Build on Trainium: Responsible AI call for proposals - Fall 2025

About this CFP

What is Build on Trainium?
Build on Trainium is a $110MM credit program focused on AI research and university education to support the next generation of innovation and development on AWS Trainium. AWS Trainium chips are purpose-built for high-performance deep learning (DL) training of generative AI models, including large language models (LLMs) and latent diffusion models. Build on Trainium provides compute credits to novel AI research on Trainium, investing in leading academic teams to build innovations in critical areas including new model architectures, ML libraries, optimizations, large-scale distributed systems, and more. This multi-year initiative lays the foundation for the future of AI by inspiring the academic community to utilize, invest in, and contribute to the open-source community around Trainium. Combining these benefits with Neuron software development kit (SDK) and recent launch of the Neuron Kernel Interface (NKI), AI researchers can innovate at scale in the cloud. The program can also support model inference on AWS Trainium chips and the closely-related AWS Inferentia chips.

What are AWS Trainium and Neuron?
AWS Trainium is a purpose-built AI chip developed by AWS for accelerating building and deploying machine learning models. Built on a specialized architecture designed for deep learning, Trainium accelerates the training and inference of complex models with high output and scalability, making it ideal for academic researchers looking to optimize performance and costs. This architecture also emphasizes sustainability through resource-efficient design, reducing environmental impact. Amazon has established a dedicated Trainium research cluster featuring up to 40,000 Trainium chips, accessible via Amazon EC2 Trn instances. These instances are connected through a non-blocking, petabit-scale network using Amazon EC2 UltraClusters, enabling seamless high-performance ML training. Trn instances is optimized to deliver substantial compute power for cutting-edge AI research and development. This unique offering not only enhances the efficiency and affordability of model training but also presents academic researchers with opportunities to publish new papers on underrepresented compute architectures, thus advancing the field.

Focus on Responsible AI (RAI)
The focus of this CFP is on Responsible AI (RAI) - a major field of AI research that is critically needed for greater adoption of AI technology. Broadly speaking, AWS defines RAI to include the dimensions of 1/ Fairness, 2/ Explainability, 3/ Privacy and Security, 4/ Safety, 5/ Controllability, 6/ Veracity and robustness, 7/ Governance, and 8/ Transparency. Within RAI the five topics below are priority topics for research, based on their expected benefits to builders, customers and the scientific community.

Please develop your grant proposal addressing one or more of these topics, or make a strong case for another topic within RAI.

1. AI safety and alignment. It is critical to ensure that artificial intelligence systems remain beneficial, controllable, and aligned with human values as AI becomes increasingly capable and autonomous. We seek proposals that utilize Trainium to advance:
   
   1. Robustness and Reliability: ensuring AI systems perform consistently and safely under all conditions, including adversarial attacks, distribution shifts, and edge cases
   2. Alignment and Value Learning: ensuring AI systems understand and act in accordance with human values and intentions, including inverse reinforcement learning, reward modeling, superalignment and preference learning
   3. Scalable oversight: maintaining meaningful human control over AI systems as they become more capable to include reasoning, tool usage and distributed across agents.
   4. Fairness, privacy protection and bias mitigation: Training foundation models and multi-agent AI systems with inbuilt fairness objectives, bias mitigation and protection of privacy
2. Multi-lingual language models. Humans worldwide speak thousands of languages, but only a small fraction of these languages are used to natively train LLMs. There is a growing demand for advanced AI solutions for every human language. We seek work that:
   
   1. Introduces novel deep learning architectures, techniques, data and software that advance multi-lingual LLMs
   2. Trains models that excel in rarer human languages and releases datasets and tools, such as data preparation and tokenization tools for non-Latin alphabets
   3. Trains multilingual small language models (SLMs) for applications like guardrails for Asian languages
   4. Develops standardized benchmarks in knowledge, reasoning and other competencies in non-English languages with a focus on fairness and cultural inclusion
3. Representation engineering (RE). RE is focused on understanding, analyzing, and manipulating the internal representations learned by artificial neural networks. This field investigates how models encode, transform, and utilize information to make decisions. We seek proposals that utilize Trainium for RE, including:
   
