yasp
yasp is a German-Canadian deep tech startup building yasp.compile, an Agentic AI Compiler that autonomously generates hardware-optimized GPU kernels for AI training and inference via a single API call, targeting AI developers and enterprises seeking to escape NVIDIA lock-in.
- Company typePrivate
- Founded2023
- HeadquartersMunich, Germany
- Headcount11–50
- GTM typeB2B
- OfferingSoftware
What yasp does
yasp (legal entity yasp.ai GmbH) is a German-Canadian deep tech startup founded in 2023 that builds yasp.compile, an Agentic AI Compiler designed to automate the optimization of AI model training and inference on heterogeneous hardware. The product is an LLM-based agent that analyzes PyTorch computational graphs, generates hardware-specific GPU kernels (CUDA, Triton, HIP), and iteratively refines them through compilation checks, numerical testing, and hardware-in-the-loop profiling. The compiler applies techniques including kernel fusion, quantization, and algebraic rewrites—demonstrated with outcomes such as up to 10x faster training, up to 24x speedup on core kernels, up to 6.25x on the IBM Granite Mamba layer, and up to 3x end-to-end over torch.compile on IBM Granite 4.0. A 2.91x speedup on MiniGPT running on AMD Radeon PRO V710 GPUs via Azure was jointly validated with Microsoft and AMD.
The company targets two primary segments: AI developers and ML engineers who need performance without manual tuning, and enterprises seeking hardware flexibility and cost control across AI infrastructure. Distribution is hybrid: an early access program and single API call integration act as product-led growth levers, while demo booking and regional offices in Munich (European HQ) and Montreal (North American hub) support direct enterprise sales. Marquee technical partnerships include IBM (Granite optimization), AMD and Microsoft Azure (AMD GPU inference on Azure NVads V710 PRO), and the PyTorch Foundation, which yasp joined in December 2025. Pricing follows a subscription SaaS model (monthly or annual), with quote-based enterprise pricing and no publicly disclosed tiers.
The business is at an early commercialization stage. yasp operated in stealth for roughly two years before launching its public demo in December 2025 and its Early Access Program in January 2026, with a third US office planned by end of 2025. It closed a $5 million seed round in September 2025 led by Capnamic, with participation from Start-up BW Innovation Fonds & MBG BW and business angels. As of the available data, revenue is not publicly disclosed and the company has 11–50 employees.
yasp firmographics
Firmographics- Name
- yasp
- Legal name
- yasp.ai GmbH
- Website
- https://yasp.ai/
- Company type
- Private
- Founded year
- 2023
- Operating status
- Operating
- Headcount range
- 11–50 employees
- Short description
- yasp is a German-Canadian deep tech startup building yasp.compile, an Agentic AI Compiler that autonomously generates hardware-optimized GPU kernels for AI training and inference via a single API call, targeting AI developers and enterprises seeking to escape NVIDIA lock-in.
- Ownership category
- akta.pro rank
yasp industry classification
Industry- Product category
- AI Compiler / Developer Tools
- NAICS
- Software Publishers (5132), Custom Computer Programming Services (541511)
- SIC
- Services-Prepackaged Software (7372), Services-Computer Programming Services (7371)
- akta.pro primary industry
- AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers) (HDAAAAAI)
- akta.pro secondary industries
- Model Development & Training Platforms (AutoML, Notebooks, Feature Stores) (HDAEANAB), Model Training & Hyperparameter Optimization (HDAAABAE)
Keywords
Where yasp is headquartered
LocationHeadquarters
- HQ city
- Munich
- HQ country
- Germany
- HQ region
- Europe
Offices4 records
Markets served
yasp business model
Business model- GTM type
- B2B
- Offering type
- Software
- Cost components
- Personnel, Technology or R&D, Marketing or Sales, Operations
Revenue model
- Subscription SaaS: Monthly or annual subscription model for access to the Agentic AI Compiler technology. Customers pay flexibly to access the platform.
- Early Access Program: Limited early access program for organizations to trial the compiler and shape product roadmap while benefiting from performance gains.
