I'm a gpu company, and i'm looking to sell to startups that train their own ai models but can't afford to sign their own gpu contracts directly. ideally, they have both inference and model training needs.

Shadeform is a startup (launched in 2023 and part of the Y Combinator S23 batch) that provides a platform for teams struggling with GPU availability, which suggests they target companies that cannot afford their own GPU contracts directly. They offer solutions for both inference and model training needs, as indicated by their services for deploying models for inference and aggregating data for training jobs.

Anarchy is a startup that offers a no-code training platform for building and training datasets, which indicates they train their own AI models. They provide tools for both model training and inference, as evidenced by their support for fine-tuning and distillation. Given their small team size and the fact that they offer a free tier and relatively low-cost Pro-Tier pricing, it is likely they cannot afford to sign their own GPU contracts directly.

Strong Compute is a startup (YC batch W22) that focuses on accelerating AI model training and provides on-demand GPU clusters. They offer solutions for both model training and inference needs, and their mission includes making GPU resources more accessible, which suggests they target companies that may not afford their own GPU contracts directly.

Cerebrium is a startup (Y-Combinator batch W22) that provides a platform for deploying machine learning models to serverless GPUs, indicating they train their own AI models. They offer solutions for both inference and model training needs, as evidenced by their support for streaming endpoints and various applications like transcribing podcasts and generating logos. Additionally, they enable the use of AWS/GCP credits to offset GPU costs, suggesting they target startups that may not afford direct GPU contracts.

Empower is a startup (Y-Combinator batch S23) that provides a serverless LLM hosting platform for fine-tuned models, indicating they train their own AI models. They offer both inference and model training needs as they support deploying and fine-tuning LoRA models. Empower's pricing structure, including a free tier and a low-cost startup tier, suggests they target startups that may not afford direct GPU contracts.

Sieve is a startup (Y-Combinator W22 batch) that focuses on AI applications in video and audio understanding and generation. They provide production-ready APIs and tooling for combining various models, indicating they train their own AI models. Their small team size (6 employees) and focus on providing cost-effective solutions (e.g., cheaper audio transcription API) suggest they may not afford to sign their own GPU contracts directly. They also have both inference and model training needs as evidenced by their work on AI lipsyncing, audio enhancement, and video AI avatars.

Mystic is a startup with a team size of 11 and backed by Y Combinator, indicating it is in the early stages of growth. The company provides a serverless cloud platform for deploying ML models, which suggests they train their own AI models. They offer solutions for both inference and model training needs, as evidenced by their support for ML deployments from ideation to scale-up, low-latency inference, and dynamic scaling. Additionally, their pay-as-you-go structure and support for using existing cloud credits indicate that they cater to companies that may not afford to sign their own GPU contracts directly.

Diffuse Bio is a startup in the healthcare industry, specifically focused on drug discovery and delivery using generative AI. They are developing foundation and generative models for proteins, which indicates they train their own AI models. Given their early stage and small team size, it is likely they cannot afford to sign their own GPU contracts directly. Additionally, their work involves both model training and inference needs as they develop and validate AI-generated proteins.

Airtrain AI is a startup (Y Combinator S22 batch) that provides a no-code compute platform for Large Language Models (LLMs), enabling AI practitioners to build, fine-tune, and serve open-source LLMs. This indicates that they train their own AI models and have both inference and model training needs. Additionally, they target businesses looking to scale AI prototypes into production-grade products and address challenges such as large AI bills, suggesting they may not afford signing their own GPU contracts directly.