Cactus Compute logo

Cactus Compute

AI InfrastructureVerified90% conf

Creates lightweight AI models distilled from larger models for on-device deployment

cactuscompute.com ↗

📍 San Francisco, CA

Verified Data

💰
Est. Revenue<$100K ARR

“Based on YC S25 cohort status (very early stage), pre-revenue focus on developer adoption, and typical early-stage AI infrastructure startups before monetization”

🚀
FundingYC S25 funding from Y Combinator and Wellington Management
🔗linkedin.com↗
👥
Users500,000+ weekly inference tasks
🔗ycombinator.com↗
🧑‍💻
Team Size10-30 employees
🔗linkedin.com↗
📈
GrowthNeedle model reached top of Hacker News with 230+ points; 280+ GitHub stars
🔗news.ycombinator.com↗
🏷️
StageSeed (YC S25)
📅
Founded2025

Company Profile

ModelOpen core / Developer Platform
VerticalMobile app development, AI device startups, Enterprise AI
ClientsMobile app developers, hackathon participants from Google DeepMind event
BuyersAI engineers, mobile application developers, privacy-conscious enterprise tech teams building on-device AI solutions
PricingFree for hobbyists and personal projects, paid commercial license for production apps

Contact

✉️ [email protected] LinkedIn

Strategic Analysis

Strategy

Open core model targeting mobile developers with free tier for hobbyists and paid commercial licenses. Betting on privacy-first, on-device AI as a key differentiator against cloud-based solutions. Land-and-expand strategy starting with individual developers and scaling to enterprise teams.

Tactics

Open-sourced core inference engine to drive adoption, premium commercial licensing for production use. Heavy developer community building through hackathons and GitHub presence. Strategic partnership with Google DeepMind for Gemma 4 integration and Y Combinator ecosystem leverage.

Competitive Positioning

Competes with cloud AI inference providers by focusing on on-device, privacy-first solutions. Differentiates on speed (sub-150ms transcription) and mobile optimization versus general-purpose inference engines. Positioned as the premium choice for privacy-conscious mobile developers.

Marketing Approach

Developer-led growth through open source releases and technical blog content. Hackathon sponsorships and community events as primary brand channels. Hacker News launches and GitHub stars as key distribution mechanisms for technical audience.

Notable

YC S25 cohort, Needle model reached top of Hacker News, Google DeepMind partnership

🔗 Source ↗

Tech Stack

C/C++CUDAReact NativeReact NativeFlutterKotlinKotlinSwiftSwiftARM CPU KernelsZero-copy memory mapping
🔗 Source ↗

Recent News

Needle: We Distilled Gemini Tool Calling into a 26M ModelMay 13, 2026Cactus Compute released 'Needle', an extremely lightweight 26M parameter function-calling model distilled from Gemini, capable of running on low-power devices like watches and glasses at high speed.Cactus Compute + Google DeepMind Gemma 4 HackathonMay 10, 2026A major AI hackathon hosted at Y Combinator in collaboration with Google DeepMind, focusing on building on-device voice agents using Gemma 4 and the Cactus inference engine.Launch of Sub-150ms Hybrid Transcription EngineApr 15, 2026Roman Shemet announced a new hybrid engine for Cactus that achieves transcription speeds under 150ms on mobile devices by optimizing model kernels for phone chips.

Related AI Infrastructure Companies

Tokenwise logo
Tokenwise
AI Infrastructure
Headroom logo
Headroom
AI Infrastructure
S
SIA
AI Infrastructure
Kuberns logo
Kuberns
AI Infrastructure
Redis Iris logo
Redis Iris
AI Infrastructure
SpadeBox logo
SpadeBox
AI Infrastructure

Discovery Sources

Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model
May 13, 2026

Signals

growth rateNeedle model reached top of Hacker News with 230+ points; 280+ GitHub stars🔗 source ↗
team size10-30 employees🔗 source ↗
user count500,000+ weekly inference tasks🔗 source ↗
funding raisedYC S25 funding from Y Combinator and Wellington Management🔗 source ↗
trend indicatorOn-Device AI🔗 source ↗
trend indicatorFunction Call Model🔗 source ↗
trend indicatorMachine Learning🔗 source ↗
trend indicatorAI🔗 source ↗

Evidence

github.com

We distilled Gemini 3.1 into a 26m parameter Simple Attention Network that you can even finetune locally on your Mac/PC.

linkedin.com

YC S25 funding from Y Combinator and Wellington Management

ycombinator.com

500,000+ weekly inference tasks

linkedin.com

10-30 employees

news.ycombinator.com

Needle model reached top of Hacker News with 230+ points; 280+ GitHub stars