T
Together AI operates a purpose-built GPU cloud for model training, fine-tuning, and inference, with a tech stack centered on PyTorch, Kubernetes, and custom inference engines (vLLM, SGLang, Triton). The adopting list reveals heavy infrastructure investment—moving into RoCEv2, InfiniBand, and Ceph—while projects focus on distributed GPU scheduling, multi-petabyte storage systems, and cost optimization, indicating they're scaling toward larger, multi-region deployments and addressing the operational complexity of managing GPU fleets at cloud scale.
Tech Stack 138 technologies
Core StackPython PyTorch TypeScript Go Cypress GitHub Actions AWS Rust Java Terraform Pulumi Stripe Kubernetes Apache Airflow Kafka FlashAttention k6 Argo CD AWS EKS CUDA Triton AWS CDK Kinesis C/C++ vLLM SGLang AWS Lake Formation IAM VPC TeamCity+102 more
What Together AI Is Building
◆Challenges
- Cost optimization for storage
- Scaling data processing infrastructure
- Lower the cost of modern ai systems
- Handling millions events daily
- Scaling physical backbone
- Scalability challenges
- Cost efficiency
- Vendor management
- Cost optimization across major infrastructure builds
- Optimizing autoscaling
▲Active Projects
- Efficient inference and rl‑driven training
- Runtime inference services at scale
- Usage-based billing system
- Run complex demonstrations and pocs of together’s entire stack
- Global data center strategy development
- Sales enablement tools
- Design multi-petabyte ai/ml storage systems
- Build kubernetes-native storage operators
- Implement intelligent prefetching and model-weight distribution
- Distributed gpu scheduling system
Hiring Activity
75 roles · 20 in 30d
Department
Seniority
Company intelligence
