Startups Are Moving From Bits to Atoms
What this video is
⚡ a 36-minute video, readable in 60 secondsThis is YC partner commentary (Garry Tan and Jared Friedman are cited) on trends across recent Y Combinator batches: a resurgence of hard tech, defense, robotics, and space startups; an emerging AI 'harness war' between system-of-record incumbents like Salesforce and coding-agent tools; and a fast-growing wave of startups selling data or RL environments to AI labs. It matters because it quantifies how agentic coding is compressing startup revenue timelines and reshaping which sectors and founder types YC is funding.
[01:19] YC's hard tech companies grew from 8% to 20% of the batch over the last 12 months
Key takeaways
- [01:40] Median YC company revenue by end of batch rose from about 8K to 20,000 monthly, up from zero at acceptance
- [02:16] Within the batch, robotics grew from 1% to 6-7%, industrial manufacturing from 4% to 10%, and defense from 1.5% to 5%
- [10:08] An Nvidia A100 GPU's hourly rental cost is now appreciating rather than depreciating with age, as AI compute demand outstrips supply
- [13:20] A benchmark reportedly showed Pi robotics' Astra improving from about 10% to 60-70% task completion
+ 7 more takeaways
- [16:04] Salesforce has released its own Slack AI harness, framed as the start of a new 'harness wars' against Codex, Claude Code, OpenCode, and Hermes
- [17:37] An OpenAI post showed GPT-5.6 underperformed on ARC-AGI-3 until two API settings tripled scores with 6x fewer tokens, and a custom harness reportedly pushed it past 90%, up from low double digits a couple months earlier
- [18:00] Companies doing full-stack end-to-end work rose from 10% to over 25% of the batch, alongside median MRR growth to about 20K
- [22:00] Some YC companies are now going from zero to seven-figure revenue within the 3-month batch, versus roughly 18 months previously
- [23:15] YC has funded more than a dozen companies in the last two years each making over $10 million a year selling data or RL environments to AI labs, some making hundreds of millions
- [25:02] Big AI labs are reportedly spending about a billion dollars on RL environments
- [29:49] Solo founders made up about 5% of accepted YC companies a year ago; that figure is now 18-19%
Jump to a moment
0:00— Intro1:06— Startups Are Moving From Bits to Atoms3:29— Why AI Is Making Hard Tech Easier5:21— Defense, Manufacturing, and the Return of Hard Tech9:48— AI Compute Is Becoming a Physical Infrastructure Problem12:34— Robotics Is Approaching Its ChatGPT Moment15:40— Software Isn’t Dead. It’s Becoming the Harness17:59— Why AI Startups Are Growing Faster22:52— The Hidden Boom in Data and RL Environments27:08— Why Robotics Will Need Specialized Models29:38— The Rise of the Solo Founder32:54— Why Experienced Founders Are Back34:59— What Founders Should Do Right Now
Their links, sorted & clickable
🤝 Work with them1
Apply to Y Combinator🔗 Other links1
Work at a startupWas this brief useful?