Yasp, a pioneering deep tech startup dedicated to advancing AI technologies, has announced the completion of its $5 million seed funding round, which was led by Capnamic. In addition to Capnamic, participants included Start-up BW Innovation Fonds & MBG BW as well as several seasoned business angels.
This investment will propel the development of Yasp’s Agentic AI Compiler, moving the company closer to its ambitious goal of launching this product by the end of 2025. The Agentic AI Compiler promises to provide an innovative and hardware-agnostic solution designed to substantially accelerate both AI training and inference, addressing critical industry needs.
As the global landscape of artificial intelligence evolves, investment in this sector is projected to soar to a staggering $1.8 trillion by 2030. This unprecedented growth presents considerable challenges for enterprises striving to scale their AI development efforts.
Notably, over 60% of corporate budgets dedicated to AI are currently allocated to software and infrastructure. As AI models become increasingly sophisticated and complex, future developmental efforts are expected to demand even greater financial resources.
Additionally, developers are grappling with the persistent challenge of vendor lock-in, which stifles innovation and limits long-term strategic planning. Consequently, achieving efficiency and flexibility has emerged as a paramount concern for organizations seeking to maintain their competitive edge, underscoring the urgent need for innovative solutions that can both accelerate AI advancements and manage financial expenditures effectively.
Yasp’s Agentic AI Compiler is ingeniously engineered to tackle these pressing challenges. This breakthrough technology is meticulously designed to accelerate the processes of AI model training and inference while simultaneously improving overall performance and reducing time-to-market.
The compiler integrates seamlessly with popular frameworks, such as PyTorch, thereby transforming the conventional methods developers use to build and deploy AI models. By simplifying advanced hardware optimization to just a few lines of code, Yasp enables developers to harness the full potential of the underlying hardware.
Through a simple API call, engineers can access hardware-specific compilation and optimization tailored to their AI models. In a marked departure from traditional runtime compilation methods, Yasp’s innovative approach involves generating custom kernels that are specifically designed for the target hardware.
This flexibility allows for scaling from individual kernels to entire models with ease. Furthermore, the backend optimizer utilizes intelligent algorithms to refine the model by precisely evaluating the interplay between the model, available hardware resources, and user requests, resulting in a substantial reduction in overhead. The result is the generation of lean, hardware-aware executables optimized for both training and inference tasks.
Initial tests conducted with Yasp’s Agentic AI Compiler have already yielded promising results, demonstrating remarkable performance improvements across a diverse array of training and inference workloads. Notably, these enhancements have been achieved without compromising model performance or accuracy, highlighting the efficacy of the compiler’s design and functionality.
As part of its strategic expansion, Yasp has recently established offices in Munich and Montreal, positioning itself to tap into key technology hubs.
The company plans to utilize the newly acquired funding to accelerate product development, broaden support for additional hardware platforms, and enhance integration with various widely used AI development tools. Yasp is resolutely committed to launching the Agentic AI Compiler by the end of the year, with a clear vision of empowering developers and enterprises to unlock new frontiers in AI innovation while maintaining control over their operational costs.
KEY QUOTES:
“We are building a product to directly address a critical and growing bottleneck in AI development: the time-consuming process of model training and deployment. Our vision is to empower developers and enterprises to achieve faster results, without sacrificing flexibility or accuracy.”
Stefan Krassin, Co-Founder and CEO of yasp
“Our approach is fundamentally different: instead of requiring developers to rewrite their complex models. With one API call, yasp rewrites and optimizes a model’s training and inference layers for the user’s target hardware.”
Reza Rahimi, CTO of yasp
“yasp is redefining what is possible in AI development by bridging the gap between software innovation and hardware performance. Their Agentic AI Compiler empowers engineers to achieve unprecedented speed and efficiency, while maintaining flexibility and avoiding vendor lock-in. We are excited to support yasp as they unlock new levels of productivity and scalability for the entire AI ecosystem.”
Christian Siegele, Managing Partner at Capnamic
“yasp’s technology is a game changer for enterprises looking to harness the full potential of their AI investments. By making advanced hardware optimization accessible and seamless, yasp is removing critical barriers in AI adoption and innovation. We are proud to support their mission and look forward to seeing their impact across industries.”
Cornelius Baral and Sven Schatz from Start-up BW Innovation Fonds & MBG BW
