Y Combinator's The State of Startups in 2026: skim's analysis identifies 11 key moments, with 1 potential conflict of interest flagged. Y Combinator partners Diana, Garry, Jared, and Harj break down internal data from recent startup batches, revealing a surge in hard tech companies from 8% to 20%, median end-of-batch monthly revenue jumping from $8,000 to $20,000, and solo founders rising to nearly 19%.
Category: Business. Format: Panel Discussion. YouTube video analyzed by skim.
skim AI Analysis
Credibility assessment: Insider Batch Data. The hosts share direct proprietary metrics from recent Y Combinator cohorts, offering first-hand data on startup revenue, founder demographics, and sector breakdowns. However, these figures are self-reported internal statistics presented alongside promotional enthusiasm for their own portfolio companies.
Bias assessment: YC Ecosystem Bullishness. As Y Combinator partners, the hosts naturally champion the startup model, their funded portfolio companies, and the transformative upside of AI and hard tech, ending with an explicit call to apply to their accelerator.
Originality: 82% — Proprietary Cohort Metrics. Rather than offering generic tech commentary, the panel brings concrete internal batch numbers on hard tech growth, median revenue acceleration, solo founders, and stealth RL environment startups.
Depth: 78% — Cross-Stack Synthesis. The discussion connects macro trends across physical infrastructure, silicon precision, optical interconnects, SaaS system-of-record moats, and agentic workflows, grounding claims in specific startup examples.
Key Points (11)
1. Diana Shares Surge in Hard Tech and Batch Revenue
Timestamp: 00:01:01 to 00:03:29
Analyzing companies accepted over the past 12 to 18 months, Diana reports that hard tech startups rose from 8% to 20% of the Y Combinator batch while median monthly revenue by batch end jumped from $8,000 to $20,000. Physical categories like robotics, US manufacturing, defense, semiconductors, and power have tripled or quintupled.
Significance (High): Demonstrates a structural shift in early-stage venture away from pure software toward physical infrastructure, accompanied by faster initial monetization.
Sources in support: Diana (Co-host and Partner at Y Combinator), Garry (Co-host and Partner at Y Combinator)
2. PhD Founders and AI Codegen Accelerate Hard Tech
Timestamp: 00:03:29 to 00:05:21
Jared notes that one in six founders in the current summer batch holds a PhD, a sharp rise driven by research-heavy fields like silicon photonics. Garry and Jared add that AI code generation and smart models accelerate scientific research and let hardware startups build complex software with tiny teams instead of hiring thousands of engineers.
Significance (High): Lowers the capital and headcount barrier for full-stack hardware startups, making deep tech economically viable for small founding teams.
Sources in support: Diana (Co-host and Partner at Y Combinator), Garry (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator)
3. Space, Defense, and Domestic Metal Supply Chains
Timestamp: 00:05:21 to 00:09:48
Driven by SpaceX's IPO and modern drone warfare, founders are building satellite startups like Exosat and Beyond Reach Labs alongside defense firms like Icarus and Nine Mothers. To supply these new defense startups moving too fast for legacy vendors, Knox Metals is reviving domestic metal manufacturing in empty Detroit factories at software-like growth rates.
Significance (Medium): Shows how a wave of defense and aerospace startups creates secondary demand for modernized domestic industrial suppliers.
Sources in support: Diana (Co-host and Partner at Y Combinator), Garry (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator)
4. AI Compute Demand Sparks Silicon and Optical Tech
Timestamp: 00:09:48 to 00:12:34
Diana explains that surging AI compute demand has caused older Nvidia A100 hourly prices to appreciate, spurring startups across data center construction, power, and custom chips like Lamb Labs and Ternary-based Bot. Jared highlights Dipole Labs, which builds fully optical photon switches to eliminate electronic routing bottlenecks between GPUs.
Significance (Medium): Illustrates how LLM characteristics like lower floating-point precision and GPU communication bottlenecks open opportunities for semiconductor and photonics challengers.
Sources in support: Diana (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator)
5. Robotics Gains Momentum and VC Appetite Returns
Timestamp: 00:12:34 to 00:15:05
The panel notes that robotics is approaching a breakthrough as new models jump from roughly 10% to 60-70% on task benchmarks. After a decade of venture capitalists favoring only B2B SaaS, investor appetite flipped rapidly toward hard tech during the winter batch when SaaS stocks dipped and Claude Code surged.
