Latest from Vellex

Autonomous drones rely on pre-trained AI models to navigate and inspect infrastructure, but these static models fail when field conditions diverge from training data. The traditional fix requires transmitting raw sensor data back to a remote server, consuming critical bandwidth and creating crippling latency. This article explores how the high power requirements of digital processors keep AI retraining locked in data centers, and how Vellex Computing uses milliwatt-level analog hardware to move the retraining process directly onto the drone, ensuring continuous adaptation in remote and contested environments.

Earth observation satellites rely on pre-trained AI models to process massive volumes of imagery on-orbit, but these static models degrade as hardware decays and atmospheric conditions change. The traditional fix requires transmitting gigabytes of raw training data back to Earth, consuming the limited downlink bandwidth meant for high-value intelligence. This article explores how the power constraints of digital processors keep AI retraining locked in data centers, and how Vellex Computing uses milliwatt-level analog circuits to move the retraining process directly onto the satellite, ensuring autonomous adaptation without the downlink bottleneck.

When field-deployed AI models encounter unexpected conditions, they degrade. The traditional fix relies on sending massive amounts of raw data back to the cloud for retraining. This article explores the severe connectivity constraints—from limited bandwidth to crippling latency—that make cloud dependency a critical flaw for edge devices. Discover how Vellex Computing eliminates this bottleneck entirely by using analog semiconductors to train models directly on-device, enabling continuous, real-time adaptation for drones, satellites, and industrial IoT without requiring an internet connection.