Y Combinator’s Summer 2026 batch brings together startups working across some of the most important areas of emerging technology, from artificial intelligence and robotics to biotechnology, defense, energy infrastructure, fintech and advanced materials.
The YC S26 companies reflect a broader shift in startup building: many are not simply developing another software application, but are attempting to solve difficult problems in the physical world using AI, automation, new computing architectures and specialized hardware.
This article takes a closer look at the startups in the YC Summer 2026 batch, focusing on four things for each company: what the company does, the problem it is addressing, its solution and the founders behind it.
1. Fabraix
Summary: Fabraix builds AI red-teaming agents that continuously test AI systems for security vulnerabilities.
Problem: AI systems evolve rapidly, while manual red-teaming is expensive and difficult to repeat continuously.
Solution: Its Nyx platform autonomously generates attacks, identifies vulnerabilities and provides reproducible attack paths and fixes.
Founders: Ibrahim Abdu and Ahmed Aly, with backgrounds in AI security, reliability engineering and fraud detection.
2. TareBio
Summary: TareBio develops personalized and population-level vaccines by identifying antigens already triggering immune responses.
Problem: Traditional vaccine development struggles with diseases where effective immune targets are difficult to predict.
Solution: It studies disease-specific immune responses to identify relevant antigens and design vaccines around them.
Founders: Eta Atolia, an MD-PhD researcher from UCLA-Caltech who developed the underlying technology.
3. OpenVector
Summary: OpenVector turns existing cameras into autonomous AI systems that understand events and take actions in real time.
Problem: Most cameras only record footage, leaving businesses dependent on humans to continuously monitor what is happening.
Solution: Users describe what they want to detect in natural language, while OpenVector builds vision systems that can trigger actions automatically.
Founders: Andrey Gizdov and Vishal Urlam, with backgrounds in computer vision, camera systems and embedded engineering.
4. OneTriangle
Summary: OneTriangle is building an inference cloud designed to reduce AI serving costs and latency through multi-model inference.
Problem: AI inference is increasingly expensive and latency-sensitive as applications generate more tokens and require larger models.
Solution: Its architecture uses smaller models for prefill and larger models for decoding while transferring KV caches between them.
Founders: Hannah Chung and Medha Venkatapathy, with backgrounds spanning computer science, physics, economics and financial technology.
5. Atomarine
Summary: Atomarine is developing offshore, nuclear-powered floating data centers to provide scalable AI computing infrastructure.
Problem: Traditional data centers face constraints around electricity, land, grid connections, cooling and deployment timelines.
Solution: It plans standardized data-center barges powered by nuclear energy and cooled using seawater.
Founders: Dimitris Koutentakis and Emile Germonpre, with backgrounds in engineering, nuclear technology, finance and data-center infrastructure.
6. WonderTx
Summary: WonderTx uses AI to discover first-in-class drugs, initially focusing on turning injectable treatments into oral alternatives.
Problem: Many validated biological targets lack a suitable chemical starting point, making drug discovery a difficult zero-shot problem.
Solution: Its extrapolative AI models combine with agentic design-make-test workflows to accelerate compound discovery and testing.
Founders: Abraham Heifets, Saulo and Cam, combining AI drug-discovery, generative chemistry and technology leadership experience.
7. Litmus
Summary: Litmus creates AI-powered engineering assessments using a company's own codebases, tickets and job requirements.
Problem: Traditional coding interviews often fail to reflect real-world engineering work, while work trials are difficult to scale.
Solution: Candidates complete realistic, company-specific tasks in their own environment while Litmus evaluates both process and outcome.
Founders: Shaivi Rau and Elena Zhao, with experience across startups, venture capital, Two Sigma and Meta.
8. Marker
Summary: Marker is an AI-native consultancy and platform that helps enterprises rebuild workflows around AI agents.
Problem: Companies struggle to identify valuable AI use cases and move them from experimentation into production.
Solution: Its forward-deployed engineers map business processes, identify opportunities and deploy customized AI agents through the Marker platform.
Founders: David Pastewka, Richard Berwick and Will Drevno, who previously built Drapr and Twindom together.
9. Most Robotic
Summary: Most Robotic develops autonomous systems for evaluating robot performance without constant human supervision.
