Together AI Unveils Enhanced NVIDIA GPU Clusters at GTC 2025
In a significant announcement at the NVIDIA GTC 2025, Together AI has unveiled new advancements in GPU technology, showcasing the deployment of NVIDIA Blackwell GPUs and the introduction of Instant GPU Clusters. These developments are expected to significantly enhance AI performance and scalability for enterprises, according to together.ai.
Enhanced GPU Capabilities
Together AI, as an NVIDIA GTC Gold Sponsor, is rapidly deploying NVIDIA Blackwell GPUs at scale, promising next-generation AI performance. Early adopters, including Zoom, Salesforce, and fal.ai, have reported a twofold increase in training and inference speeds using the NVIDIA HGX B200 on the Together AI platform. This leap in performance is attributed to the advanced architecture of the Blackwell GPUs, which are set to accelerate AI capabilities even further.
Introduction of Instant GPU Clusters
The company has also launched the Preview release of its Together Instant GPU Clusters. These on-demand clusters can host up to 64 NVIDIA GPUs, interconnected via NVIDIA Quantum-2 InfiniBand and NVLink. Designed for rapid deployment, these clusters can be provisioned in minutes, providing AI teams with powerful infrastructure to meet immediate computational demands.
Expanding AI Frontiers
Together AI's efforts are part of a broader strategy to provide businesses with open AI infrastructure that is both flexible and high-performing. Customers like Cartesia and DuckDuckGo are leveraging these capabilities to push the boundaries of AI innovation. The company emphasizes that its infrastructure allows enterprises to maintain control over their models and data without the limitations of closed AI platforms.
Scalable AI Solutions
Together AI offers three distinct GPU cluster solutions tailored to varying levels of AI compute demand:
- Instant GPU Clusters: Up to 64 NVIDIA Hopper GPUs, available for rapid access and high-performance workloads.
- Dedicated GPU Clusters: Ranging from 64 to 1,000 GPUs, optimized for large-scale training and inference.
- Custom GPU Clusters: Designed for hyperscale deployments, supporting up to 100,000+ NVIDIA GPUs.
Partnership with NVIDIA
Through strategic collaboration with NVIDIA, Together AI ensures that its infrastructure is optimized for peak performance. The integration of NVIDIA AI Enterprise provides a production-ready software stack, enhancing model training efficiency and overall workload performance.
As the AI landscape continues to evolve, Together AI's innovations in GPU technology and infrastructure are positioned to support the next generation of AI research and applications, empowering enterprises to achieve unprecedented levels of AI performance and scalability.