   1. Analysis and Visualization: techniques for understanding how neural networks encode information, including feature visualization, attribution methods, and dimensionality reduction approaches
   2. Transparency and Interpretability: methods for explanation generation, decision verification, and model inspection
   3. Modification and Control: steering how models learn representations, including architecture design, training objectives, and direct manipulation of learned representations and unlearning
   4. Security and Counter-Hallucination: detection and prevention of adversarial attack or hallucinations using RE, including through data path injections.
4. Sustainability, Efficiency and Small Language Models (SLMs). The rapid advancement of generative AI technologies has raised resource consumption and environmental impact to a top concern of decision makers and application developers. Meanwhile, advances in SLMs enable more efficient and performant LLMs. Using cost and resource allocation as a proxy for carbon emissions offers benefits both to reduce carbon and to increase access to AI. We seek proposals that advance:
   
   1. Training and inference efficiency: improving the efficiency of training and inference using novel architectures, training curriculum and schedule optimization, LLM Ops and other techniques
   2. Novel SLMs: novel architectures, datasets and training strategies that advance the state of the art in SLMs such as pruning, matrix factorization, quantization, neural architecture search, and other strategies.
   3. Resource consumption of existing GenAI solutions: utilizing tools such as the Neuron SDK and Neuron Kernel Interface, research novel approaches to reduce computational overhead. Develop new algorithms, refine model architectures, and create software optimizations that significantly improve efficiency without compromising AI capabilities
   4. Real-time monitoring and optimization techniques for GenAI resource consumption: design intelligent systems capable of dynamically adjusting resource allocation based on workload demands and resource constraints. Predictive models for resource usage, adaptive algorithms for power management, and innovative approaches to balance performance and efficiency in live GenAI environments
5. Deep learning models for synthetic data generation. VAEs, GANs and other techniques can generate fully-synthetic data such as text, images and even entire databases. These datasets can accelerate application development while protecting privacy and is indispensable for high-risk and regulated industries. The related techniques of data augmentation can accelerate the training of neural models. We seek proposals that utilize Trainium for:
   
   1. Foundational research: conceptual breakthroughs in realism, efficiency, scale and user control over model developer for synthetic data generation
   2. Hierarchical, multimodal and structured data: novel techniques for using models to generate fully-synthetic hierarchical and logically-constrained data as found in relational databases or multimodal and time-dependent data
   3. Advanced data augmentation: novel data augmentation methods and models for single- and multi-modal data

Timeline

Submission period: October 1 — November 12, 2025 (11:59PM Pacific Time)
Decision letters will be sent out in March 2026

Award details

Principal Investigators (PIs) may receive the following:

• Applicants are encouraged to request AWS Promotional Credits in one of two ranges:
  
  • AWS Promotional Credits, up to $50,000
  • AWS Promotional Credits, up to $250,000 and beyond
• AWS Trainium training resources, including AWS tutorials and hands-on sessions with Amazon scientists and engineers

Awards are structured as one-time unrestricted gifts. The budget should include a list of expected costs specified in USD, and should not include administrative overhead costs. The final award amount will be determined by the awards panel.

Your receipt and use of AWS Promotional Credits is governed by the AWS Promotional Credit Terms and Conditions, which may be updated by AWS from time to time.

Eligibility requirements

Please refer to the ARA program rules on the rules and eligibility page.

Proposal requirements

PIs are encouraged to exemplify how their proposed techniques or research studies advance kernel optimization, LLM innovation, distributed systems, or developer efficiency. PIs should either include plans for open source contributions or state that they do not plan to make any open source contributions (data or code) under the proposed effort. Proposals for this CFP should be prepared according to the proposal template and are encouraged to be a maximum of 3 pages, not including Appendices.

Selection criteria

Proposals will be evaluated on the following:

• Creativity and quality of the scientific content
• Potential impact to the research community and society at large
• Interest expressed in open-sourcing model artifacts, datasets, and development frameworks
• Intention to use and explore novel hardware for AI/ML, primarily Trainium and Inferentia.

Expectations of Recipients

To the extent deemed reasonable, award recipients may acknowledge the support from ARA (“(e.g., “Research reported in this [publication/press release] was supported by an Amazon Research Award, [Cycle /Year].”). Award recipients will inform ARA of publications, presentations, code and data releases, blogs/social media posts, and other speaking engagements referencing the results of the supported research or the Award. Award recipients are expected to provide updates and feedback to ARA via surveys or reports on the status of their research. Award recipients will have an opportunity to work with ARA on an informational statement about the awarded project that may be used to generate visibility for their institutions and ARA.