Keep pulling the thread on Tuhin Srivastava.
Baseten has grown 30x over the last year.
Baseten is expected to generate more than $1 billion in revenue this year.
Tuhin Srivastava estimates that 99% of the potential enterprise market for AI, measured by inference count, has not yet adopted the technology.
The DeepSeek model can be run in production for approximately 20% of the cost of running models from OpenAI or Anthropic, with comparable or better latency and reliability.
Approximately 95% of the tokens served on Baseten's dedicated inference platform are for custom models that have been modified by customers.
Procuring a cluster of 1,024 NVIDIA B200 GPUs from a top-tier cloud provider currently requires a 3-to-5 year contract with a 20-30% upfront payment of the total contract value.
Baseten has a 400% annual net dollar retention (NDR) rate.
None of Baseten's top 30 customers have ever churned.
In a compute-constrained environment, owning compute capacity is the most important strategic asset.
Tuhin Srivastava predicts that no competitor will be able to effectively challenge NVIDIA's market position in the next couple of years.
Tuhin Srivastava believes there will not be enough compute capacity to meet the demand for AI model capabilities over the next five to ten years.
Tuhin Srivastava believes the AI inference market is the "last market," suggesting that even with the advent of AGI, the primary economic activity will be inference.