Startups Directory

Nexthop AI builds custom networking infrastructure purpose-built for AI workloads in hyperscale cloud data centers. The company combines custom hardware, software, and optical integration using a JDM (Joint Development Manufacturer) model to deliver tightly integrated solutions for GPU clusters. Nexthop's Disaggregated Spine architecture achieves up to 30% reductions in energy consumption and infrastructure costs while supporting extreme throughput (50-100+ terabits) compared to traditional networking vendors like Cisco and Juniper.

Together AI is an AI Acceleration Cloud platform enabling developers and researchers to train, fine-tune, and deploy generative AI models at scale. The platform supports over 200 open-source models across all modalities (chat, image, audio, vision, code, embeddings) with 2-3x faster inference than hyperscalers, OpenAI-compatible endpoints, and infrastructure scaling from 16 to 100,000+ GPUs. Founded by Stanford professors and serial entrepreneurs, Together AI is positioned as the cloud infrastructure layer for the open-source AI ecosystem, with recent expansion into data transformation through the Refuel.ai acquisition.

Dash0 is an OpenTelemetry-native observability platform that helps engineering teams monitor metrics, logs, traces, and resources with built-in alerting, dashboards, and AI-powered troubleshooting. Unlike legacy observability tools, Dash0 generates vendor-agnostic, standardized data that isn't locked into proprietary formats. The platform solves the observability crisis—too noisy, too expensive, and too complex—with transparent consumption-based pricing and an AI copilot called Agent0 for intelligent incident investigation.

Decart is an AI research lab building the fastest and most efficient generative AI models for real-time video generation and manipulation. Their flagship products—Oasis (real-time AI game world) and Mirage (live video editing via text prompts)—achieve 10x better efficiency than competitors like OpenAI's Sora, reducing inference costs from $100/hour to $0.25/hour. The company combines a proprietary LLM inference engine (built from scratch in C++ and CUDA) with advanced video generation capabilities, serving developers, enterprises, and streaming/gaming platforms.