Why SPC Backed Baseten
Powering a multi-model future
Ruchi Sanghvi · June 22, 2026
When Tuhin and Phil first walked into SPC in early 2019, the AI revolution was still in its infancy. A few dozen founders and researchers at SPC were reading linear algebra textbooks, discussing the scalability of transformer-based approaches, and exploring the potential of machine learning (as it was called at the time). But the new models at the time, BERT and GPT-2, still felt more like toy demos than something that a real business would use.
Like most great founders, Tuhin and Phil saw a different future. They played with those demos and saw possibilities. They saw these models bringing intelligence and customization to workflows that previously lacked it. They envisioned a future where every product-building company would train fleets of specialized models to serve as core differentiators of their products. Alongside their former colleague Amir, they started whiteboarding out dozens of ideas for how to build for this future. At the top of their list: an infrastructure stack to enable turnkey ML infrastructure and model customization. We signed the term sheet the day after the company incorporated.
The journey to the multi-model future was not always smooth. Closed source model releases dominated the headlines and the industry collectively stopped believing in a multi-model future. Tuhin, Phil, and Amir didn't forget, they kept building. Then came Stable Diffusion, followed quickly by Whisper. The Llama family started delivering frontier performance, and Deepseek brought reasoning to the open-weight community. Enterprises began scrambling to use and customize these models, but they ran into a mountain of infrastructure challenges.
I had experienced the pain of previous platform shifts. At Facebook, then Dropbox, I watched cloud go from curiosity to the default. We felt the pain of moving workloads to the cloud, looked for new vendors who could help absorb the infrastructure pain, and realized we should have moved sooner. I know that it's not easy to absorb the pain of infrastructure during a platform shift, especially when AI is the biggest platform shift in history.
For the last 6+ years, Baseten has delighted customers by absorbing the pain of infrastructure and delighting them with production-ready models. They handle the parts of running inference that are almost impossible to do well in-house: GPU allocation, autoscaling, and bin-packing of dozens of model variants across heterogeneous hardware. All while adapting their infrastructure to a rapidly changing AI landscape. It's no surprise that hundreds of companies, including our portfolio companies like Gamma and Rox, are happy customers.
That brings us to today. We are at an inflection point. The AI platform shift is well underway and inference will be the largest market ever created. The inevitability of a multi-model future is becoming more clear.
Baseten is leading us towards this future. One where every company can develop and run specialized models that truly understand their business. One, as Tuhin says, where "intelligence should be owned by all of us." Much like it was in 2019, this future is clear to Tuhin, Phil, Amir, Pankaj, and the team. That's why we are proud to back Baseten once again in this round.
Baseten is hiring across engineering, research, and GTM. If you are excited to build a multi-model future, join the team!
