As AI companies mature, the fight for high-quality data has become one of the most competitive areas in the industry, launching companies like Mercor, Surge, and, most prominently, Alexandr Wang’s Scale AI. But now that Wang has moved on to run AI at Meta, many funders see an opening — and are willing to fund companies with compelling new strategies for collecting training data.
The Y Combinator graduate Datacurve is one such company, focusing on high-quality data for software development. On Thursday, the company announced a $15 million Series A round, led by Mark Goldberg at Chemistry with participation from employees at DeepMind, Vercel, Anthropic, and OpenAI. The Series A comes after a $2.7 million seed round, which drew investment from former Coinbase CTO Balaji Srinivasan.
Datacurve uses a “bounty hunter” system to attract skilled software engineers to complete the hardest-to-source datasets. The company pays for those contributions, distributing over $1 million in bounties so far.
But co-founder Serena Ge (pictured above with co-founder Charley Lee) says the biggest motivation isn’t financial. For high-value services like software development, the pay will always be far lower for data work than conventional employment — so the company’s most important edge is a positive user experience.
“We treat this as a consumer product, not a data labeling operation,” Ge said. “We spend a lot of time thinking about: How can we optimize it so that the people we want are interested and get onto our platform?”
That’s particularly important as the needs of post-training data grow more complex. While earlier models were trained on simple datasets, today’s AI products rely on complex RL environments, which need to be constructed through specific and strategic data collection. As the environments grow more sophisticated, the data requirements become both more intense for both quantity and quality — a factor that could give high-quality data collection companies like Datacurve an edge.
As an early-stage company, Datacurve is focused on software engineering, but Ge says the model could apply just as easily to fields like finance, marketing, or even medicine.
Techcrunch event
San Francisco
|
October 27-29, 2025
“What we’re doing right now is we’re creating an infrastructure for post-training data collection that attracts and retains highly competent people in their own domains,” Ge says.
