What Nexa AI does

Nexa AI is a developer infrastructure company that builds NexaSDK, a unified toolkit for running AI models locally on device across mobile, desktop, automotive, and IoT hardware. The company's product surface covers five SDK targets — Android (Java/Kotlin via Maven Central), iOS/macOS (xcframework), Python (pip), a multi-OS CLI, and Docker for Linux ARM64 — and runs on NPUs, GPUs, and CPUs interchangeably, including Qualcomm Snapdragon (Hexagon NPU, Adreno GPU), Apple Silicon (ANE/MLX), and x86 CPUs. At the core is NexaML, a proprietary inference engine that Nexa describes as built from scratch at the kernel level rather than as a wrapper over an existing runtime, supporting three model formats (GGUF, MLX, and the proprietary .nexa format) and multiple AI modalities spanning text (LLM/VLM), image (CV/OCR/detection/depth/ImageGen), audio (ASR/TTS/diarization), embeddings, and reranking. Functional positioning emphasizes privacy-preserving, offline-capable, low-latency inference as alternatives to cloud-based AI APIs.

The business model is product-led and developer-first. Distribution is self-serve across public package registries, direct downloads, Docker Hub, GitHub, and Hugging Face, with developers onboarded through documentation and active Discord/Slack communities. The Builder Bounty Program pays up to $1,500 USD for developers who ship open-source applications on NexaSDK, and the Nexa Wishlist lets developers vote on which third-party models Nexa should optimize next. Monetization mechanics center on token-based SDK access for downloading NPU-optimized models, with GGUF and community models distributed without token gating; pricing is not publicly disclosed and there is no evidence of named enterprise contracts or disclosed ARR. Customer segmentation is horizontal and developer-led, targeting mobile app developers, desktop/PC integrators, embedded/IoT engineers, and automotive use cases.

Nexa AI is now operating as part of Qualcomm AI Hub following acquisition by Qualcomm Technologies, Inc., with the SDK redistributed through aihub.qualcomm.com. The company was founded in 2023, remains private, and runs with 11-50 employees out of Cupertino, California.

Name

Nexa AI

Legal name

Nexa AI

Website

https://nexa.ai

Company type

Private

Founded year

2023

Operating status

Operating

Headcount range

11–50 employees

Short description

Nexa AI builds NexaSDK, a developer toolkit for running AI models locally across NPUs, GPUs, and CPUs on mobile, desktop, automotive, and IoT devices. Powered by its proprietary NexaML kernel-level inference engine, it targets privacy-preserving, offline-capable on-device AI. Nexa is now part of Qualcomm AI Hub.

Ownership category

akta.pro rank

Product category

Edge AI Inference Software

NAICS

Software Publishers (5132), Software Publishers (513210)

SIC

Services-Prepackaged Software (7372)

akta.pro primary industry

On-Device Inference Runtimes & SDKs (mobile/embedded) (HDAAAJAB)

akta.pro secondary industries

AI Application Enablement Platforms (Copilot/Agent Frameworks, SDKs) (HDAEANAJ), Developer Tooling & SDK Platforms (BPAMAOAD), Platform Engineering & Internal Developer Platforms (IDP) (BPAEAKAC)

Keywords

• On-device AI inference
• Edge AI SDK
• AI model runtime
• Local LLM deployment
• NPU acceleration

Headquarters

HQ city

Cupertino

HQ country

United States

HQ region

North America

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Marketing or Sales

Revenue model

1. SDK Access Token: Access tokens required to download and use NPU-optimized models. Users create accounts at sdk.nexa.ai and generate tokens via Deployment section.

Pricing tiers

Go-to-market motion2 records

Distribution channels5 records

Marketing channels5 records

Core offering

Nexa AI develops and distributes NexaSDK, a cross-platform developer toolkit that lets developers run AI models locally on NPUs, GPUs, and CPUs across Android, iOS/macOS, Windows, Linux, and automotive/IoT devices. The SDK is powered by the proprietary NexaML unified inference engine and exposes multimodal AI capabilities (LLM, VLM, ASR, TTS, Computer Vision, Embeddings, Reranking, ImageGen) via Python, CLI, REST API, and native mobile SDKs.

Product overview

Nexa AI offers NexaSDK, a unified developer toolkit for running AI models locally across hardware platforms (NPUs, GPUs, CPUs). The core is the NexaML inference engine supporting GGUF, MLX, and .nexa model formats. The product suite includes platform-specific SDKs for CLI/Go (Desktop), Python (Desktop), Android (mobile), iOS/macOS (mobile/desktop), and Docker (Linux/IoT). Key capabilities include LLM text generation, VLM multimodal processing, ASR/TTS speech processing, Embeddings/Reranker for search, Computer Vision (OCR, detection), and ImageGen. Nexa AI was acquired by Qualcomm and is now part of Qualcomm AI Hub.

