The First Dedicated YC GPU Cluster - With Together AI
YC and Together AI have partnered to launch the first dedicated YC GPU cluster, addressing the critical compute bottleneck faced by AI-native startups. This initiative provides flexible, cost-effective access to GPU resources, enabling companies from early-stage research to major players to train, fine-tune, and run inference on AI models, and mitigating the financial strain of long-term compute commitments.
This discussion introduces the partnership between Y Combinator (YC) and Together AI, establishing the first dedicated YC GPU cluster to provide AI-native startups with essential compute resources. Vipul Ved Prakash, CEO and co-founder of Together AI, explains that the company, founded four years prior by academics, aimed to create a platform for innovation in generative AI, covering the entire lifecycle from model building and post-training open models to large-scale serving.
Together AI’s Broad Customer Base and Services
Together AI supports over 8,000 customers, ranging from small research groups in early experimentation phases to major industry players like Cursor, Cognition, and ElevenLabs. These companies utilize Together AI for training models and serving them at scale. Ankit Gupta from YC highlights the diverse landscape of AI-native startups in the YC portfolio, including those training models from scratch, AI-native services built on open-source models, data providers, and application-layer companies. This variety leads to heterogeneous compute needs, which Together AI’s flexible offerings are designed to meet, helping companies efficiently deploy capital for rapid iteration.
Evolution of AI Company Building and the Compute Bottleneck
Building an AI company today differs significantly from 2018. Ankit notes that many traditionally non-software industries (e.g., pharma, healthcare) are now operating as software-first entities due to AI, expanding the scope for YC startups. However, this shift introduces new bottlenecks, particularly in compute. Previously, acquiring GPU instances from cloud providers like AWS was relatively straightforward. Now, the sheer access to capacity, let alone favorable pricing, is a major challenge. The increasing value of tokens and the associated computational flops mean that the cost of compute continues to rise, making secure and cost-effective access a critical differentiator.
YC’s Commitment to Research-Focused AI Startups
YC has become a significant seed funder for research-oriented AI companies. Examples include the original OpenAI (founded as YC Research), Deepgram (voice AI), Vappy (distribution), Miso Labs (emotive voice models), and companies in biology, healthcare, and robotics. Many of these are research bets requiring hard technical problem-solving, where compute access is a primary bottleneck. YC views securing compute as crucial for attracting top founders and enabling their success during and after the program.
Rationale for the YC-Together AI Partnership
YC chose Together AI for several reasons:
- Speed and Flexibility: Together AI’s ability to rapidly deploy a cluster and support diverse needs, from single nodes for scaling startups to companies requiring 256 GPUs on day one, was key.
- Engineering and Research Expertise: Together AI’s strong engineering and research team provides best practices and support, helping founders navigate complex compute cluster management. This ensures that startups can focus on their core product rather than becoming compute infrastructure experts.
- Deep Research: Ankit acknowledged Together AI’s long-standing research contributions (e.g., Flash Attention, Mamba architecture), which directly benefit their product offerings.
Together AI’s Research and Production AI Focus
Vipul explains that Together AI’s research focuses on the “science of production AI” – optimizing the underlying linear algebra and novel accelerators. This is critical for efficiency in a capital-intensive domain where training and token production costs are high. Their work on attention mechanisms (like Flash Attention, which optimizes large context windows) and new architectures (like Mamba, used in models such as Neotron) aims to accelerate workloads and achieve the best possible unit economics. This research provides a significant advantage to both early-stage startups and scaled companies using their platform.
Compute Planning Advice and Solving Financial Bottlenecks
Vipul advises early-stage companies to leverage the vast infrastructure, open models, tooling, and datasets available today for efficient development. However, compute planning remains complex. The partnership aims to address a critical financial hurdle: the initial capital outlay for securing compute capacity. Many startups found that the upfront cost for a 24-month compute commitment often exceeded their current cash balance. The YC-Together AI cluster allows startups to make shorter commitments (a few months) and plan their compute needs several months in advance, easing cash flow constraints and enabling more effective capital deployment.
Current Impact and Future Vision
YC companies using the cluster have provided positive feedback, appreciating the support team, optimized workflows, and the ability to plan compute needs in advance. This flexibility allows companies to book varying amounts of compute for different phases (e.g., small for initial work, large for a training run three months later). The cluster maintains 100% utilization across the portfolio while offering individual companies the flexibility they need. Both parties are excited about expanding the partnership, viewing it as essential for AI startups. Ankit emphasizes YC’s role as a “founders union,” providing shared resources like the GPU cluster to give startups the leverage typically associated with larger companies, empowering them to succeed regardless of their technological focus.