Problem: Robot testing requires people to watch rollouts, judge success and repeatedly reset physical environments.
Solution: Its evaluation system uses AI to judge robot performance and is designed to eventually automate the complete evaluation-and-reset cycle.
Founders: Claire Mao and Lucy Cai, with backgrounds in robotics, AI, aerospace research and autonomous systems.
10. Computable
Summary: Computable is building an exchange where companies can buy, sell and hedge GPU computing capacity for specific time periods.
Problem: Long-term GPU contracts can leave companies paying for unused capacity while offering little flexibility when prices change.
Solution: Its marketplace allows users to purchase exact GPU windows, resell unused capacity and lock in future prices.
Founders: The founding team comes from financial-market and technology firms including Jump Trading, Coinbase and Citadel Securities.
11. Applied Electrodynamics, Inc.
Summary: Applied Electrodynamics builds radio-wave imaging systems that can see through walls and solid materials to reveal hidden pipes, cables and structural elements.
Problem: Much of the built environment is opaque, making hidden defects and infrastructure difficult to identify and causing costly rework.
Solution: Its camera-like radio imaging technology creates high-resolution images and point clouds through materials such as sheetrock, brick and composites.
Founders: Bilgehan Avser, Brian Huppi, George McLean and Paul Leutheuser, with backgrounds at Apple, Humane, Nest and advanced imaging research.
12. Hardware Intelligence
Summary: hardware intelligence develops AI-powered tools for chip design, with Wave helping engineers debug semiconductor designs using natural language.
Problem: Chip verification and debugging are time-consuming, involving massive waveform files and expensive legacy software.
Solution: Wave uses an AI verification agent to trace errors, suggest fixes and re-simulate designs, while WaveZip compresses waveform data significantly.
Founders: Athreya Anand and Rishov Sarkar, former Google, AWS, Tesla, AMD and Siemens EDA engineers with Georgia Tech backgrounds.
13. Edviro
Summary: Edviro builds AI that operates energy infrastructure by creating a continuously updated model of how facilities consume and manage energy.
Problem: Facility data is fragmented across building systems, sensors and spreadsheets, making it difficult to predict problems or optimize operations.
Solution: Edviro combines infrastructure data to detect anomalies, simulate interventions, recommend actions and coordinate maintenance workflows.
Founders: Hursh Shah and Tanuj Siripurapu, both 19-year-old repeat founders with backgrounds in energy, AI, engineering and software.
14. Greypoint Industries
Summary: Greypoint Industries develops LEGION, a drone swarm that uses radio signals to locate hidden battlefield emitters such as drone operators and jammers.
Problem: Military operators can hide behind terrain and structures, making them difficult to locate reliably on the battlefield.
Solution: LEGION triangulates radio emissions using coordinated drones to identify, geolocate and track hidden targets.
Founders: Nic Chu, Darcy Zhang, Steven Xu and Jordan Khoo, with experience spanning defense, UAV detection, satellite sensors and military systems.
15. IMPACT Drones
Summary: IMPACT Drones provides autonomous interceptor drones designed to protect civilian infrastructure such as data centers, airports and power grids.
Problem: Conventional air-defense systems can cost millions per shot, making them difficult to deploy economically against lower-cost drone threats.
Solution: Its containerized systems launch autonomous interceptor drones that use onboard vision to physically neutralize hostile drones.
Founders: Jasper Lortije and Alessandro Fiume, combining aerospace and drone engineering with sales and operations experience.
16. Shiraz AI
Summary: Shiraz AI builds robots that can learn new manufacturing tasks from a single demonstration by a human worker.
Problem: High-mix manufacturing changes frequently, while traditional automation is expensive, rigid and difficult to redeploy.
Solution: Its robot foundation model uses a demonstration video as context, allowing robots to adapt to new task variations without extensive retraining.
Founders: Navid Aghasadeghi and Mojtaba Mozaffar, with senior robotics experience at Boston Dynamics, Rethink Robotics, Amazon Robotics and Northwestern.
17. Hebbian Robotics
Summary: Hebbian Robotics provides APIs that help robotics companies search, analyze and verify the quality of physical AI training data.
