Substrate lands $8M funding to bring ‘Lego blocks’ approach to enterprise AI

Substrate, a startup founded by tech industry veterans Rob Cheung and Ben Guo, quietly emerged from stealth last week to launch its artificial intelligence development platform. The company also announced it has raised $8 million in a funding round led by Lightspeed Venture Partners to grow its team and expand its product offerings.

Substrate aims to democratize AI by providing a unified platform for enterprises to build, deploy, and manage machine learning models and pipelines. Its flagship offering is an API that enables developers to create complex AI workflows by stitching together high-quality open-source models that have been curated and optimized by Substrate.

The company believes its platform will make it significantly easier and more cost-effective for enterprises to harness the power of advanced AI capabilities like large language models (LLMs) and other generative AI techniques. This could accelerate the adoption of AI in industries ranging from content creation to business analytics to customer support.

Breaking down complex problems into manageable pieces

“The main problems right now with integrating the current generation of AI, and LLMs in particular, are accuracy, cost, and latency,” explained Rob Cheung, co-founder and CEO of Substrate, in an interview with VentureBeat. “Substrate addresses all three by enabling developers to break down a big complex problem into many smaller, more constrained problems that are easier to solve.”

Cheung drew an analogy to how Google Search works behind the scenes, parsing queries using 15-20 different machine learning models operating in concert. “All of that is jammed into one giant prompt, you might get good answers 10% of the time, but 90% of the time you’re really in the dark about how that giant prompt is going to steer it,” he said. “If you break it down, and you have a good way to run a broken down description of the problems, that solves a lot of the accuracy issues.”

While tech giants like Google have built extensive infrastructure to optimize and orchestrate huge numbers of ML models, most companies lack those capabilities. “We think it makes a lot of sense to centralize all that performance optimization work in one place and offer it as a service, because that’s really what people want to do,” said Cheung. “One of our big customers, Substack — they’re really just not interested in running machine learning infrastructure. They want Lego blocks to build out their ML workload and just have it work.”

Curated models and better abstractions boost productivity

Substrate co-founder Ben Guo said the company’s experiences with early customers like Substack, which used the platform to generate summaries and topic categories for blog posts, demonstrate the value of its approach. “It enables them to use all their models in one place, running on one cluster, which enables much faster run speeds for pretty big workloads, as well as lower costs and better reliability,” he explained.

In addition to performance benefits, Guo believes Substrate’s curated set of plug-and-play models will appeal to enterprises that don’t want to wade through the constantly growing landscape of open-source AI. “Part of what people want is for us to read the literature, cut through the noise, and pick the most interesting and useful models as they come out,” he said.

Substrate also aims to create a better developer experience by providing simple abstractions and templates for common enterprise use cases. “We’re taking a step back, looking at the landscape and trying to figure out the platonic ideal for these abstractions, which I think nobody is really doing right now,” Guo told VentureBeat. “It’s kind of like what I learned at Stripe — there’s a lot of hidden value in creating very simple abstractions,” like enabling a payments integration in just seven lines of code.

Bringing the cloud platform playbook to enterprise AI 

As large language models and other AI building blocks become more powerful and accessible, platforms like Substrate could play a key role in helping enterprises translate those raw capabilities into real-world applications and business value. With a more abstracted, full-stack approach to AI development, Substrate aims to do for machine learning what cloud platforms did for general-purpose computing — making it simpler and more economical for companies to build and deploy powerful software.

Substrate’s $8 million funding round will enable it to expand its platform, grow its team, and scale up its go-to-market efforts to reach more enterprise customers. With seasoned founders and strong early traction, the startup seems well-positioned to become a major player in the rapidly evolving world of enterprise AI.