At a fireside chat at Microsoft's Ignite conference in Orlando this week, CEO Satya Nadella conceded to moderator Walter Isaacson that one of his goals was for Microsoft to once again become a curious company that admits it doesn't know it all.
Part of being curious is coming out of its comfort zone, something we saw additional evidence of with announcements, especially with SQL Server, coming out of the conference this week.
Release of SQL Server 2017 on both Windows and Linux was obviously not a surprise -- Microsoft declared its intentions roughly 18 months ago. But it was the latest step in a process of making Linux a first-class citizen, especially on the platform that really matters to Microsoft, the Azure cloud (for the record, SQL Server on Linux is also available in an on-premise edition).
Review: SQL Server 2017 adds Python, graph processing and runs on Linux
The definitive SQL Server 2017 story has already been told by Big on Data bro Andrew Brust on these pages. There was little suspense in the announcement given that Microsoft disclosed its Linux intentions well over a year ago. The port was well thought out in that the result is a native Linux experience, complete with support for the form of command line and package development to which Linux developers are accustomed. It's a bit surprising how the SQL Server got there (under the covers, there is an invisible Windows boot that is part of the process). But SQL Server 2017 on Linux really looks and feels like a Linux database. And as Andrew noted, the benchmarks are there showing its performance.
Not surprisingly for a first release, the Linux version is not yet at full parity with Windows. In some cases, it's just a matter of time; the next dot release will likely offer the support for R and Python in-database support that are now part of Windows. In other cases, there are questions whether Windows-oriented features, like SQL Server Reporting Services, will be in demand by the Linux base.
Of course, this being 2017, it was virtually impossible to escape the onslaught of announcements and strategies centered on machine learning and artificial intelligence. Actually, as Microsoft has been no stranger to ML (especially with its experiences with Bing and Cortana), this was more of a matter of keeping its ear to the ground.
Microsoft has always adhered to a democratization theme in its product strategy, as its platforms and applications began in the workgroup and worked up to the enterprise. Nonetheless, the reality with ML and AI is that they are heavily hyped, but the supply of skilled practitioners dwarfs the demand. The typical Microsoft SMB customer is not about to afford a data scientist, even if they could locate one.
That was the theme behind Azure Machine Learning Studio, a cloud service that Microsoft has offered for several years. It provides a walled-garden approach that offers a more codeless, drag and drop approach to building machine learning models without having to dive into the guts of coding or choosing from the infinitely expanding range of frameworks and libraries that are becoming available in the wild. Microsoft's democratization approach to AI and ML is also evident with its intent to populate applications like Office with features that go beyond grammatical highlighting (the green squiggly lines in Word) to actually pointing out where you have redundant paragraphs.
But Microsoft realizes it has to address the deeper end of AI and ML. Part of that involves offering a portfolio of APIs to its own intellectual property, much as AWS and Google already have. Microsoft has released a number of "cognitive" APIs addressing speech and image recognition, just like its major cloud rivals. These are areas where each major cloud provider will be constantly playing leapfrog.
Serving the elite audience of data science and AI programmers is not just a matter of keeping up with the Joneses. While Azure, like AWS and Google Cloud, offer their own ML cloud services, development of more ambitious cognitive or deep learning services can have knock-on effects downstream, enriching Microsoft's core portfolio of business and productivity applications. So, Microsoft is working with several selected Global 2000 clients on cognitive services for Customer Support/Experience applications.
While those are clearly the types of one-off engagements for which Microsoft is not known, the results could eventually seed the core of the portfolio, like Dynamics 365 with novel features that go beyond chatbots or provide more reliable next-best offers, or with Excel where it (and partners) could provide libraries of ML functions that could be invoked in your rudimentary spreadsheet.
As part of addressing the deeper end of AI/ML, Microsoft is introducing Azure Machine Learning Workbench that addresses data preparation, model development (through integration of Jupyter notebooks), and deployment. As a cohort, those organizations buying big data analytic services in the cloud are more likely to have data scientists, and clearly, Microsoft must serve them.
And ideally, those ML models get tracked and versioned, because just as organizations are increasingly under scrutiny for ensuring data privacy and data sovereignty, they are likely to ultimately be accountable for the ML models that are deployed against the data.
Microsoft is obviously not alone here. Offering platforms for deploying and managing the lifecycle of ML models is a case of keeping up with the Joneses. For instance, providers like IBM and Cloudera have already introduced data science platforms designed to bring together data scientists with data engineers and developers, so that the models that data scientists design survive intact when they actually get deployed. It's an area also attracting a significant third party ecosystem from the Dataikus to the Data Robots and Domino Data Labs of the world. Given the collaboration capabilities of some of the third party offerings, we would not be surprised if Microsoft eventually makes another acquisition here.