Google LLC has reacted to the launch of rival Amazon Web Services Inc.’s SageMaker machine learning service by dramatically slashing the cost of its own offering.

Technology news website VentureBeat first reported the price cuts, citing a blog posted online by Google that was later removed. However, the new prices for its managed cloud machine learning service are reflected in the company’s official documentation.

The cuts are fairly significant, with customers now being charged 43 percent less than before for Google’s basic tier compute service for training machine learning algorithms, VentureBeat reported.

It’s not clear why Google’s blog post was pulled, but it could have something to do with a couple of other new services the company is planning to launch soon. The deleted post alluded to the general availability of a new online prediction feature. In addition, it mentioned a private alpha test of online prediction for gradient-boosted decision trees using the scikit-learn and XGBoost frameworks.

A quick Google search found no mention of those services being available yet, which suggests the company probably intends to keep them under wraps for the time being. Google declined to comment beyond pointing to the new pricing on the site.

Still, the machine learning price cuts are real, coming just two weeks after Amazon launched its rival service AWS SageMaker, which is said to provide an easy way to train and deploy machine learning models. The service comes with prebuilt “notebooks” for common problems and built-in algorithms, as well as what Amazon described as “one-click” training of the models after specifying data sources to draw from.

“There is no other solution out there that lets you do close to as easy as this,” AWS Chief Executive Andy Jassy said at the time of its launch.

Google’s own ML Engine service offers very similar functionality, with its own prebuilt systems and automatic algorithm tuning.

The price cuts suggest that Google is worried about protecting its early lead in machine learning services, which is one of the few areas in the cloud where it’s believed to dominate. Google is also facing pressure here from the likes of Microsoft Corp. and Facebook Inc., which recently teamed up to announce an open-source machine learning project called the Open Neural Network Exchange that provides a shared model representation for interoperability and innovation in the AI framework ecosystem. It’s also compatible with machine learning frameworks such as Cognitive Toolkit, Caffe2, and PyTorch, all of which are rivals to Google’s own offering. Microsoft also has a separate project called Brainwave, which is a deep learning acceleration platform powered by Intel Corp.’s field-programmable gate arrays.

This “anti-Google alliance” is most likely what compelled Google to cut its prices for its machine learning services, Holger Mueller, principal analyst and vice president of Constellation Research Inc., said in an interview.

“At the end it’s going to be a mixture of capabilities, mindshare and total cost of ownership that will win the battle for the machine learning crown,” Mueller said. “We know Google’s Tensorflow is currently leading in mindshare, but the rest of the game is open and it’s still early days for providers. The race will ultimately end with an autonomously running deep learning networks on enterprise data.”

Google has also reportedly tried to clarify its pricing structure for machine learning, publishing “per-hour” prices for each of the training machines it offers, something that was previously unavailable.

Image: Ben Nuttal/Flickr