Problem: Robotics teams struggle to prove data quality and detect issues such as drift, duplicates and collection inconsistencies at scale.
Solution: Its APIs provide quality metrics, compliance checks, drift monitoring and duplicate detection for robotics datasets.
Founders: Brandon Ong and Kingston Kuan, with backgrounds in robotics AI, LLMs, high-throughput systems and video surveillance.
18. Petrarch
Summary: Petrarch is a marketplace for specialized operational data locked inside companies, starting with manufacturing and industrial datasets.
Problem: Frontier AI labs increasingly need proprietary operational data, while valuable datasets inside companies often remain unused and unmonetized.
Solution: Petrarch sources, cleans, de-identifies and legally clears enterprise datasets before reselling them to AI companies.
Founders: Ian Lee, Sudhish Swain and Samuel Hahn, all Harvard dropouts with experience across AI data, software engineering and GTM.
19. Exosat
Summary: Exosat is building thousands of low-cost LEO satellites as a neutral, sovereign alternative to existing satellite communication networks.
Problem: Nations and enterprises face concerns around dependence on US- or Chinese-controlled satellite infrastructure and the high cost of alternatives.
Solution: Exosat uses lower orbits, automotive-grade components and non-ITAR/EAR-controlled designs to target satellites costing around $300,000 each.
Founder: Edward Ge, a repeat space-tech founder who previously built Aethero and has experience launching satellite hardware.
20. Kara
Summary: Kara develops ultra-high-purity diamond wafers and components for thermal management in AI chips, data centers, aerospace and advanced electronics.
Problem: Heat limits the performance of GPUs, RF systems, lasers and other high-value hardware while increasing cooling requirements.
Solution: Kara uses engineered diamond materials with high thermal conductivity to help hardware operate cooler, faster and more efficiently.
Founders: Aarin Jhaveri and Anuveer Chadha, combining diamond-industry expertise with materials, quantum and chemical engineering experience.
21. Aktoria Robotics
Summary: Aktoria Robotics provides a global teleoperation network that allows robot companies to use trained human operators when autonomous systems need assistance.
Problem: Many supposedly autonomous robots still require human intervention, while building an in-house teleoperation team is costly and slow.
Solution: Aktoria combines operators-as-a-service with low-latency teleoperation software and 24-hour coverage across India, Mexico and the US.
Founders: Shreenabh Agrawal and Jaskaran Singh Walia, both with Carnegie Mellon robotics/ML backgrounds and research experience.
22. Lamb Labs
Summary: Lamb Labs develops Model Processing Units (MPUs) that hardcode AI model weights directly into silicon for faster and more efficient inference.
Problem: LLM inference is heavily constrained by memory movement, resulting in high energy use, latency and cloud costs.
Solution: Its chips keep model weights on-chip and use customized architectures and quantization to reduce the memory bottleneck.
Founders: Thomas Lanning and Niki Kotecha, with advanced physics, engineering and AI research backgrounds from Oxford, Cambridge and Imperial College London.
23. Ooak Data
Summary: Ooak Data converts internal company data into privacy-safe digital twins that AI agents can use for training and evaluation.
Problem: AI agents struggle with messy, permission-heavy workflows that depend on real organizational structures and cross-tool relationships.
Solution: Ooak connects to company tools and creates anonymized digital twins preserving organizational structures, permissions and workflows.
Founders: Thomas Aubry, Pierre-Louis and Grégoire; Aubry brings data and AI experience from PayLead, Samsung AI and autonomous-vehicle work.
24. Decawork
Summary: Decawork is an agent control platform helping IT teams securely deploy, govern and maintain AI agents across organizations.
Problem: Rapid agent adoption can create security risks, excessive permissions, fragmented deployments and unclear ownership.
Solution: Decawork provides agent identities, scoped credentials, approval workflows, gateway checks, monitoring and audit logs.
Founders: Aman Raj and Sarthak Aggarwal, with backgrounds in enterprise AI, compliance, NVIDIA, OpenAI, Meta and Ema.
25. Studio
Summary: Studio builds simulation models that predict how interconnected consumer groups may respond to products, prices, campaigns and other market decisions.
Problem: Surveys and historical data often observe consumers individually, while real markets are shaped by interactions between different groups.
