Anna Patterson has had a storied career in Silicon Valley. She founded three startups, including search engine upstarts Xift and Cuil, as well as recall.archive.org, which became the Internet Archive. She was the vice president of engineering at Google, and later started Gradient Ventures, an AI-focused seed fund. And she isn’t done building.
Patterson told TechCrunch that a few years ago she had the itch to start something new again, but was unsure she had another startup in her. But after a breast cancer diagnosis in 2023 and being out of work a lot of that year, she realized she could either go back to her old life or start something new.
It’s clear she chose the latter: Her new startup, Ceramic.ai, says it provides foundational AI training infrastructure that enterprises can use to train large language models faster using fewer GPUs than the “current state-of-the-art.” Ceramic claims its model can use long contexts and work with any clusters, and its goal is to help models scale by 100 times.
Patterson said that she got the idea for the company when she realized that the existing infrastructure for building LLMs had too many variables, and were too complicated for enterprise adoption at scale.
“Whenever you’re trying to stretch an infrastructure 10x, it usually works no problem,” Patterson said. “But when you’re trying to stretch it, you know, 100x or more, you often have to kind of take a step back and rethink it. So, I thought to myself, if this is some infrastructure that we’re going to run for the next 10 years, is this how I would do it?”
That thought process culminated in Ceramic AI in January 2024, which Patterson founded with Tom Costello, the company’s chief scientist. The startup has been operating in stealth since then, and has already signed partners, including AWS and Lambda — though it isn’t generating any revenue yet. Patterson says she wanted to build credibility, awareness, and trust with potential customers before focusing on sales.
The company recently raised a $12 million seed round that was led by NEA and saw participation from IBM, Samsung Next, and Earthshot Ventures. Patterson said NEA was a natural choice to lead the round because of the firm’s technical prowess. The funding will mainly be put toward sales and continued development.
But after three startups, a career’s worth of senior roles, and years as an investor, what’s different this time? Patterson says her approach to building startups has changed since she spent time on the other side of the table as a VC investor. She feels she’s under more time-pressure now compared to when she built her first few companies.
“When I was younger building companies, I felt this, I mean this is kind of sad, but I felt this really relaxing time of like running a just engineering, no-sales company,” she said. “Whereas now, you know after being a VC for a number of years, I definitely feel more time pressured.”
She thinks that is a good thing though, saying it’s better to get your product in front of customers earlier and get feedback and iterate.
That is likely pertinent for Ceramic because it isn’t the only company looking to help enterprises scale their foundational models.
One of its biggest competitors is Together AI, which also looks to help companies “turbocharge” their model building work, and has raised more than $530 million in venture capital. MosiacML also targeted accelerated LLM building and raised $37 million before it was acquired by Databricks in 2023 for $1.3 billion.
For now, Ceramic has its work cut out if it wants to carve out a place for itself in this fast-moving market.