- Programs
- AWS Startups Custom-Model Inference Fellowship
AWS Startups Custom-Model Inference Fellowship
An invite-only, 12-week technical engagement for startups running inference workloads. Get $100K in AWS credits, dedicated 1:1 support from AWS GenAI Specialists, and hands-on guidance to optimize your inference stack on AWS.
Fellowship Benefits
*Startups must also qualify for the AWS Activate credit program.
$100K in AWS Credits
Distributed in two tranches: $50K at program start across all AWS services, and $50K at month 3 contingent on demonstrated progress toward your POC goals.
Dedicated Specialist Support for 12 weeks
Work directly with AWS inference specialists for the full program duration — architecture review, hands-on implementation, and performance optimization.
Assigned Technical Lead Specialist
Get assigned a dedicated Technical Lead Specialist matched to your inference journey category — not a generic mentor, a hands-on engineer working alongside your team in your environment.
Joint PR & Co-Marketing Opportunities
Participate in case study development, blog posts, conference speaking slots, and co-promotion with AWS and corporate partners.
Post-Program Transition Support
Finish with a custom transition plan for long-term success, including a clear path from POC to production.
Application timeline
Accepting Applications
Applications for the fellowship are open during this time. Please apply using the "Apply Now" links.
Sep 9 - Oct 3, 2026
Reviewing Applications
All applications are reviewed and assessed for fellowship fit.
Oct 4 - Oct 18, 2026
Applicants Notified
All applicants are notified of their acceptance or rejectance through their account managers.
Oct 19, 2026
Program Schedule
Kickoff & POC Charter
Work with your assigned lead specialist to create a POC Charter - what you will focus on during the fellowship.
Weeks 1-2
Implementation Phase
Work alongside your specialist to improve your inference stack on AWS.
Weeks 2-10
Closeout & Next Steps
Create a transition plan for post-program success with your specialist and account team.
Weeks 11-12
Frequently asked questions
Startups with active inference workloads (production or pre-production) looking to migrate to or optimize on AWS. Applicants should have a well-defined inference use case, an existing or anticipated bottleneck (cost, latency, throughput, or capacity), and engineering resources to commit to a 12-week POC engagement.
Ideal candidates are startups that are:
- Running inference workloads today
- Open to or already running inference on open-weight models (e.g., Llama, Qwen, GLM, Kimi, Minimax, DeepSeek), rather than exclusively using proprietary models like Claude or GPT
- Capable of model customization or familiar with Kubernetes-based deployments
This program is NOT for:
- Token usage optimization on Bedrock managed APIs
- Workloads exclusively using proprietary models (e.g., Claude, GPT) with no intent to move to open-weight models
- Companies without an active or planned inference workload
After acceptance:
- Weeks 1-2: 60–90 minute Inference Deep-Dive Strategy Call to produce your POC Charter (target architecture, success criteria, timeline)
- Weeks 2–10: Hands-on implementation with mandatory weekly check-ins and dedicated specialist support
- Weeks 11–12: Validation, benchmarking, and transition planning for post-program success
Yes. Startups from everywhere around the globe are welcome. Talk with your account manager for more information.
The Fellowship focuses on four primary inference challenges:
- Latency-Critical Inference — need frontier-model accuracy at significantly lower latency
- High-Traffic Production Scaling — capacity constraints with current providers at scale
- Agentic Inference — multi-model orchestration hitting cost/latency ceilings
- Domain Accuracy & Fine-Tuning — need specialized accuracy beyond frontier APIs
These are the most common patterns we see, but they are not exhaustive. If your inference challenge does not fit neatly into one of these categories, we still encourage you to apply — select “Other” on the application and describe your use case.