Full Transcript
Loading transcript…
Stanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Applications, Applied AI
Watch with subtitles, summary & AI chat
Add the free Subkun extension — works directly on YouTube.
- Watch
- Subtitles
- Summary
- Ask AI
Founders, engineers, and investors interested in the economics and infrastructure of AI development and deployment.
TL;DR
This video features Tuhin, CEO of Base 10, discussing the exponential growth of AI inference and the critical role of infrastructure in powering AI companies. He highlights how Base 10 optimizes inference for custom AI models, enabling businesses like Whisper Flow and Abridge to achieve superior performance and reliability compared to relying solely on frontier models or generic cloud providers.
Key Takeaways
- Inference is projected to increase by a factor of one billion, signifying a massive scaling in AI computation needs.
- The founder's journey highlights a transition from finance to engineering and entrepreneurship, emphasizing the value of early-stage technology and small teams.
- Bay 10 focuses on production inference, powering high-growth AI companies by optimizing infrastructure for custom machine learning models.
- Companies like WhisperFlow and Abridge leverage Bay 10 for optimized, reliable, and low-latency inference across multiple cloud providers.
- Bay 10's value proposition includes superior performance, multi-cloud reliability, and a robust developer platform, addressing limitations of direct cloud provider solutions.
- While frontier models dominate current inference spend, custom models are seen as key to building profitable and defensible AI businesses.
- Open-source models offer a significant cost advantage, running 70-90% cheaper than frontier models, despite a ~90-day lag in capabilities.
- The AI supercycle is driven by the increasing demand for inference, which serves as the core mechanism for delivering AI value.
In This Video
- 00:09Inference is About to Explode
The speaker predicts a billion-fold increase in inference, introducing the guest who is driving this growth.
- 00:31Tuhin's Entrepreneurial Journey
Tuhin, CEO of Bay 10, shares his winding path from finance to machine learning research and startups.
- 02:07Falling for Early-Stage Tech
After moving to San Francisco, Tuhin fell in love with building products in small teams at early-stage tech companies.
- 02:57Founding Bay 10
Tuhin and co-founders started Bay 10 in 2019 to build infrastructure for the rapidly growing machine learning field.
- 03:56Bay 10's Customer Examples
Bay 10 powers companies like Whisper Flow (speech-to-text) and Abridge (healthcare scribe) with optimized inference.
- 06:16The Future of AI Inference
The thesis is that AI will be massive, with inference as the key value delivery. Custom models are crucial for profitability.
- 07:28Why Choose Bay 10?
Bay 10 offers superior performance, reliability, and a developer platform, solving inference stack pain points.
Questions & Answers
What is the main focus of Ban?
Ban focuses on production inference, powering the fastest-growing AI companies by providing optimized infrastructure for machine learning models.
What is Whisper Flow and what does Ban do for them?
Whisper Flow is a speech-to-text app. Ban runs optimizations and infrastructure for Whisper Flow, ensuring low latency for their voice typing experience.
What is Abridge and how does Ban support it?
Abridge is an ambient scribe for healthcare, integrated with EMRs. Ban runs dozens of their models, including speech-to-text and those for clinical notes, ensuring reliability and speed.
Why do companies choose Ban over cloud providers like AWS or GCP?
Companies choose Ban for performance optimizations, reliability across clouds, and a developer platform that offers flexibility and security, which are difficult to achieve independently on general cloud providers.
What is the economic advantage of using open-source models compared to frontier models?
Open-source models are about 90 days behind frontier models but can be run 70-90% cheaper, making them a cost-effective option for many AI applications.
Key Terms
- Inference — The process of using a trained machine learning model to make predictions or generate outputs.
- Production Inference — Running trained AI models in a live environment to serve end-users or applications, focusing on speed, reliability, and scalability.
- Ambient Scribe — Technology that listens to conversations (e.g., doctor-patient) and automatically generates notes or summaries.
- Frontier Models — The most advanced and capable large-scale AI models available at any given time, often proprietary and computationally intensive.