An unmanned aerial vehicle in flight. Image: SiMa.ai
An unmanned aerial vehicle in flight. Image: SiMa.ai

A new integrated drone AI solution delivers 50 trillion operations per second of AI compute at less than 10 watts, combining SiMa.ai’s Modalix MLSoC with AVerMedia’s industrial-grade platform.

California-based AI computing company SiMa.ai and Taiwanese edge AI solutions provider AVerMedia said the system is designed to meet drones’ strict size, weight, and power constraints while supporting AI applications.

The integrated solution is already available for customer evaluation, with production quantities supported through both companies.

At the hardware level, AVerMedia’s D137SM carrier board is specifically designed for drone applications and integrates with SiMa.ai’s Modalix MLSoC.

AVerMedia’s D137SM carrier board integrates with SiMa.ai’s Modalix MLSoC for AI-powered drone applications. Image: SiMa AI
AVerMedia’s D137SM carrier board integrates with SiMa.ai’s Modalix MLSoC for AI-powered drone applications. Image: AVerMedia

The companies said the integrated design also reduces hardware integration work, helping developers move AI systems from development toward production.

AI Software and Multimodal Processing

SiMa.ai is also providing its Palette Neat development environment to speed Physical AI development with pre-optimized pipelines and reusable components. The company said it can cut development cycles from weeks or months to hours or days.

SiMa.ai’s Palette Neat development environment is designed to accelerate the deployment of AI models on its Modalix MLSoC hardware. Image: SiMa AI
SiMa.ai’s Palette Neat development environment is designed to accelerate the deployment of AI models on its Modalix MLSoC hardware. Image: SiMa AI

“Physical AI applications require more than raw AI performance,” said Liran Bar, SiMa.ai’s vice president of strategic sales. He pointed to the need for efficient compute, low power consumption, software, and production-ready hardware.

The Modalix MLSoC runs multiple AI workloads on a single chip, including convolutional neural networks, vision transformers, large language models, and large multimodal models. 

It can handle the full Physical AI pipeline, from sensor capture and multimodal processing to vision processing and application logic.

SiMa.ai and AVerMedia will showcase the production-ready system at the August 27 Drone Seminar in Taipei, Taiwan.

You May Also Like

BAE Boosts AI Expertise for Assets Management With Oxford Dynamics 

Oxford Dynamics and BAE Systems integrate AI into Prophesea to boost real-time asset management and UK defense readiness.

AI, Drones Take Command in Taiwan’s Next Defense Upgrade

Shih Chi-liang from the ministry’s integrated assessment department said three prototype projects are already lined up for 2026.

Lockheed Simulates 114 Years of AI Combat Testing in ‘AI Fight Club’ Exercise

Synthetic aerial combat scenarios tested how autonomous agents operate under sustained battlefield pressure.