Inside the a16z Startup Attacking AI’s $400B Spending Problem [PNvKCxJ0B33]
Granola is the AI notepad for professionals in back-to-back meetings. Get your first month of Granola free: In this episode, I go behind the scenes with Inference.net. Founded by Sam Hogan Heutmaker, Ibrahim Ahmed and Amarjot Singh, Inference has raised an $11.8M Seed from Multicoin Capital and a16z CSX to help companies build specialised AI models that are faster, cheaper and purpose-built for their products. AI adoption is accelerating at a remarkable pace, but as more businesses build AI-powered products, they’re discovering that running those models at scale can become incredibly expensive. Many companies are paying frontier AI prices for workloads that don’t actually require frontier intelligence, creating an opportunity for a new generation of AI infrastructure companies. The team at Inference believes there’s a better approach. Rather than relying entirely on third-party APIs, they help companies distil large foundation models into smaller, specialised models that customers can own, optimise and deploy themselves. Their bet is that companies won’t want to rent intelligence forever. As AI becomes core infrastructure, more businesses will want to own the models powering their products. That long-term vision goes well beyond reducing inference costs. By owning the model, businesses gain greater control over one of the most important parts of their technology stack, without being completely dependent on a handful of frontier AI providers. Inside this episode: - Why Inference believes many companies are paying frontier AI prices for workloads that don’t require frontier models - How model distillation turns large foundation models into smaller, specialised models built for specific use cases - The origin story of Inference, from open-source projects to raising an $11.8M Seed - Inside a product planning meeting as the team builds new observability tools - The team’s daily stand-up as they build, test and ship new features - How Inference is getting its products in front of customers - Why the team believes observability is the gateway to custom model training - Sam Hogan’s advice for founders building startups in one of the fastest-moving technology markets in history But if Inference's thesis proves true, how quickly can a seed-stage startup wedge itself into one of the most competitive and heavily funded layers in all of AI? Check out Inference: Join my founder newsletter: Let’s connect: X: Instagram: LinkedIn: Episode Chapters: 00:00 Introduction 1:58 Interview with Sam Hogan Heutmaker (Founder, CEO) 3:40 Analysis of Inference.Net 7:51 Granola, the AI meeting assistant 8:54 Product Meeting 12:40 Standup Meeting 13:36 Interview with Ibrahim Ahmed (Co-Founder, CTO) 15:44 Interview with Mike Pollard (Founding GTM Engineer) 17:06 Sam’s advice for founders