About
Cactus Compute (Y Combinator S25) builds Cactus, an open-source on-device AI inference engine that runs language, speech and vision models directly on smartphones, wearables and other low-power devices without cloud dependency.
Capital graph
Founders / team· 2
Comparable companies· by market
This company
BusinessLatest roundTotal raisedDate- Cactus Compute
Cactus Compute (Y Combinator S25) builds Cactus, an open-source on-device AI inference engine that runs language, speech and vision models directly on smartphones, wearables and other low-power devices without cloud dependency.
—
United States
BusinessLatest roundTotal raisedDate- Autoloops
Low-latency real-time inference for open-source LLMs for voice agent developers (serverless API + on-demand GPU clusters), with in-house inference engine Hanoi
— - RadixArk
Open-source AI infrastructure; maintains the SGLang inference engine and Miles reinforcement learning framework for cross-hardware model deployment and training.
Seed · $100M · 2026-05 · Valuation $400M
$100MValuation $400M - PyTorch Foundation
Hosts and governs open-source AI infrastructure projects (PyTorch framework, vLLM inference engine, DeepSpeed, Ray distributed computing, etc.), providing a vendor-neutral, openly governed technical foundation for model training, inference and deployment to AI developers, research institutions and member companies worldwide.
—
China
BusinessLatest roundTotal raisedDate- Qingcheng.AI
AI compute system software: Chitu LLM inference engine and BaGuaLu training acceleration system, adapted to Chinese AI chips including Ascend, Hygon, MetaX and Moore Threads
Series A · 2026-10
— - DeepKernel
Sovereign AI infrastructure: trusted execution and confidential computing so enterprises using large models keep control of data, agent trajectories and accumulated intelligence
Seed · 2026-09
— - ExponTech
In-house distributed software-defined storage SDS2.0, including AI-native KV Cache storage WQS, distributed block storage WDS and parallel file storage WFS, for AI inference/training storage and domestic IT substitution needs of government, enterprise and LLM vendors
Series B · 2026-10
—
Europe
BusinessLatest roundTotal raisedDate- SLNG
Voice AI execution-layer platform for developers and enterprises building voice agents: unified multi-model access, regional compliance routing, lower latency and cost
$4M
$4M - Clastix
Kubernetes multi-tenancy and hosted control plane platform kMetal (Capsule, Kamaji open-source projects) for AI lab and cloud provider clusters
Seed · $3M · 2026-09
$3M - Literal Labs
Develops custom logic-based AI models designed to be faster, more energy-efficient and more interpretable than traditional neural networks
Pre-Seed · $6M · 2025-06
$6M
Other / unspecified
BusinessLatest roundTotal raisedDate- DesignedAI
AI model inference runtime/software framework Metep that optimizes model execution across hardware, cutting compute, power and memory costs
— - Headroom Labs
Token optimization and context compression as Python/TypeScript libraries, proxy or MCP server for AI agent teams; works with Claude, OpenAI, Bedrock, LangChain, LiteLLM, Vercel AI SDK; claims 60%–95% token cuts on structured data
— - Paritok
Context/token compression gateway and open-source 4B compression model for AI coding agents, reducing LLM API costs.
—
Confirmed rounds only. Matched on subsector, sector, stage and funding scale — each verifiable on the company profiles.
Investors that may fit
Sequoia Capital
8 Training & Inference Stack deals in the last 24 months · 5 in Silicon Valley
Request intro →Spark Capital
6 Training & Inference Stack deals in the last 24 months · 6 in Silicon Valley
Request intro →Index Ventures
5 Training & Inference Stack deals in the last 24 months · 5 in Silicon Valley
Request intro →Menlo Ventures
4 Training & Inference Stack deals in the last 24 months · 4 in Silicon Valley
Request intro →Ranked by each firm’s deals in this sector over the last 24 months, weighted by nearby stage and shared geography. Indicative only — no rating judgment.