Data platform giants like Databricks and Snowflake do great when it comes to building data pipelines and running low-latency analytics to generate AI features, but they don’t solve the need for fresh data and complex compute requirements at AI inference time. That’s according to Chalk, the AI startup that today announced it has raised $50 million to build AI inference data pipelines.
Chalk was founded in 2022 by three engineers, Marc Freed-Finnegan, Elliot Marx, and Andy Moreland, to develop a real-time data platform for AI inference. The trio had experience building AI systems at startups like Affirm, Haven (acquired by Credit Karma), and Index (acquired by Stripe), as well as industry giants like Google and Palantir, and saw a wider need for better AI inference systems.
The engineers developed the Chalk data platform with a special focus on speeding up the AI inference process and delivering access to “ultra-low latency” data to power AI apps, such as detecting identity theft, qualifying loan applicants, boosting energy efficiency, and moderating content.
Developers interact with the Chalk platform by declaring machine learning features in Python, which is then executed in parallel feature pipelines atop a Rust-powered compute engine. This engine then “resolves features directly from the source” at inference time, which eliminates stale data and brittle ETL data pipelines of existing AI data platforms while also improving latency.
Over the past three years, Chalk’s unique approach to AI inference has attracted a number of customers, including Doppel, Nowst, Sunrun, Whatnot, Socure, Found, Medely, and iwoca, among others. The San Francisco company has been particularly successful at helping customers in the financial services industry build AI inference pipelines.
“Chalk helps us deliver financial products that are more responsive, more personalized, and more secure for millions of users,” stated Meng Xin Loh, a senior technical product manager at MoneyLion. “It’s a direct line from infrastructure to impact.”
“Chalk has transformed our ML development workflow. We can now build and iterate on ML features faster than ever, with a dramatically better developer experience,” stated Jay Feng ML Engineer at Nowstaw. “Chalk also powers real-time feature transformations for our LLM tools and models–critical for meeting the ultra-high freshness standards we require.”
When the co-founders started Chalk, they knew real-time inference was critical for fintech, said Marc Freed-Finnegan, Chalk’s CEO. “Over the years, we’ve discovered that its importance extends far beyond fintech–to identity verification, fraud prevention, healthcare, and e-commerce,” he wrote in a blog post today.
With a few notches on its AI inference belt, Chalk is now ready to scale up operations and make some more noise in the space. In particular, Chalk sees the large data platform like Snowflake and Databricks being susceptible to the market’s shift away from AI training towards AI inference.
Chalk co-founders (left to right): Elliot Marx, Marc Freed-Finnegan and Andrew Moreland
“AI compute is shifting rapidly from training to real-time inference, creating new demands for fresh data and complex computations at the exact moment decisions are made,” Freed-Finnegan wrote. “Existing solutions have enabled large, complex training workflows and feature stores (low-latency caches of pre-processed data), but real-time inference remains underserved.”
The CEO says Chalk addresses this gap “by providing infrastructure designed explicitly for instantaneous, intelligent decisions. “Our mission remains clear: to deliver intuitive, powerful data infrastructure that integrates seamlessly with developers’ favorite tools,” he says.
Aydin Senkut, the founder and managing partner at Felicis, one of the venture capital firms that led Chalk’s Series A round, said that Chalk is poised “to become the Databricks of the AI era.”
“It’s one of the fastest-growing data companies we’ve ever seen,” Senkut stated. “The team has fundamentally redefined how data moves through the AI stack, a crucial advancement for chain-of-reasoning models. What’s even more remarkable is Chalk’s ability to deliver 5-millisecond data pipelines at massive scale–something that, until now, was considered out of reach.”
The Series A round, which included participation by Triatomic Capital and existing investors General Catalyst, Unusual Ventures, and Xfund, valued Chalk at $500 million. That’s about what Databricks was valued around 2017, just before the company embarked upon a remarkable string of venture-fueled growth. As it raked in billions in venture money from 2018 through 2024, Databricks’ annual recuring revenue also grew, from about $100 million in 2018 to about $3 billion in ARR at the end of 2024, when the company brought in a whopping $10 billion Series J round at a valuation of $62 billion.
Will Chalk ever reach those great heights? Only time will tell.
Related Items:
Future Proofing Data Pipelines
Cutting-Edge Infrastructure Best Practices for Enterprise AI Data Pipelines
How to Build a Better Machine Learning Pipeline
