Machine Learning Is Invading a Cloud Near You

FYI - this blog was published on Tractica web site recently.

The popularity of artificial intelligence (AI) technology has led to application developers demanding more compute power. AI applications come in different shapes and sizes, so the need for compute performance varies quite a bit. Training and inference also have different performance requirements. The training phase requires higher compute performance capacity, while lower compute performance suffices for inference.

Compute performance needs can range from a personal computer (PC) with a graphics card, to a server with a graphics processing unit (GPU) or a more specialized box developed for AI algorithms. However, the costs involved in developing such infrastructure could be quite high. For example, the cost of a server runs in the tens of thousands of dollars and a specialized box like the NVIDIA DGX-1costs $129,000, which is a sizable investment for anyone. In addition, labor and software maintenance costs add up over time.

Cloud Service Providers Jump at an Opportunity

Sensing an opportunity, many cloud service providers (CSPs) are introducing platforms that are oriented specifically toward AI application developers. Amazon has introduced G2 GPU instances and IBM is offering its SoftLayer GPU service. Google recently introduced Google Cloud application programming interfaces (APIs) with a focus on machine learning, Microsoft has Azure machine learning, and CSP Nimbix is offering a field programmable gate array (FPGA)-based service for machine learning acceleration. These instances can be rented with minimal notice and be scaled up or down as needed. The CSP maintains them, so the application developer need not worry about ensuring whether the software is right version or not. These services are also very affordable. Amazon EC2, for example, costs as low as $97.40 for batch prediction. In addition, they support a range of deep learning frameworks from which application developers can pick and choose.

Benefits across the Machine Learning Space

This is good news for a wide range of companies in the machine learning space. Hardware and chip companies can now look forward to having established cloud companies as guaranteed buyers and a volume market. Software intellectual property (IP) companies can use the marketplace provided by CSPs to sell their products. Development tool companies can make their tools available on these platforms and optimize them for a specific platform. Software service providers can train their engineers on these platforms and offer rapid ramp up to their customers.

However, the biggest potential impact could be that this will open up the market to a large number of application developers just like the mobile platform did several years ago. Microsoft, for example, provides a direct link to application developers to publish their applications in the marketplace. A quick look at the page shows a plethora of ideas coming from early developers who are eager to apply machine learning to real life problems. This will only improve over time.

Significant Market Growth Ahead

Tractica forecasts that the market for AI applications will increase from $643.7 million in 2016 to $36.8 billion by 2025 at a CAGR of 57%. The availability of low-cost, on-demand, and cloud-based AI frameworks will be key enablers of this growth.