Before ChatGPT: Titan ML’s Early Bet on Deep Learning Infrastructure
Timing is everything in startups, but only in retrospect. In a recent episode of Category Visionaries, Meryem Arik shared how Titan ML bet on deep learning infrastructure when most investors thought the market was too small – and what happened when ChatGPT changed everything overnight.
The Early Signs
Back in 2021, Titan ML saw signals that others missed. “We had a very strong intuition when we started it a couple of years ago that deep learning would be huge,” Meryem recalls. “And that inference would be a really key part of the problem.”
This wasn’t just wishful thinking. They observed real adoption patterns: “Back then were working with computer vision models and were working with older style transformer models, like but springs. And back then there were definitely use cases of where people are using these. So, for example, the financial services were using Bert style models a huge amount.”
Navigating Investor Skepticism
Despite these early signals, convincing others wasn’t easy. “We had investors telling us that they didn’t think that NLP and language AI would be a big enough market,” Meryem shares. This skepticism came “from the investor side, who are meant to have a lot of forethought.”
Rather than wait for market validation, Titan ML focused on solving concrete problems. Their thesis was simple: “ML engineers are spending way too much of their time on building infrastructure rather than solving the problems that are really core to their business.”
The ChatGPT Acceleration
Then everything changed. “I think what we found in the beginning half of 2023 is that businesses wanted to, quote unquote, ‘do AI,’ but hadn’t figured out how to do it,” Meryem explains. “So there was just like a bit of a, almost a panic.”
The challenge shifted from proving market size to helping businesses implement effectively. “Now we’re starting to see businesses have methodologies and ways that they can prove AI value, and now we’re just working on kind of capturing that demand.”
Adapting the Go-to-Market Motion
The rapid market evolution required quick adaptation. “The things that we expected to see within about five years came within about a year, which is fantastic,” Meryem notes. But this acceleration created new challenges.
Their educational approach became even more critical. “What we have found has worked really well with developers is less salesy and more educational piece,” Meryem shares. “We are very happy to talk to our clients about open source alternatives to what we’re doing or how they can build it themselves or the underlying technology underneath it.”
Looking Forward
The vision hasn’t changed, but the timeline has compressed dramatically. “We would like Titan ML infrastructure and Titan ML technology to be a core and probably invisible part of almost every single LLM deployment and generative AI deployment,” Meryem explains.
For founders building in emerging markets, Titan ML’s experience offers valuable lessons. Having strong conviction in your thesis matters more than early market validation. But when the market does catch up, being prepared to scale quickly becomes crucial.
The key is maintaining focus through both the skepticism and the hype. As Meryem puts it, their goal remains consistent: making infrastructure “analogous to deploying a database when you don’t even really think about it, and it’s just kind of part of your stack.”
In the end, being early isn’t about predicting the future perfectly – it’s about being ready when the future arrives faster than anyone expected.