Ambarella

Ambarella designs and sells low-power system-on-chip (SoC) processors and software for edge and physical AI applications and intelligent automation. The company’s products enable edge devices to perceive, reason, and act on multi-modal sensor data (cameras, 4D radar, lidar, thermal, NIR) for real-time autonomous and semi-autonomous operation. Ambarella combines its CVflow AI accelerator with video/image processing, deep learning, and sensor fusion to support advanced edge AI workloads, including object detection, classification, tracking, segmentation, stereo perception, and, in newer generations, transformer-based models and large AI networks.

Key platform and technologies

• CVflow AI processing architecture: a hardware engine that executes data-flow graphs for AI and vision tasks, including transformer networks (up to 34 billion parameters on some SoCs) and multi-modal sensor fusion.
• Multi-modal fusion: integrates camera, 4D imaging radar, lidar, thermal, and NIR data for comprehensive environmental perception.
• Process nodes and product families: CV7, CV3, and N1 families built on 4 nm, 5 nm, and 2 nm process nodes, with a focus on high performance per watt; tape-out of the first 2 nm design completed.
• N1 SoC: supports on-device generative AI workloads, including LLM inference with models up to 34B parameters and multi-modal vision-language capabilities (VLMs, VLAs).
• CV3 family: central domain controllers for L2+ to L4 automated driving and robotics; integrates CVflow AI, sensor fusion, planning, HD radar, and in some variants a hardware security module (HSM).
• CV7 family: high-performance, power-sensitive edge AI processors for automotive, security, and robotics; designed for complex perception and multi-modal processing.
• B8 SerDes: mixed-signal data transport for adding cameras or bridging to other automotive subsystems.
• Software platform: Cooper development platform and Cooper Foundry software stack, with SDKs, APIs, model porting tools, and a developer zone that includes optimized models and no-code/low-code agent blueprints to speed edge AI application development.
• Advanced radar software: adaptive AI radar algorithms and a centralized 4D imaging radar architecture that pairs with CVflow for enhanced perception.
• Software modules: HD radar processing, deep-learning-based low-light and HDR image processing, monocular/stereo perception, and AD stack components optimized for CV3.

Markets and application areas

• IoT / Cameras
  • Enterprise and public security cameras (IP surveillance) with on-device AI for people counting, tracking, facial recognition, license-plate recognition, and other analytics.
  • Home security cameras and doorbells with intruder/pet detection and on-device AI.
  • Portable video devices and body-worn cameras with on-device AI for low-light performance, efficiency, and extended recording.
  • Consumer devices (action cameras, drones, 360° cameras) with high-resolution video, object detection, and stabilization.
• Automotive
  • Front-facing ADAS cameras, interior cabin monitoring (CMS/DMS), telematics, electronic mirrors, and automotive data loggers.
  • Central domain controllers (CV3-AD family) supporting L2+ to L4 autonomous driving stacks with multimodal perception, deep fusion, planning, and in-cabin sensing.
• Robotics and industrial
  • Fixed robotics, warehousing automation, mobile robotics (drones, autonomous ground robots) using edge AI for navigation, object detection, and manipulation.
• Edge AI boxes/servers
  • Edge devices aggregating video, audio, and other streams with on-device generative AI inference (multi-modal LLM/VLM/VLA architectures).

Customers and revenue concentration

• Sales model: sells to OEMs and ODMs; end products are typically designed by OEMs, with ODMs or Tier-1s often handling design-in and fulfillment.
• Fiscal 2026 revenue concentration:
  • WT Microelectronics Co., Ltd. (WT) served as a non-exclusive sales representative and fulfillment partner in Asia outside Japan and accounted for approximately 70% of total revenue.
  • The largest end customer was Arashi Vision Inc. dba Insta360 (served indirectly via WT and multiple ODMs).
  • The top 10 end customers accounted for about 67% of total revenue.
• Asia exposure: approximately 88% of revenue in fiscal 2026 (85% in 2025; 79% in 2024).
• Accounts receivable with WT as of January 31, 2026: approximately $24.6 million.