Solution: Studio creates live market simulations using consumer behavior and external signals to forecast outcomes and compare different strategies.
Founders: Nikita Mullangi and Cameron Malloy, with backgrounds in behavioral research, AI agents and financial simulations.
26. Vernius Systems, Inc.
Summary: Vernius Systems develops low-cost radar seekers that can turn drones into autonomous interceptors for air defense.
Problem: Many battlefield interceptors still require pilots, while autonomous missile systems can cost millions per shot.
Solution: Its Archimedes radar seeker is designed to attach to drones and enable autonomous target detection and interception at a much lower cost.
Founders: Arthur Nguyen-cao, Hisham El-Halabi and Joe Gonzalez, with backgrounds in defense technology, RF engineering and secure military communications.
27. Familiar
Summary: Familiar develops AI dubbing that preserves an actor's voice, facial performance and identity across 30 languages.
Problem: Traditional dubbing can create a mismatch between the original actor's facial performance and translated voice.
Solution: A joint audio-visual model handles voice translation, lip-sync and scene understanding together.
Founders: An Zhu Liu, Mingi Kwon and Xu Zheng, with backgrounds in voice AI, generative video and AI research.
28. 83 Sciences
Summary: 83 Sciences uses unpublished experimental data to train AI models for discovering new materials.
Problem: More than 85% of experimental research remains unpublished, leaving valuable failures and results outside datasets used by frontier models.
Solution: It partners with laboratories to train models on unpublished experimental data for applications such as batteries, semiconductors and catalysts.
Founders: Eric Riesel, Ian Naccarella and Yankang Yang, combining materials science, AI and commercialization experience.
29. OS3
Summary: OS3 builds affordable US-made semi-humanoid robots designed for physical work across industries.
Problem: Robotics development is constrained by expensive hardware and limited access to high-quality robot training data.
Solution: OS3 combines vertically integrated hardware with a video-trained foundation model designed to generalize across different tasks.
Founders: Rishabh Chanana and Chris Hailey, with backgrounds in robotics, AI, software engineering and product development.
30. Prodigy Research
Summary: Prodigy Research is developing a finance-specific AI foundation model and agents for quantitative research and trading.
Problem: Human quantitative researchers have limited capacity to discover and test trading strategies across increasingly complex financial markets.
Solution: Its AI agents identify opportunities and generate strategies that are tested through delta-neutral, risk-controlled trading.
Founders: Brothers Michael Wang and Yuhua Wang, with experience at Google DeepMind, Jane Street, Apple and Salesforce.
31. Cosmic Robotics
Summary: Cosmic Robotics builds autonomous heavy-lift robots for construction, starting with solar installations and expanding toward larger infrastructure projects.
Problem: Construction remains highly dependent on manual labor, while skilled-worker shortages contribute to project delays.
Solution: Its Cosmic-1 robot automates heavy lifting and precise solar-panel placement while integrating with existing construction workflows.
Founders: James Emerick and Lewis Jones, with backgrounds in field robotics, construction automation and aerospace engineering.
32. Neuromorphic
Summary: Neuromorphic builds general-purpose robots that can autonomously perform tasks inside existing wet laboratories.
Problem: Traditional lab automation requires expensive, fixed infrastructure and specialized workflows that are difficult to adapt.
Solution: Its robot brain combines sensors, compute, an LLM orchestrator and safety systems to execute natural-language instructions using existing lab equipment.
Founders: Vatan Aksoy Tezer, Ege Doganay and Cem Toker, with extensive robotics and AI research experience.
33. Molagri
Summary: Molagri develops species-specific biopesticides designed to target pests while avoiding beneficial insects and pollinators.
Problem: Pests are developing resistance to existing pesticides while regulators are restricting some active ingredients due to toxicity concerns.
Solution: Its modular virus-like particles combine pest-specific targeting elements with interchangeable kill mechanisms.
Founders: Zaky Hassan and Min Jin, viral structural biologists and University of Toronto researchers.
34. Synapse Semiconductor
Summary: Synapse Semiconductor develops image sensors that perform neural computation directly inside each pixel.
Problem: Conventional edge-AI systems separate cameras from GPUs, increasing power consumption, latency and hardware costs.