Significance (Medium): Solves a long-standing financing bottleneck for Y Combinator hardware startups that previously struggled to raise downstream capital.
Sources in support: Diana (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator), Harj (Co-host and Partner at Y Combinator)
6. Systems of Record Must Become AI Agent Harnesses
Timestamp: 00:15:05 to 00:17:59
Despite earlier SaaS fears, Salesforce and Snowflake posted strong recoveries because agents still rely on systems of record. Garry argues that systems of record like Slack must evolve into AI harnesses where agents and humans execute work, or risk losing their data moats through protocols like MCP as the AI harness wars heat up.
Significance (High): Defines the survival strategy for incumbent enterprise software: become the active execution harness for agents or be commoditized into a passive database.
Sources in support: Diana (Co-host and Partner at Y Combinator), Garry (Co-host and Partner at Y Combinator), Harj (Co-host and Partner at Y Combinator)
7. End-to-End Agent Workflows Drive 7-Figure Batches
Timestamp: 00:17:59 to 00:22:51
Diana reveals that Y Combinator startups building full-stack end-to-end agent workflows rose from 10% to over 25% of the batch,with some companies jumping from zero to seven figures in revenue within three months. Using examples like Juicebox, the hosts explain that enterprises pay 2x to 3x more when software completes the entire job rather than just tracking it.
Significance (High): Explains why AI startups are monetizing faster than legacy SaaS: capturing labor budgets by automating complete workflows like recruiting outreach and medical billing.
Sources in support: Diana (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator), Harj (Co-host and Partner at Y Combinator)
8. Stealth Startups Earn Millions on RL Environments
Timestamp: 00:22:51 to 00:27:07
Jared and Garry highlight a stealthy boom in startups selling specialized training data and reinforcement learning environments to frontier AI labs. In the past two years, Y Combinator funded over a dozen such companies each generating over $10 million annually, with some reaching hundreds of millions as labs reportedly spend around $1 billion on custom RL and physical-world robotics data.
Significance (High): Uncovers a lucrative, under-the-radar startup category supplying domain-specific training environments and egocentric physical data to frontier labs.
Sources in support: Diana (Co-host and Partner at Y Combinator), Garry (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator)
9. Robotics Requires Task-Specific Model Fine-Tuning
Timestamp: 00:27:07 to 00:29:20
Diana argues that unlike language models, robotics models operate in 3D physical space requiring real-time reactions, making custom fine-tuning essential for vertical tasks like data center cabling. Jared confirms that every Y Combinator robotics company using Physical Intelligence (Pi) foundation models must fine-tune them on thousands of hours of task-specific footage rather than using them out of the box.
Significance (Medium): Indicates that value in robotics will accrue to startups collecting proprietary vertical training data to fine-tune foundation models for specific physical environments.
Sources in support: Diana (Co-host and Partner at Y Combinator), Garry (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator)
10. Solo Founders Surge to Nearly 20% of the YC Batch
Timestamp: 00:29:20 to 00:32:47
Diana reports that solo founders jumped from about 5% of accepted Y Combinator companies a year ago to 18-19% today. Because AI coding tools lower the technical building barrier, single founders can gain initial traction alone before adding co-founders later in the company lifecycle, though the hosts still expect the most successful startups to bring on co-founders over time.
Significance (High): Reshapes founding team dynamics and equity splits by allowing solo builders to validate ideas and gain traction before recruiting co-founders.
Sources in support: Diana (Co-host and Partner at Y Combinator), Garry (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator), Harj (Co-host and Partner at Y Combinator)
11. Garry Explains Why Experienced Founders Are Back
Timestamp: 00:32:47 to 00:36:15
Garry observes a resurgence of founders in their late 30s, 40s, and 50s, citing Peter Steinberger as an example of how industry veterans with domain taste know exactly what to build. Jared adds that decades of experience managing human engineering teams translates directly into orchestrating large fleets of AI coding agents.
Significance (High): Shifts the competitive advantage in startups from raw youth and coding speed to domain judgment, product taste, and agent management.
Sources in support: Garry (Co-host and Partner at Y Combinator), Jared (Co-host and Partner at Y Combinator)
Neutral sources: Diana (Co-host and Partner at Y Combinator)
This analysis was generated by skim (skim.plus), an AI-powered content analysis platform by Credible AI. Scores and classifications represent the platform's AI-generated assessment and should be considered alongside other sources.
