Diana Hu, YC managing partner and AR pioneer, says companies are hitting 7-figure revenue in 3 months — a pace that was impossible when she went through YC in 2017

Diana Hu joined Y Combinator as a managing partner after co-founding Escher Reality, an AR SDK company that went through YC's Summer 2017 batch before being acquired by Niantic. That founder-to-investor arc shapes how she reads the current moment.

Revenue pace

The clearest shift Hu identifies is how fast companies now hit meaningful revenue. Reaching seven figures in three months has become achievable for the best companies in recent batches — something she describes as impossible in 2017. Back then, 10% week-on-week growth that compounded to a few hundred thousand in ARR was considered exceptional. The difference, in her view, is agentic coding letting small teams ship more mature products faster, so once a founder finds the right customer and nails the sales motion, they can accelerate through early stages that previously took much longer.

She is careful not to universalize that benchmark. Companies building core infrastructure or defense applications should prioritize deep deployment with a few customers over headline revenue growth, even if the early numbers look modest.

“It has become more common that companies can go from 0 to 7 figures in revenue within just three months. Which when I did the batch, that would have been impossible... The best companies back then would grow, let's say, 10% week on week, which is what Airbnb did back then. And that would get them to couple 100,000 in ARR, which was a very impressive accomplishment back then.”

Platform risk then and now

Hu draws a direct line between the platform risk AR companies faced in 2017 and the question AI founders are wrestling with today. On the first day of her YC batch, Apple shipped ARKit — functionally the same SDK Escher Reality had built. The company's response was to compress its roadmap and build cross-platform support, eventually enabling code to run across Android, iOS, and headsets from a single build. The parallel she draws for today's AI founders is whether their roadmap sits in the path of what the frontier labs are already building.

Consumer and gaming

On consumer opportunities, Hu is honest that the patterns haven't resolved yet. She thinks there will be large category winners, but predicts their shape is as hard to call now as Airbnb or DoorDash would have been in the early 2010s. On gaming specifically, she argues the real bottleneck isn't build complexity — that barrier is clearly lower — it's encoding fun. Game loops that keep players returning are hard to design, hard to measure, and vary game by game. She floats the possibility that agentic game loops could self-improve based on usage data, but says nobody has figured out how to build that reliably yet.