Solution: Its RETINA chip combines sensing and computation in the same device, reducing the need for separate edge-computing hardware.
Founders: Tania Roy and Sanjeev Chauhan, with backgrounds in semiconductor research, machine learning and electrical engineering.
35. Datoric
Summary: Datoric builds secure, verified training datasets for voice AI, robotics and world models.
Problem: Collecting large amounts of human data creates challenges around quality, provenance, consent and fraudulent submissions.
Solution: Its project-specific environments use verified contributors and track the full provenance of every data submission.
Founders: Nikhil Reddy and Jeffrey Lin, with backgrounds in AI/ML, software engineering, data annotation and robotics.
36. Frontier Computing
Summary: Frontier Computing develops biological neural networks as a potential computing substrate for AI training and inference.
Problem: Conventional GPU systems face memory-bandwidth and scaling constraints when running increasingly large AI models.
Solution: Frontier grows cultured neuronal tissue at larger scales, aiming to combine biological memory and computation.
Founder: Michael Domarkas, a former Cambridge Natural Sciences student and researcher in cultured neuronal tissue and reinforcement learning.
37. Robocurve
Summary: Robocurve develops standardized, real-world benchmarks to measure the capabilities of robots and the AI models controlling them.
Problem: Robotics lacks continuous standardized benchmarks, while lab demonstrations and simulations do not always reflect real-world performance.
Solution: Its physical-world evaluations test robots on reproducible tasks using standardized hardware and scoring.
Founder: Jay Chooi, who is building Robocurve around open-source robotics evaluation and benchmarking.
38. Maingen
Summary: Maingen builds simulated industrial environments where AI agents can learn to operate power plants, factories and data centers.
Problem: Industrial expertise is largely tacit and stored in operators' experience and enterprise systems rather than online datasets.
Solution: Maingen works with operators to convert real workflows and decisions into reinforcement-learning environments and benchmarks.
Founders: Phillip Yan and David Yang, with backgrounds in power trading, Scale AI, Coinbase, Amazon and reinforcement-learning research.
39. Verdict Machine
Summary: Verdict Machine provides AI-powered cybersecurity for financial institutions managing digital assets and on-chain infrastructure.
Problem: Digital-asset attacks can exploit authorized users through malicious transactions, while existing security tools often provide fragmented visibility.
Solution: It maps on-chain environments, assesses cyber risk and checks transactions against security policies before execution.
Founders: Dan Danay, Isaac Grossman and Liran Kogan, with backgrounds in blockchain security, cryptography, Check Point and financial technology.
40. Graphify Labs
Summary: Graphify Labs builds an on-device knowledge graph that gives engineers and AI coding agents a continuously updated view of enterprise codebases.
Problem: AI-generated code is increasing faster than teams can review it, making hidden bugs and breaking changes harder to detect.
Solution: Graphify combines codebase knowledge graphs with formal verification to review pull requests and identify changes that break existing behavior.
Founder: Safi Shamsi, a knowledge-graph researcher and author with expertise in knowledge-graph RAG.
41. Meteoric
Summary: Meteoric develops autonomous drones that modify clouds over solar farms to reduce cloud-related losses in electricity generation.
Problem: Low and mid-level clouds significantly reduce solar output, while building additional solar capacity faces infrastructure and interconnection constraints.
Solution: Its electric drones are designed to modify cloud droplets without chemicals, supported by AI-based weather forecasting.
Founders: Mete Karslioglu and Eric Nilsson, both Cambridge engineers with backgrounds in aerospace, climate research, electronics and microfluidics.
42. Caution
Summary: Caution provides a verifiable confidential-computing platform that lets customers prove exactly what software is running in production.
Problem: Cloud software typically relies on trust in infrastructure providers, administrators and CI/CD systems, creating opportunities for code tampering.
Solution: Caution combines confidential computing with reproducible builds and cryptographic verification of source, configuration and runtime.
Founders: Anton Livaja, Lance Vick and Ksenia Lesko, who previously worked together at Distrust on security and confidential-computing systems.
43. Trident
Summary: Trident is an AI-powered pentesting platform that continuously attacks web applications, APIs and cloud infrastructure like a real attacker.