Manufacturing and supply chain

• Fabless model; primary foundry partner is Samsung Electronics (Austin, TX and Korea).
• Assembly and initial testing are outsourced to partners including Signetics, STATS ChipPAC, and ASE; final testing by Sigurd Corporation or King Yuan Electronics.
• No long-term foundry or assembly contracts; capacity is supplier-driven and subject to lead times and quarterly constraints.
• ISO 9001:2015 quality certifications for facilities and vendors.

Intellectual property and protection

• Patent portfolio and pending applications cover image/video processing, AI processing, camera/radar perception, and multi-market sensor fusion.
• As of January 31, 2026: 390 issued US patents (14 continuation/divisional), 149 US continuations/divisions, 14 patents in Europe, 24 in China, 8 in Japan, and 58 US patent applications pending.
• Uses non-exclusive licenses for some third-party technologies and actively manages IP risk and indemnification with customers.

Research and development

• Approximately 959 employees as of January 31, 2026:
  • United States: 252
  • Asia: 612 (Taiwan 372; China 225)
  • Europe: 95
• About 75% of employees are in R&D.
• Primary design centers: four R&D centers in the United States, China, Italy, and Taiwan.
• R&D expense (approximate annual spend):
  • 2026: $238.5 million
  • 2025: $226.1 million
  • 2024: $215.1 million
• R&D focus areas include 4/5/2 nm node designs, transformer-capable CVflow, multi-modal AI algorithms, HD radar perception, stereo depth, NN-ISP for low light, and edge LLM/VLM/VLA capabilities.

Location and corporate structure

• Incorporated in the Cayman Islands with corporate headquarters in Santa Clara, California; 58,000 sq ft facility for corporate functions.
• Global presence with offices and facilities in the United States, Taiwan, China, Hong Kong, Japan, Korea, Italy, and Germany.

Product and technology differentiation

• Proprietary CVflow AI processing architecture supporting on-device inference for multiple simultaneous AI models, including transformer networks.
• Deep sensor fusion across camera, radar, and other modalities for enhanced edge perception.
• Highly integrated SoCs with scalable CPU/GPU/ANN accelerators, low power consumption, and compact die sizes.
• Comprehensive software ecosystem (APIs, SDKs, Cooper platform) to accelerate customer development and time-to-market.
• Advanced NN-ISP image processing, HDR and low-light performance, lens de-warping, and high-efficiency video encoding/decoding.

Regulatory, risk, and governance highlights

• Security: ISO 27001-certified information security management system, with formal risk assessments and incident response planning.
• Export controls and sanctions exposure under EAR; dependency on US and international compliance regimes.
• Data privacy and cybersecurity obligations across multiple jurisdictions (EU GDPR, China PIPL/Data Security Law, CCPA).
• Intellectual property risk, including potential patent disputes and third-party license requirements.
• Global supply chain and geopolitical risks, particularly related to Taiwan, China, and trade policies.
• No material ongoing litigation disclosed as of the 2026 Form 10-K.

Public company and governance notes

• Publicly traded shares; corporate governance is governed by Cayman Islands incorporation and associated shareholder rights considerations.
• Company highlights concentration risks, currency exposure (USD-denominated sales with non-USD costs), and macroeconomic volatility as drivers of revenue and margin fluctuations.

Summary

Ambarella develops edge AI SoCs (CVflow-based) and software for intelligent vision and physical AI across IoT, automotive, robotics, and edge computing markets. Its platforms support on-device AI inference, multi-modal sensor fusion, and autonomous decision-making at the edge. Product families CV7, CV3, and N1 cover applications from security cameras to ADAS/autonomy and autonomous robots. Revenue is concentrated through WT as a major distributor/fulfillment partner and Insta360 as the largest end customer, with the majority of revenue generated in Asia. The company operates a fabless manufacturing model (Samsung as primary foundry), maintains a broad IP portfolio, and invests heavily in R&D across multiple global design centers.