Is there a next phase for cloud computing? During the past few years, cloud computing has become a mainstream element of modern software solutions just as common as websites or databases. The cloud computing market is a race vastly dominated by four companies: Amazon, Microsoft, Google and IBM with a few other platforms with traction in specific regional markets such as AliCloud in China. In such a consolidated market, it’s hard to imagine a technology being disruptive enough to alter the existing dynamics.
Artificial intelligence (AI) is the type of technology with the potential to not only improve the existing cloud platform incumbents but also power a new generation of cloud computing technologies.
The thesis of a new generation of cloud computing platforms might seem ludicrous at first but it also presents a very intriguing argument. As a technology trend, AI has not only created a brand new generation of programming frameworks but also influenced a new generation of hardware technologies required to run AI programs at scale.
Mobile and IoT didn’t change the cloud but AI can
Claiming that AI can influence a new generation of cloud computing infrastructure is an interesting proposition considering that transformational technology trends such as mobile or the internet of things (IoT) haven’t had a disruptive impact on the cloud computing landscape. However, the analysis makes sense if we factor in an important difference between movements like mobile or IoT and AI.
From the cloud platform perspective, mobile and IoT capabilities materialized as backend services that could be used from mobile applications or IoT devices. In this sense, cloud platforms were not required to provide the runtime to run IoT or mobile platforms but rather services that enable the backend capabilities of those solutions. Contrasting with that model, AI applications require not only sophisticated backend services but a very specific runtime optimized for the GPU intensive requirements of AI solutions. For instance, a next generation cloud AI platform should be able to deploy a program authored using a deep learning framework like TensorFlow or Torch across hundreds of nodes that are provisioned on demand with optimal GPU capabilities.
AI in the cloud today
The last few years have seen a tremendous level of investment on AI capabilities in cloud platforms. With companies like Google, Amazon, Microsoft and IBM leading the charge, many platform as a service (PaaS) solutions have started incorporating AI capabilities. If we analyze the current landscape of cloud AI technologies, we can identify two major groups:
- Cloud Machine Learning (ML) Platforms: Technologies like Azure Machine Learning, AWS Machine Learning and the upcoming Google Cloud Machine Learning enable the creation of machine learning models using a specific technology. However, excepting Google Cloud ML that leverages TensorFlow, most cloud ML technologies don’t allow the execution of AI programs written in mainstream AI or deep learning frameworks like Theano, Torch, TensorFlow, Caffe, etc.
- AI Cloud Services: Technologies like IBM Watson, Microsoft Cognitive Services, Google Cloud Vision or Natural Language APIs enable abstract complex AI or cognitive computing capabilities via simple API calls. This model allows applications to incorporate AI capabilities without having to invest in sophisticated AI infrastructures.
As AI technologies evolve, cloud platforms should transition from this level of basic support for AI capabilities to a model in which AI programs are as widely supported as web and databases are today.
The AI-First cloud
The AI-first cloud is a next generation cloud computing model built around AI capabilities. Even though we don’t know exactly how the architecture of AI-first cloud platforms will look, we can explore a few interesting ideas:
- Support for mainstream AI frameworks: The next generation cloud computing platforms should be able to run deep learning or AI applications implemented in mainstream frameworks such as TensorFlow, Caffe, Theano, Torch, etc. in the same way they support the implementation of web applications or background processes today. From that perspective, the AI-first cloud should not be constrained to a single AI framework but support heterogeneous deep learning frameworks that are actively being used by developers worldwide.
- GPU optimized infrastructure: In order to run arbitrarily complex AI processes, new cloud infrastructures will have to support GPU environments optimized for fast computing. We are already seeing some initial efforts in this area with Microsoft announcing the availability of the N-Series GPU instances as part of the Azure platform.
- Management tools: One of the biggest challenges of the current generation of deep learning and AI frameworks is the lack of operational management tools. The next generation cloud computing platforms are in a unique position to alleviate this challenge by providing sophisticated tools to manage and operate AI programs deployed in their infrastructure.
- AI-first infrastructure services: The next generation of cloud computing platforms will go beyond enabling the infrastructure for AI programs and will leverage AI as a first class citizen in its infrastructure and platform services. In the future, we can see AI being a key element to improve the intelligence of cloud services such as storage, compute, or security.
- Integration with mainstream PaaS services: In order to build sophisticated AI applications, the next generation of cloud computing platforms should provide seamless integration between AI and deep learning frameworks and the existing catalog of cloud services included in cloud platforms.
Can AI power the next phase of cloud computing?
Cloud computing is a well-established technology trend vastly dominated by companies like Amazon, Microsoft and Google. From that perspective, it seems hard to imagine a technology trend disrupting the current cloud computing landscape. However, AI brings some very unique characteristics that can definitely influence the next generation of cloud computing platforms. AI requires a new computing infrastructure and the support for brand new programming paradigms and frameworks. In the near future, we should expect the cloud incumbents to incorporate AI capabilities as a foundational element of its infrastructures and, maybe, we will see the emergence of a new generation of cloud platforms powered by AI. We are certainly poised to enter the era of the AI-first cloud.
This article is published as part of the IDG Contributor Network. Want to Join?