Problem: Traditional annual pentests quickly become outdated as companies continuously ship new code and attackers automate exploitation.
Solution: Trident continuously identifies and chains vulnerabilities into verified attack paths, providing proof of exploitation and remediation guidance.
Founders: Dekai Li and Junaid Mahmood, with backgrounds in cybersecurity, CTFs, bug bounties and computer science.
44. Palisade (OS Security)
Summary: Palisade uses AI to detect and automatically fix operating-system vulnerabilities across large fleets of devices.
Problem: Many security products focus on networks, applications and cloud infrastructure, leaving OS-level vulnerabilities difficult to detect and remediate.
Solution: It monitors systems at the kernel level and applies automated fleet-wide mitigations through kernel security rules.
Founder: Fnu Prince, an ML researcher and security engineer who has worked in offensive and defensive security since his teens.
45. Palisade (Sales Agents)
Summary: Palisade provides AI sales agents that act as personalized salespeople for visitors on online marketplaces.
Problem: Online marketplaces largely rely on search and filters, while offline businesses use salespeople to guide customers and close purchases.
Solution: Its agents understand marketplace listings, proactively assist users, navigate interfaces and complete purchases or bookings.
Founder: Jonathan Salama, a repeat founder with experience at Amazon and AI-powered shopping products.
46. Ekho Labs
Summary: Ekho Labs predicts global freight disruptions such as wars, strikes and tariffs before they impact supply chains.
Problem: Companies often discover supply-chain disruptions only after shipments are affected, leading to production delays and higher costs.
Solution: Its freight world model monitors over 1 million sources and provides early container-level forecasts with recommended actions.
Founder: Alina Park, a repeat founder with experience in signal detection, defense technology and AI-based geolocation.
47. Salem Robotics Inc.
Summary: Salem Robotics develops software that turns robots into autonomous inspection and survey workers for hazardous facilities.
Problem: Radiation and safety teams often perform manual inspections that are time-consuming, dangerous and prone to reporting errors.
Solution: Its system enables robots to navigate facilities, operate equipment, collect measurements and generate auditable reports.
Founders: Caleb Horan and Janak (Crasun) Panthi, both UT Austin robotics researchers with extensive experience deploying robots in nuclear environments.
48. Agnost AI
Summary: Agnost AI analyzes AI-agent conversations to identify silent failures, behavior drift and recurring workflows.
Problem: AI agents can fail without obvious errors, while many repetitive tasks are unnecessarily handled by expensive frontier models.
Solution: Agnost converts production traces into evaluation datasets and specialized models that can perform recurring workflows more efficiently.
Founders: The source identifies Agnost AI as a two-person team but the retrieved section does not provide founder names.
49. Assemble
Summary: Assemble builds AI agents that execute IT implementation work inside enterprise platforms such as Salesforce, SAP, Workday and ServiceNow.
Problem: Enterprise systems accumulate complex custom logic, making even routine changes slow, expensive and difficult to execute safely.
Solution: Its agents scope dependencies, make and test changes, while providing audit trails, approvals and rollback controls.
Founders: Shaurnav Ghosh, Aliyan Ishfaq and Shrish Janarthanan, with backgrounds in Stanford AI, Amazon, Apple, LangChain and cybersecurity.
50. CoArena
Summary: CoArena is a crowdsourced benchmark for computer-use AI, alongside Coasty, its reliability-focused computer-use agent.
Problem: AI agents and RPA tools often struggle with messy legacy software, where failures and silent errors can be costly.
Solution: CoArena generates real-world evaluation data, while Coasty adds error recovery, verification and auditability to computer-use workflows.
Founders: Nitish Kovuru and Prateek Jannu, with backgrounds in AI, machine learning and computer-use agents.
51. Tensr
Summary: Tensr builds autonomous robotic factories that manufacture robots, with a 12,000 sq. ft. facility in the Bay Area.
Problem: US robot manufacturing remains expensive and heavily dependent on manual work across design, supply chain, assembly and quality control.
Solution: Tensr automates the manufacturing process end-to-end using robotics and deep learning, from customer order to completed production batches.
Founders: Eric Berndt, Adith Sundram and C.K. Wolfe, who previously worked together on UC Berkeley's autonomous IndyCar program.
