AI hardware ecosystem. Morin highlighted ongoing collaborations with domestic silicon developers such as Axelera, Fractile, Kalray, SiPearl, and SpiNNcloud, positioning ZML as a software enabler for regional chipmakers competing globally. The startup’s trajectory also underscores Paris’s growing capacity to retain top engineering talent and attract venture capital within the AI infrastructure stack. While Nvidia maintains a commanding presence in the accelerated computing market, ZML’s cross-platform approach signals a structural shift toward interoperable, cost-efficient inference pipelines. As enterprises scale model deployments, the ability to abstract hardware dependencies and optimize workloads across heterogeneous systems will likely influence procurement strategies. ZML’s near-term success will depend on developer adoption, independent performance benchmarks against established frameworks, and the eventual execution of its monetization roadmap. The company’s focus on practical efficiency and ecosystem neutrality positions it to capture market share among organizations seeking to future-proof their AI infrastructure.