Infrastructure Startups funded by Y Combinator (YC) in San Francisco 2026

Fivetran automates data movement out of, into and across cloud data platforms. We automate the most time-consuming parts of the ELT process from extracts to schema drift handling to transformations, so data engineers can focus on higher-impact projects with total pipeline peace of mind. With 99.9% uptime and self-healing pipelines, Fivetran enables hundreds of leading brands across the globe, including Autodesk, Conagra Brands, JetBlue, Lionsgate, Morgan Stanley, and Ziff Davis, to accelerate data-driven decisions and drive business growth. Fivetran is headquartered in Oakland, California, with offices around the world. 

Building the communication infrastructure needed for fully autonomous agents

Sandbox infrastructure for AI agents, with full-environment forks and snapshots across files, processes, memory, and databases.

Minicor is a platform for building and running desktop automations at scale. One API call triggers a full desktop workflow on a Windows VM. On-premise, cloud, or Citrix. Deploy your automations with self-healing, video replay, and horizontal scaling built in. Go live with any customer using desktop software at scale.

Compression middleware that removes context bloat in milliseconds, lowering costs and improving end-to-end latency. Compression is especially effective across natural language workloads. In a blind LLM arena case study with one of our customers, compressed requests increased user preference, lowered costs, and lifted purchase volume by 5%.

AI SRE agents that automatically learn each customer’s system so they work just like an in-house engineer.

Terminal Use is an orchestration platform for background agents. Purpose-built for agents that use filesystems. Our platform is CLI-first, making it easy for your coding agents to experiment and continuously improve your deployed agents.

Piris Labs offers a full-stack inference service that eliminates the AI data movement bottleneck. By pairing proprietary photonic hardware with a vertically optimized software stack, we minimize the "memory wall" associated with expensive GPUs. This allows us to deliver the same performance as traditional clusters at a fraction of the cost. Our technology improves effective FLOP utilization and reduces latency, finally making the unit economics of trillion-parameter models sustainable. Founded by a team of MIT physicists and Meta AI experts.

Cascade builds software infrastructure for autonomous intelligence with self-improving safety and reliability models that continuously maintain, update, and prevent threats and failures at scale.

We enable AI voice agents to collect payments over the phone through a single API call, handling card processing and PCI compliance across all payment gateways

Your agent processed a refund without looking up the order ID, costing you thousands. You only found out three hours later from a support ticket. Evals, output scoring, and observability can reduce the likelihood of mistakes like these occurring - but there's no solution that inspects and prevents an action as it’s about to execute. Salus does that. We’ve built an API that wraps around your agent and checks its actions at run time, blocking incorrect ones and providing immediate feedback to guide retries. Kevin and Vedant were roommates at Stanford, where they both studied computer science.

Autumn is building the first real-time signal intelligence platform for GTM teams. We monitor posts, commits, blogs, and announcements, surfacing buying signals the moment they appear. Define your ICP and the signals that matter, and we deliver a condensed, real-time feed filtered by intents.

Get your personal OpenClaw instance on the cloud and connected to your Slack or other messaging app in 3 minutes. Pre-installed security features and common skills.

Oximy helps enterprises understand and manage AI usage across their workforce. Used by security, finance, and transformation teams to track adoption, control spend, and manage governance as AI scales.

APIs weren’t built for agents. Orthogonal fixes that. Developers and agents get instant access to hundreds of APIs through our MCP or SDK. No API key management, no billing headaches, pay as you go. API providers list once and become instantly discoverable.

Moda is the reliability & monitoring layer for agents. It surfaces patterns across agent hallucinations, laziness, forgetfulness, and tool call failures. While also providing analytics on user frustration, net promoter score, retention, and churn rates. In the next 10 years, with more AI agents spawning and trying to get PMF, developing an agent that actually works for users is critical. Everyone has access to the same models, thus the ones that win on a product level will be the ones who learn fastest from real usage. Moda is that learning layer.

