The tech giant complements its existing offer with Edge TPU and offers world’s first fully-integrated ecosystem to create AI applications

By Dr. Alexander Lorz and Ann Sophie Löhde

Two years ago, Google introduced its Tensor Processing Units (TPUs) – specialised chips for AI tasks in their data centres. Now, Google offers its cloud expertise as a new Edge TPU. The small AI chip can perform complex Machine Learning (ML) tasks on IoT devices.

The Edge TPU was developed for “inference”, the part of machine learning where an algorithm performs the task it was trained to do, such as detecting a defect on a product in a production line. In contrast to that, Google’s server-based TPUs are optimized for training machine-learning algorithms.

These new chips are designed to help companies automate tasks such as quality control in factories. For this type of application, an edge device has some advantages over using hardware that has to send data for analysis – requiring a high-speed and stable internet connection. Inference on an edge device is generally safer, more reliable, and delivers faster results. That’s the sales letter, at least.

Of course, Google is not the only company that develops chips for this type of edge application. However, unlike its competitors, Google offers the entire AI stack. Customers can save their data in the Google Cloud, train AI algorithms on TPUs; and then run the trained AI algorithms on the new Edge TPUs. And most likely, customers will also create their machine learning software with TensorFlow – the coding framework developed and operated by Google.

This kind of vertical integration provides tremendous benefits for Google and its customers. On the one hand, Google ensures a perfect connection between the different systems, on the other hand, the customer can program his AI in the most efficient way – using only one platform.

Google Cloud’s Vice President for IoT, Injong Rhee, said: „Edge TPUs are an addition to our cloud TPUs to accelerate the training of ML algorithms in the cloud and then run inference on edge. Sensors are becoming more than just pure data collectors – they can locally take smart decisions in real-time.

Interestingly enough, Google is offering these Edge TPUs as a development kit, which enables the customer to test the devices in their environment. This contradicts Google’s current policy to keep their AI hardware a secret. However, if Google wants to convince customers all over the world to apply only their integrated AI tools, they first need to allow them to test their effectiveness. Hence, offering this development kit is not just a strategy to win new customers, but also shows Google’s efforts to become the leading partner for the company’s AI development efforts.

As soon as this announced development kit is available, our experts will run some tests and provide you with feedback. For further questions, please do not hesitate to contact us at

Source: Google Blogs.