How YC Startups Can Access Flexible GPU Compute

TL;DR

YC and Together AI are launching a dedicated GPU cluster that gives YC startups faster access to compute, favorable pricing, technical support, and commitments measured in weeks rather than years. The partnership supports varied workloads, from single-node experiments to deployments involving hundreds of GPUs, while helping founders train, post-train, and serve generative AI models at scale.

Transcript

[music] Hey everyone, I'm here with Bippol, CEO and co-founder of Together AAI and we've got some exciting news. YC and Together are partnering to bring online the first dedicated YCGPU cluster. This gives our portfolio of AI native startups easier access to the compute they need to build and scale. Vipple, thanks for being here. >> So great to be ... Read More

Key Insights

• YC and Together AI are launching the first dedicated YC GPU cluster to give portfolio companies easier access to the computing capacity needed for building and scaling AI products, especially when general cloud capacity is difficult to secure.
• Together AI is a cloud platform covering the generative AI lifecycle, including building models, post-training open models, and serving models at scale. Its stated purpose is to create a platform where a broader group can participate in AI innovation.
• Together AI has more than 8,000 customers, ranging from small research groups still defining their projects to established AI-native companies such as Cursor, Cognition, and ElevenLabs that train models or rely on scaled model serving.
• AI-native startups have heterogeneous compute requirements because some train models from scratch, others build services on hosted open models, and still others operate across data and application layers. A flexible provider can allocate resources according to each workload.
• Compute capacity is a major startup bottleneck because companies can no longer assume that large numbers of GPU instances will be readily available on demand. Access to capacity can matter alongside price when model and workload requirements expand.
• The YC cluster offers commitments lasting only a few weeks rather than two years, according to the discussion. This arrangement helps startups conserve capital and adjust capacity as their experimentation, training, and scaling needs change.
• YC chose Together AI partly because the cluster can accommodate markedly different starting points, from a company needing one node immediately to another purchasing 256 GPUs on its first day and anticipating growth to several thousand.
• Engineering support is important because founders who previously used managed compute may not have operated their own clusters. Together AI can share operational best practices and reduce the amount of infrastructure knowledge teams must develop entirely from scratch.

Questions & Answers

Q: Why are YC and Together AI creating a dedicated GPU cluster?

YC and Together AI are creating the cluster to make computing capacity easier for YC startups to obtain as they build and scale AI systems. The partnership addresses constraints in availability, pricing, support, and contract length. It is designed for companies with varied requirements, including early research teams, model developers, and application businesses whose resource needs can change quickly.

Q: What services does Together AI provide to AI companies?

Together AI provides a cloud service designed for the full generative AI lifecycle. Its capabilities include helping customers build models, post-train open models, and serve models at scale. The platform supports both access to compute and inference services, allowing customers at different stages to experiment, develop their systems, train models, and operate those models for users.

Q: What types of customers use Together AI?

Together AI says it serves more than 8,000 customers. They range from small research groups that are still experimenting and deciding what to build to prominent AI-native companies such as Cursor, Cognition, and ElevenLabs. Customer workloads include obtaining compute, accessing inference services, training models, and having Together AI serve models at scale on their behalf.

Q: Why has GPU access become a bottleneck for AI startups?

GPU access has become a bottleneck because startup compute requirements are increasing while capacity is harder to obtain. The discussion contrasts current conditions with 2018, when a company could obtain around a thousand GPU instances from AWS without significant difficulty or long reservations. Modern startups must consider actual availability as well as pricing, support, and commitment length.

Q: How does the YC GPU cluster help startups manage capital efficiently?

The cluster helps startups manage capital by combining secured capacity and favorable pricing with shorter commitments. According to the discussion, startups can commit for a few weeks instead of two years. That flexibility lets teams direct spending toward the workloads they currently need, iterate rapidly, and avoid locking scarce capital into capacity before their technical and commercial requirements are clear.

Q: What range of compute needs can the dedicated cluster support?

The dedicated cluster is intended to support widely different compute requirements across YC companies. Some startups need only a single node and immediate access while they begin scaling. Others may seek 256 GPUs on their first day and expect to expand to several thousand. The partnership aims to support these cases through one flexible infrastructure and service relationship.

Q: Why did YC choose Together AI as its cluster partner?

YC chose Together AI because the company prioritized speed, could bring a cluster online quickly, and could support startups with substantially different workloads. YC also valued Together AI's engineering and research team, which can teach infrastructure best practices to founders who have trained models on managed systems but have not necessarily operated their own computing clusters.

Q: Why is technical support important for AI startup infrastructure?

Technical support matters because strong model engineers do not always have experience managing compute clusters themselves. Some founders previously trained models where infrastructure was already managed for them. Together AI handles parts of the environment directly and helps founders understand the portions they must manage, reducing the difficulty of adopting cluster operations and encoding established best practices.

Summary & Key Takeaways

• YC and Together AI are bringing a dedicated GPU cluster online for YC startups. The partnership addresses a growing constraint for AI-native companies: obtaining enough compute capacity without accepting long commitments. Participating startups can access favorable pricing, technical support, and infrastructure suited to both early experimentation and rapid growth.

• Together AI provides cloud services across the generative AI lifecycle, including model building, post-training of open models, and large-scale serving. Its more than 8,000 customers range from small research groups to companies such as Cursor, Cognition, and ElevenLabs, demonstrating the variety of workloads that its platform supports.

• YC chose Together AI because the provider emphasized deployment speed, flexible allocation, and engineering support. The cluster can serve startups seeking one node as well as companies requesting 256 GPUs initially. This flexibility helps research-focused founders use capital efficiently while tackling technical problems across software, biology, healthcare, and robotics.