Pricing tiers
| Model | Billing | Price |
|---|---|---|
| Subscription | Monthly | Subscription model with flexible monthly or annual billing |
Go-to-market motion3 records
Distribution channels3 records
Marketing channels5 records
yasp product offering
Product offeringCore offering
yasp offers yasp.compile, an Agentic AI Compiler that automatically generates optimized, hardware-specific kernels for AI model training and inference through a single API call. The compiler uses an LLM-based agent to analyze computational graphs, generate and iteratively refine CUDA, Triton, and HIP kernels, and tune performance for the target hardware (NVIDIA GPUs, AMD GPUs, CPUs, or accelerators). Developers submit PyTorch model workloads and receive custom-tuned kernels without manual code rewriting or hardware-specific tuning, eliminating vendor lock-in across heterogeneous hardware.
Product overview
yasp offers a single unified product: yasp.compile, the Agentic AI Compiler. This product serves as an intelligent compiler layer that bridges AI model development and heterogeneous hardware infrastructure. With a single API call, developers can automatically generate custom kernels tailored to target hardware (GPUs, CPUs, or accelerators) for both training and inference workloads. The compiler uses AI agents to analyze computational graphs, identify bottlenecks, and produce optimized code without requiring developers to rewrite their models or manually tune performance.
Differentiator
Problem solved
Functional benefit
Brands
- yasp.compile: The Agentic AI Compiler - a breakthrough technology designed to accelerate AI model training and inference, improve performance, and reduce time-to-market.
Products and services
- yasp.compile (Agentic AI Compiler) An agentic AI compiler that automatically generates optimized, hardware-specific kernels for AI model training and inference via a single API call. It serves AI developers, ML engineers, AI researchers, and enterprises that need peak performance across NVIDIA GPUs, AMD GPUs, CPUs, and accelerators without manual kernel tuning or code rewrites.
Quantifiable outcome
- Up to 10x faster AI training and lower costs
- +6 more outcomes
Companies that use yasp
Customer profileNamed customers3 records
Segments3 records
Ideal customer profiles3 records
yasp technology and API
TechnologyTechnology focussed Yes
API detail
- Has API
- Yes
- API docs
- API detail
Core technology
AI maturity
App detail
Integration3 records
AI capability4 records
Feature6 records
yasp partnerships and signals
Strategic signalPartnerships
Four partnerships are on record, tiered core and flagship.
- PyTorch Foundationcoreyasp joined as a member of the PyTorch Foundation starting December 2025, signaling commitment to agentic AI innovation within the open-source ecosystem hosted by the Linux Foundation. Membership indicates alignment with PyTorch ecosystem and potential deeper technical integrations.
- AMDcoreStrategic collaboration to unlock high-performance AI inference on AMD Radeon PRO V710 GPUs. yasp generates optimized HIP kernels specifically tuned for AMD hardware. Partnership enables competitive inference performance vs NVIDIA A10 at lower cost.
- Microsoft AzurecoreCollaboration with Microsoft Azure to demonstrate AI inference performance on AMD GPUs in Azure's cloud infrastructure. Azure NVads V710 PRO virtual machines provide the deployment platform for yasp's optimized models.
- IBMflagshipIBM collaboration showcasing Granite 4.0 model optimization. yasp achieved up to 3x end-to-end speedup over torch.compile on IBM's Granite Hybrid 4 model family. Demonstrates production-ready optimization of IBM's open-source models.
Scale indicators9 records
Recent moves12 records
Expansion highlights6 records
yasp competitors and assessment
Company assessmentDirect peers
- Together AI: AI cloud platform focused on fast, cost-efficient open-model training and inference across heterogeneous hardware. Comparable in its positioning around hardware flexibility and inference cost economics.
- Lightning AI: PyTorch-native AI development platform with a strong focus on training efficiency and multi-hardware deployment. Comparable as a developer-facing layer that abstracts hardware complexity for AI teams.
