Back to research
Data as of 7 min read

Turning 3 Open-Source Acceleration Repos Into an Inference Cloud With 190K+ Monthly Checkout Visits

Founded in 2025 by Zeyi Cheng and David Li (Singapore HQ, multi-million angel round in Apr 2025), WaveSpeedAI unifies 1,000+ image, video, audio, 3D, and LLM models behind one low-latency pay-as-you-go API. Its growth engine funnels developers from Cheng's high-star GitHub speedups (stable-fast, ParaAttention, Comfy-WaveSpeed) and Day-0 multilingual model-API SEO sprints alongside a 10% Tolt affiliate cap—driving ~190K–230K monthly checkout visits.

01

Executive Summary

What this business is

Customers

AI app developers, ComfyUI power users, and creative platforms (such as Freepik and Draw Things) needing Day-0 API access to fast-moving media models.

Offering

Pay-as-you-go multimodal inference API hosting 1,000+ image, video, audio, 3D, and LLM models with low-latency, zero-cold-start execution.

Business model

Open-source developer funnel + Day-0 multilingual programmatic SEO for trending model APIs + B2B integrations and capped YouTube affiliate payouts.

Pricing

Pay-as-you-go API usage billing; exact MRR and gross margins aren't publicly verified on TrustMRR, though checkout traffic runs at 190K–230K monthly visits.

Takeoff mode
Open-Source + Model-API SEO
Open-Source Funnel
stable-fast (1.3k★) · Comfy-WaveSpeed (1.2k★)
Funding & Roots
Multi-$M angel (Apr 2025) · Singapore HQ

New media models (Wan 2.1, Hunyuan, Seedance 2.5, Seedream 5.0) drop almost biweekly, leaving app developers scrambling for fast hosted endpoints. Zeyi Cheng already earned deep developer credibility via open-source Diffusers and ComfyUI inference speedups—making WaveSpeedAI the fastest Day-0 hosted API whenever a new media model breaks out.

Although verified MRR isn't public, ~190K–230K monthly checkout visits alongside commercial integrations (Freepik, Novita AI, MiniMax, Draw Things) point to high-volume API spend. Setting a $150-per-customer lifetime cap on its 10% Tolt affiliate payouts neatly protects inference gross margins against high-spend enterprise accounts.

Founded in 2025 by CEO Zeyi Cheng (author of stable-fast, ParaAttention, and Comfy-WaveSpeed, who previously accelerated image inference 10x at a Silicon Valley infra startup) and co-founder/CTO David Li (ex-Meicai, DP Technology). Registered in Singapore with Hangzhou engineering roots per Preqin, it raised a multi-million-dollar angel round in Apr 2025.

Preparing your reading

A work in progress.

2026.11.08 · V0.4

Refined product categories and extension rankings. Added one-line Agent setup and three membership options.