Compresr provides an API that compresses LLM context without losing what matters. It’s a drop-in for agents and RAG that cuts token costs and improves accuracy.

We're building the world's largest library of videos by combining public videos collected from the open web and private videos aggregated from thousands of different sources. Both feed into a single, continuously updating index of billions of videos that we clean, search, and label to build custom datasets for AI teams. If a lab needs 100K hours of cooking videos featuring hand-object interactions with no scene changes, we query our index, extract the matching subset, and deliver a clean, structured dataset in days.

Cumulus Labs is a serverless GPU cloud with a proprietary inference stack, purpose-built for AI teams who want faster performance, lower costs, and zero infrastructure work. If your team uses GPUs — for inference, training, fine-tuning, or any compute workload — Cumulus replaces the painful parts of your stack with a platform that just works. Most teams today are stuck choosing between bad options. Self-hosting inference means wrestling with vLLM or SGLang configurations, debugging CUDA issues, and babysitting infrastructure that breaks at scale. Managed API providers like OpenRouter or Fireworks are convenient but expensive, and you're paying their margin on top of the compute. Serverless GPU platforms give you flexibility but hit you with slow cold starts, idle billing, and no help on the inference layer — you're still on your own to make models run fast. Cumulus eliminates that tradeoff. The platform assigns containers in seconds, scales to zero when you're idle, and scales up to as many instances as you need with no waitlists or approvals. You're billed by the second for exactly the compute you use — nothing more. For inference, Cumulus ships Ion — a proprietary engine that supports all major LLMs, VLMs, and MoE architectures out of the box. Ion is optimized for latency and throughput beyond what teams typically achieve managing vLLM or SGLang themselves, with zero configuration required. Whether you're serving your own fine-tuned model or hosting an open-source model like Llama or Mixtral, Ion handles the performance layer so your team doesn't have to. Cumulus also supports checkpointing, model compilation, and LoRA serving, and the team forward-deploys custom optimizations directly for customers. For training, fine-tuning, and general container workloads, teams bring any job to Cumulus and run it on the same serverless infrastructure — no cluster management, no GPU debugging, no orchestration setup. If you're paying for inference APIs, Cumulus lets you run the same models yourself for less. If you're self-hosting, Cumulus makes your models faster without the operational burden. If you need GPUs for any workload at all, Cumulus is the simplest and most performant way to get them.

Chamber puts your AI infrastructure on autopilot. Our agentic AI enabled platform orchestrates, governs, and optimizes your AI infrastructure so teams can run ~50% more workloads on the same GPUs without manual intervention. It operates like an autonomous infrastructure team: continuously monitoring GPU clusters, forecasting demand, detecting unhealthy nodes, and reallocating resources in real time to where they create the most impact. Your ML teams move faster, infra waste drops, and GPU bottlenecks disappear.

Tensol lets you deploy and manage AI employees for your team - built on OpenClaw. Each AI employee runs 24/7 in its own isolated VM, has full context of your company, and can monitor Slack, track Sentry errors, update your CRM, and more. We handle the infrastructure, security, and deployment. Tensol also lets you create specialized AI employees for different jobs - engineering, sales, support, in a couple of minutes. You connect the tools your team already uses (Slack, GitHub, Sentry, HubSpot, Linear, Gmail, and more) with one-click integrations. Every action is logged with a full audit trail, credentials are injected at network-level, and each AI employee gets its own dedicated identity and sandboxed environment, with enterprise-grade security built-in.

Didit is an all-in-one identity platform that lets humans prove who they are starting with a face scan. It powers the fastest identity verification in the market, fights fraud by default, and unifies all identity checks with simple pay-per-use pricing.

Luel is a sourcing and licensing platform for rights-cleared multimodal training data at scale. We work with frontier AI teams to provide high-quality bespoke data collections and off-the-shelf datasets.