52. Praxis Robotics
Summary: Praxis Robotics captures real-world human demonstrations and turns them into high-quality multimodal training data for robotics and humanoid companies.
Problem: Robotics lacks an internet-scale dataset of diverse human physical actions because real-world data is difficult to collect across environments.
Solution: It partners with businesses to capture video, motion, haptics and 3D data, then standardizes, synchronizes and quality-checks the datasets.
Founders: Rohan Seelamsetty, Dev Karpe and Tommy Li, combining robotics research with operations, finance and industrial technology experience.
53. Markov
Summary: Markov builds data infrastructure for computer-use AI using synchronized screen recordings, keyboard/mouse actions and expert annotations.
Problem: Frontier AI models still struggle with long, multi-step desktop workflows because high-quality expert computer-use data is limited.
Solution: Markov collects expert demonstrations across productivity, finance, CAD and other professional software for training computer-use models.
Founders: Dev Mandal and Harish Ashok, with backgrounds in AI labs, robotics, hardware workflows and computer science.
54. Belvedir
Summary: Belvedir is building an automated factory for creating private, customized AI models for companies.
Problem: Custom models can offer better cost, performance and privacy but are traditionally too complex and expensive for smaller companies.
Solution: Belvedir automates data collection, training, benchmarking and private deployment, continuously improving models as new usage data arrives.
Founder: Zachary Yu, a repeat founder with previous experience building reinforcement-learning environments for frontier AI labs.
55. Rasyn
Summary: Rasyn develops AI foundation models for chemistry, focusing on complex formulations such as coolants, paints, adhesives and thermal materials.
Problem: Formulation development can take years of laboratory experimentation, while existing chemistry tools are fragmented across workflows.
Solution: Its predictive models accelerate formulation design, while its Marigold platform connects chemistry tools through an AI agent with chemistry-aware validation.
Founders: Ansh Tiwari, Daood Hashmi and Ayush Chauhan, with backgrounds in Caltech, NASA JPL, chemical research and large-scale software.
56. DeepReach Inc.
Summary: DeepReach builds a global network for collecting diverse real-world human data to train physical AI and robotics models.
Problem: Physical AI needs data from millions of real workplaces, but traditional collection labs cannot provide enough environmental diversity.
Solution: DeepReach provides wearable capture hardware and software to local entrepreneurs who collect data across warehouses, farms, workshops and other real-world environments.
Founders: Tim Li and Chris Liu, combining workforce-network experience with computer vision and machine-learning research.
57. Nebula Security
Summary: NebuSec builds AI-powered cybersecurity solutions that continuously identify, validate and remediate vulnerabilities across code, cloud infrastructure and AI systems.
Problem: Modern software changes rapidly, while traditional security testing can struggle to continuously monitor code, dependencies, cloud infrastructure and AI-generated code for exploitable vulnerabilities.
Solution: Its platform provides continuous code security, autonomous vulnerability scanning, AI pentesting, dependency monitoring and cloud security, with verified findings and remediation support.
Founders: Eten Zou (CEO & Co-founder), Yuan Tan (CTO & Co-founder), Frank Wu and Xiaochuan Yu, with backgrounds spanning cybersecurity research, vulnerability discovery, CTF competitions and security research at organisations including Microsoft.
What YC S26 Tells Us About the Next Wave of Startups
The YC Summer 2026 batch offers a snapshot of where startups are heading in 2026. Across 56 companies, AI is expanding beyond chatbots into robotics, cybersecurity, manufacturing, healthcare, logistics, semiconductors, finance and scientific research.
A key theme is the shift from AI that assists humans to AI that performs real-world work. Startups are building autonomous agents, physical AI systems, robotics and tools that operate across software and physical environments.
The batch also highlights growing opportunities in AI infrastructure, specialized datasets, inference, custom models, evaluation systems and AI hardware.
Overall, YC S26 reflects a broader shift toward combining AI, automation and specialized technology to solve complex problems across both digital and physical industries.
To compare companies, documents, and availability explore unlisted companies on Precize. For ongoing IPO and private-market updates, browse the Precize blog. Stay updated with unlisted companies through our Precize Community. If this article was useful, you can share it with other investors through the Precize Referral Program.