- Anyscale: Operates Ray-based AI compute platform with a strong optimization and scaling story for training and inference workloads. Comparable in targeting AI infrastructure efficiency, though broader than kernel-level optimization.
- Modular AI: Developer platform for AI inference with its own compiler/runtime (Mojo) targeting heterogeneous hardware. Closely comparable as a hardware-agnostic AI performance engineering layer competing for the same developer mindshare.
- OctoAI (OctoML): Optimizes and runs AI models efficiently across hardware accelerators using compiler-based approaches. Direct overlap on the value proposition of automated hardware-aware AI optimization as a managed service.
Broad incumbents
- DeepSpeed (Microsoft): Microsoft's open-source deep learning optimization library for training and inference. Comparable as an incumbent optimization stack bundled with a major framework ecosystem that yasp must displace.
- PyTorch / torch.compile: First-party PyTorch compiler (torch.compile/Inductor) shipped free with the dominant framework. The primary incumbent yasp's performance claims are explicitly benchmarked against.
- Hugging Face: Dominant AI model hub and inference platform with growing optimization tooling. Comparable as a layer sitting between AI developers and deployed models where performance economics increasingly matter.
- NVIDIA TensorRT: NVIDIA's high-performance deep learning inference optimizer. Comparable on the optimization layer for production AI, with the advantage of being first-party on the dominant accelerator platform.
- Google XLA / JAX: Open-source compiler and numerical computing stack optimizing ML workloads across CPUs, GPUs, and accelerators. Comparable as an established hardware-agnostic compiler that competes for the same optimization layer.
Market position
Strengths4 records
Weaknesses4 records
Competitive moat5 records
Key risks6 records
Key highlights6 records
Customer concentration
yasp social profiles
Digital presenceyasp financial estimates
Financial estimateRevenue estimate
Valuation estimate
yasp leadership team
Management profileNumber of profiles
Profiles4 records
yasp funding detail
Funding detailFunding overview
Funding rounds2 records
Investors5 records
Funding detail is available on the Subscription and Enterprise plan.Contact sales →
yasp M&A and investment
M&A and investmentM&A
Investments
M&A and investment is available on the Subscription and Enterprise plan.Contact sales →
Frequently asked questions about yasp
What does yasp do?
yasp offers yasp.compile, an Agentic AI Compiler that automatically generates optimized, hardware-specific kernels for AI model training and inference through a single API call. The compiler uses an LLM-based agent to analyze computational graphs, generate and iteratively refine CUDA, Triton, and HIP kernels, and tune performance for the target hardware (NVIDIA GPUs, AMD GPUs, CPUs, or accelerators). Developers submit PyTorch model workloads and receive custom-tuned kernels without manual code rewriting or hardware-specific tuning, eliminating vendor lock-in across heterogeneous hardware.
Is yasp a public or private company?
yasp is a private company. It is classified as venture growth investor backed and is currently operating.
When was yasp founded?
yasp was founded in 2023. It employs 11 to 50 people.
Where is yasp based?
yasp is headquartered in Munich, Germany, in the Europe region.
How does yasp make money?
Two revenue lines are on record. Subscription SaaS are the primary driver. The others are early Access Program.
Who are yasp's main competitors?
Direct peers on record are Together AI, Lightning AI, Anyscale, Modular AI and OctoAI (OctoML). Broad incumbents are DeepSpeed (Microsoft), PyTorch / torch.compile, Hugging Face, NVIDIA TensorRT and Google XLA / JAX.
Does yasp have an API?
Yes. yasp.compile is an Agentic AI Compiler accessible via a single API call. Developers submit their PyTorch model workloads through the API, and the compiler automatically profiles, generates optimized kernels, and tunes performance for target hardware without requiring code rewrites.
What industry is yasp in?
yasp's product category is AI Compiler / Developer Tools. Its primary akta.pro industry code is HDAAAAAI, AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers), with a secondary code of HDAEANAB, Model Development & Training Platforms (AutoML, Notebooks, Feature Stores). Its NAICS code is 5132 and its SIC code is 7372.