Captain delivers the most accurate general-purpose retrieval engine for unstructured data. Connect file stores and effortlessly retrieve knowledge with much higher accuracy than RAG (Avg: 78% → 95% + citations).

Hyperspell (YC F25) gives AI agents memory. It connects to tools like Slack, Gmail, Notion, and Drive so agents can recall, reason, and remember across company knowledge. Developers use Hyperspell’s API to turn disconnected agents into AI coworkers that understand their work. Conor and Manu founded Hyperspell after building their own workplace agent and seeing how limited AI is without memory. They have over 15 years of experience in machine learning and have scaled API-based products to $30M ARR.

S2 is a serverless datastore for real-time, streaming data. The more we worked on data systems, the more we felt like there was a missing building block. The seamless experience of object storage simply did not exist for durable streams – so we set out to fix that! S2 reimagines streams as an unlimited and access-controlled web resource. Like if Kafka and S3 had a baby. - Building agents? Use granular streams to persist context in real time, make workflows auditable, and coordinate between agents. - Publishing fast-moving data? Streams support high fanout and can be accessed directly over the internet, so they are a great fit for broadcasting real-time feeds. - Sandboxing? Emit to per-instance streams, and read history or live events just as easily. - Local-first and multiplayer experiences? Propagate updates reliably and efficiently. Stop dealing with clusters, and start streaming.

We build infrastructure enabling AI agents to reliably perform complex tasks in production

1. Multi-model Agents SDK: build production-grade tool-calling agents in 5 lines of code using our open-source, vendor-agnostic SDK. Connect multiple models across providers (eg. Opus 4.6 + Gemini 3 Flash + GPT 5.4) within the same agent to optimize your agentic workflow's cost, latency, efficiency, and performance. 2. Agents-as-a-Service (AaaS): serve your agent or service as a hosted endpoint that anyone - including OpenClaw and Claude Code - can call.

Truth Systems provides a compliance agent to govern the use of generative AI. It translates operational policies into intelligent, real-time guardrails that monitor every AI interaction. This empowers companies to proactively block non-compliant actions and align access to AI tools with user competency. By eliminating the risks and operational blind spots associated with uncontrolled AI adoption, our platform enables enterprises to fully leverage the technology's potential without compromising security. We are establishing the new standard for AI governance, building a future where every professional can innovate safely and confidently.

Kernel is crazy fast, open source infra for AI agents to access the internet 🧠🕸️ - Built on unikernels for crazy fast cold starts and low resource usage - Sub-150ms cold starts on managed headful browsers - Built-in stealth mode with residential proxies and automatic CAPTCHA solving - Full observability: session recording, live view, and full logs - Persistent sessions that maintain cookies and auth across invocations - Native support for all browser and computer use frameworks - SOC 2 compliant and HIPAA-ready Trusted by Cash App, Rye, and 1000+ others. Backed by Accel & Y Combinator (S25).

DeepAware builds an AI-driven automation system for GPU-intensive data centers. Our reinforcement-learning scheduler, real-time market integration, and unified dashboard slice energy waste by up to 30%. Coming soon: autonomous “robot-hand” inspections and maintenance to enable 24/7 operations with minimal staff. More info: https://deepawareai.com/ Jobs/Internship: https://www.deepawareai.com/careers

Pangolin is an open-source, identity-based remote access platform built on WireGuard that enables secure, seamless connectivity to private and public resources.