Differentiator

Problem solved

Functional benefit

Brands

• NexaSDK: Developer toolkit for running AI models locally across NPUs, GPUs, and CPUs, powered by the NexaML engine
• NexaML
• Builder Bounty Program

Products and services

• NexaSDK A unified developer toolkit for running AI models locally across NPUs, GPUs, and CPUs, powered by the proprietary NexaML inference engine. Supports GGUF, MLX, and .nexa model formats with Day-0 support for new model architectures, and is available via Python, Android, iOS/macOS, Docker, and CLI for developers.
• Nexa CLI Command-line interface for downloading models (nexa pull), running inference (nexa infer), listing/removing models, and starting REST API server (nexa serve). Cross-platform support for macOS, Windows, and Linux.
• NexaSDK REST API Local OpenAI-compatible REST API for text generation, embeddings, image generation, and document reranking, with authentication via NEXA_TOKEN. Runs on port 18181 with Docker and CLI server modes for local deployment by developers.
• NexaAI Python SDK Python library for on-device AI inference supporting LLM, VLM, ASR, Embeddings, Reranker, CV, TTS, ImageGen, and Diarize modules. Supports macOS (MLX), Windows x64 (GGUF/CUDA), and Windows ARM64 (Snapdragon NPU) for Python developers.
• Nexa AI Android SDK Android SDK (v0.0.24) for on-device AI inference on Android devices, supporting LLM, VLM, Embeddings, ASR, Reranking, and Computer Vision with NPU (Qualcomm Hexagon), GPU (Adreno), and CPU acceleration. Minimum deployment target is ARM64-v8a for Android developers.
• Nexa AI iOS & macOS SDK iOS and macOS SDK (Beta) for on-device AI inference, supporting LLM, VLM, Embeddings, ASR, and Reranker models. Uses ANE (Apple Neural Engine) acceleration for ASR and Embeddings, GPU/CPU for other modules. Minimum: iOS 17.0, macOS 15.0.
• NexaSDK Docker Docker container solution for running NexaSDK on Linux ARM64 systems, optimized for Qualcomm NPU devices (Dragonwing IQ9) with support for server mode (REST API) and interactive CLI for embedded and IoT deployments.

Segments4 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

AI capability10 records

Feature5 records

Partnerships

One partnership is on record.

• QualcommcoreStrategic or Co-development PartnerNexa AI is now part of Qualcomm AI Hub. This represents a significant strategic acquisition/integration where Nexa AI's on-device AI SDK technology is incorporated into Qualcomm's AI ecosystem. Nexa SDK is now hosted and distributed through Qualcomm AI Hub at aihub.qualcomm.com.

Recent moves6 records

Expansion highlights6 records

Direct peers

• Ollama: Open-source tool for running large language models locally on consumer hardware. Directly comparable as an on-device/local LLM inference platform targeting developers, with overlapping model format support (GGUF) and cross-platform desktop coverage.
• llama.cpp: Open-source C/C++ inference engine for running LLMs locally on CPUs, GPUs, and various accelerators. Competes directly with NexaML as a low-level on-device inference engine, and is the de facto upstream project Nexa SDK supports via GGUF.
• MediaTek NeuroPilot: MediaTek's on-device AI SDK for Dimensity and other mobile SoCs. Directly comparable as an NPU-focused developer toolkit for on-device AI inference on mobile hardware, competing for the same embedded/edge developer audience.

Broad incumbents

• Apple Core ML: Apple's first-party on-device ML framework for iOS, macOS, and Apple Silicon. Overlaps with Nexa AI's iOS/macOS SDK in the same target audience (Apple developers) with a bundled, OS-integrated solution that competes on convenience and zero cost.
• Google ML Kit: Google's mobile SDK for on-device ML covering vision, text, and translation. Directly comparable to Nexa AI's Android SDK in target market (Android developers) and use cases (on-device CV, NLP, translation), with deep Google services integration.
• Qualcomm AI Engine SDK: Qualcomm's first-party Neural Processing SDK for Snapdragon. Now Nexa AI's parent company; previously a competitor offering NPU-targeted model conversion and inference — the acquisition effectively consolidated these capabilities under one roof.
• TensorFlow Lite (LiteRT): Google's lightweight inference runtime for mobile, embedded, and edge devices. Competes in the same on-device AI inference space across Android, iOS, and Linux/IoT — broader reach but higher complexity than Nexa SDK.

Emerging players

• Hugging Face: AI model hub and inference tooling with growing on-device offering (transformers.js, space support for local inference). Comparable as a developer platform where Nexa AI distributes GGUF and MLX models via Hugging Face collections.
• PyTorch Mobile: Meta's mobile deployment toolkit for PyTorch models. Overlaps with Nexa AI's mobile SDK as a developer-facing on-device ML deployment solution, though broader in scope and tied to the PyTorch ecosystem.
• ONNX Runtime: Cross-platform inference engine for ONNX models with mobile and edge support. Comparable as a low-level inference runtime targeting multiple hardware backends, competing with NexaML's kernel-level approach for model deployment.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat4 records

Key risks5 records

Key highlights6 records

Customer concentration

• GitHubgithub.com/NexaAI/nexa-sdk
• Xx.com/nexa_ai

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds1 record

Investors3 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Nexa AI