Last June, when Microsoft announced its $26.2 billion plan to buy social network provider LinkedIn, most attention focused on the rich purchase price and how Microsoft could take advantage of all that LinkedIn user data.
But other intriguing questions remain about the deal, completed in December. First is whether Microsoft (MSFT) and LinkedIn will eventually converge their computing infrastructures. Microsoft, after all, is building Azure as a competitor to Amazon Web Services, which leads the market in providing shared computing infrastructure to businesses.
Toward that end, Microsoft, like Amazon (AMZN), is aggressively recruiting large third-party software companies to use its cloud data centers instead of (or in addition to) their own facilities. LinkedIn, not to put too fine a point on it, is (or was) a large third-party software company running its own data center facilities.
Wouldn’t it be more efficient for LinkedIn to run in Azure? And wouldn’t the successful migration of all that data prove Azure’s ability to scale up and work smoothly?
Perhaps, but things are more complicated than that.
A LinkedIn spokesman told Fortune that the Mountain View, Calif.-based company will keep managing its own infrastructure to “deliver the best experience for our members and customers.” But he added: “Over time, we may decide to work with Microsoft’s infrastructure.”
You might expect Microsoft, the much larger company with its own infrastructure brain trust, to set the agenda. But that might not be the case. LinkedIn, has its own highly-respected infrastructure professionals, including principal engineer Yuval Bachar, who led Facebook’s (FB) data center networking architecture. That coupled with the fact that LinkedIn co-founder Reid Hoffman was named to the Microsoft board earlier this month hints that the newly-purchased company has clout at the highest leadership tier.
Get Data Sheet, Fortune’s technology newsletter.
Some industry followers speculate LinkedIn might actually be further down the road in designing and using its own hardware than Microsoft itself. As Microsoft faces off against AWS and Google, getting hardware costs down is critical, so it could perhaps learn some lessons from its new hires.
The theory is that controlling hardware design means the cloud company can also tightly manage all the software integration, and thus offer a better user experience. Just as importantly, companies need to roll out new features and functions (and fixes) fast. By controlling the data center soup-to-nuts, a company is better able to do that.
Three years ago, Microsoft climbed aboard the Open Compute Project, which advocates the use of inexpensive “white box” hardware made for massive cloud data centers. That means Microsoft is using more of these non-branded servers built to its own design.
But unlike Facebook, Google, or Amazon, Microsoft also has to manage many decades-long relationships with hardware partners such as Hewlett-Packard Enterprise (HPE) and Dell Technologies. The fact that HPE and Dell still sell branded servers and PCs bundled with Microsoft software makes it problematic for Microsoft to stop buying HPE and Dell gear, cold turkey.
On the networking side, Microsoft uses Arista products, which bundle Arista software with its own specialized hardware. That’s a costly combination. Earlier this month, Arista said it will offer a software-only version of the product that will run on cheaper, generic switches as well as in standard virtual machines or containers, as reported by tech news site SDXCentral. One theory is that Microsoft drove this move to get its costs down.
A Microsoft spokeswoman would not comment on speculation around LinkedIn, Arista, or HPE. She did point to a company blog touting progress to Microsoft’s Open Compute efforts posted in earlier this month. That update touches on the inclusion of the latest chip technologies into what Microsoft calls its “Project Olympus” hardware design but provides no detail on how much of Azure now runs on this type of hardware, versus pricier branded systems.
My guess is that percentage is rising. Last month, HPE blamed a pronounced dip in its server and storage revenue on one large service provider partner that did not buy as much gear as anticipated. Analysts later noted that this partner was Microsoft.
The implication behind Arista’s move to decouple hardware from software is that big cloud customers are demanding it. Network hardware giant Cisco (CSCO) Systems is also working on a version of its network operating system that will run on non-Cisco switches, according to tech news site The Information.
As more workloads get sucked up by AWS, Microsoft, Google, or other public clouds, companies have less reason to build their own data centers and do not need to buy as much of their own hardware. The fact that the big cloud providers are also designing their own switches and servers and then contracting out the manufacturing means that legacy hardware makers are even more under the gun. Which may be why Cisco, for example, is buying its way into the software business as fast as it can.
This use of less costly hardware is something about which LinkedIn knows a lot about. It’s worth noting that Innovium, an upstart networking chip maker that settled a lawsuit last year with much-bigger rival Broadcom, named LinkedIn’s Bachar to its board earlier this month. Innovium is expected to be competitive in that network chip market, likely with more capable and less expensive silicon.
The top cloud vendors, by virtue of their sheer size, can go their own way in terms of hardware. But Microsoft—unlike Facebook, AWS and Google—still has to deal with its legacy vendors and partners in a more gingerly fashion.
For those who are not commodity hardware purists, there is some wiggle room for Microsoft and it’s longtime hardware partners.
As Gartner (IT) distinguished analyst Lydia Leong tells Fortune, HPE and Dell also offer cheaper white box hardware options. And while those manufacturers may be “slightly more expensive,” she says, “they often do better quality control and so may have overall lower costs when the equipment’s full lifecycle is considered.”