You built the demo in a weekend. Now you need auth, a prod-grade database, real-time events, AI observability, cron jobs, email, monitoring… and you needed them yesterday. We spent a decade at our previous startup, building a platform while scaling to millions of users and customers like Meta & Netflix. Now we put that into an open-source framework, so others wouldn’t have to do the same from scratch. Modelence is a batteries-included production ready platform for agentic development that abstracts away boilerplate so you can just focus on building your product logic and deploy it to production without worrying about infrastructure. Shipping production apps is much harder than building a prototype that runs locally and 95% of all production apps share the same building blocks. We did a massive amount of work on our previous platforms that was not specific to our product, and we’ve created Modelence to abstract that away: - Authentication - Database setup - Hosting / CDN - AI integration (LLM observability, prompts) - Real-time events - Cron jobs - Email service - Monitoring - Analytics - Admin dashboards Andrej Karpathy recently posted about this same exact problem: https://x.com/karpathy/status/1905051558783418370 and explained it further in his YC AI Startup School talk: https://youtu.be/LCEmiRjPEtQ?feature=shared&t=1940

AgentMail is the email inbox API for AI agents. It gives agents their own email inboxes, like Gmail does for humans.

Janus automates AI evaluations by using high-fidelity simulation environments, catching failures in reasoning, compliance, tool usage, and performance. The resulting datasets benchmark products and feed post-training loops to continuously improve performance over time.

Aravolta provides a single pane of glass software for data centers to increase uptime, reduce costs, and improve performance. A single utility node onboards every asset within an hour and detects hardware failures, hotspots, and other issues. It also generates reports on important metrics like Power Usage Effectiveness (PUE), energy metering, resource utilization, and much more.

Morphik is open-source multimodal search for AI apps. We're used by customers ranging from space-tech teams searching across research papers to developers building brokerage agents.

RunRL improves LLMs and AI agents with reinforcement learning. Do you have a model that needs to get better at a certain task? Are you tired of messing around with prompts? Do you spend a lot of money on observability, and wish that all this data could let your model self-improve? Well, head on over to RunRL.com, and see what we have to offer! If you give us a model, a prompt, and a reward, we’ll make your model’s reward go up.

AI models are terrible in non-English languages because it's nearly impossible to find training data in other languages. So, we're building the world's largest and highest-quality multilingual data library.

assistant-ui helps frontend developers add AI chat to their apps. Our product is used by several dozen companies, like LangChain, Stack AI, Browser Use, and Athena Intelligence. We're building an open source Typescript/React library and a backend as a service, with first class integrations into popular agent frameworks like LangGraph and Vercel AI SDK.

AfterQuery is a research lab (backed by Y Combinator) investigating the boundaries and capabilities of artificial intelligence through novel datasets and experimentation. We believe that pushing the frontiers of AI requires both rigorous scientific inquiry and exceptional human data.

Flowtel is building the next generation operating system for hospitality, powered by modern AI agents that can handle everything from booking to room service.

Reliability on autopilot. Free you team to build what's next.

Enterprises are paralyzed by the risks of deploying generative AI. Traditional methods like RAG and fine-tuning are brittle, expensive, and fail to prevent costly hallucinations or guarantee compliance in high-stakes environments. This lack of trust stalls innovation and blocks ROI. CTGT ensures strict, auditable adherence to your policies on the fly. We deliver defensible audit trails and remediate noncompliant model behavior, so outputs become more aligned with company policies over time, combating data and model drift. CTGT is backed by Google's AI fund (Gradient), General Catalyst, and luminaries like Paul Graham. We provide the control and confidence necessary to unlock the full potential of enterprise AI.

Revyl is a AI-native mobile testing platform that automatically catches and triages bugs before they reach production.

Choose a coding agent harness, model, and task dataset and optimize context and prompts to get the best performance for long-horizon tasks

Archil transforms S3 buckets into a 30x faster, unlimited, local disk. Archil enables AI, analytics, and serverless applications to instantly access massive data sets without waiting for data transfer. Researchers use Archil for shareable, local storage of data set and model versions that never runs out of capacity.

One-click GPU instances with persistent storage, snapshots, and hot-swappable hardware, with the lowest prices anywhere.

Miru makes it easy for robotics teams to manage, version, and deploy robot configurations. Developers use our well-documented APIs and UI components to automate configuration updates for both engineers and operators. In the long term, we're building software infrastructure to accelerate the automation of the physical world.