An Insider View on the Cloud

One of my roles at Microsoft was leading the Cloud Infrastructure Long Range Plan (LRP). The LRP is a multi-year capital expenditure plan outlining the infrastructure Microsoft needs across each metro and different asset classes that make up the cloud such as servers, datacenters, power, fiber, and land. In short, my job was to make the global Microsoft cloud a physical reality.

The cloud is primary known for the idea of on-demand limitless scale with cloud computing. Pay for what you need when you need it. The classic example is the retailer that sees 10x traffic over the holidays. Instead of paying for servers year-round to support the holiday load, the retailer can scale up and down when they need it.

Building the cloud computing promise

A key part of my job was to anticipate this need and position all the dependent infrastructure to support the growth of Azure over a 5+ year horizon. Customers want to rent out virtual machines that depend on physical servers. The servers need space, power, and cooling in a datacenter. Datacenters require land, fiber, water, energy, and heavy equipment like generators. While the cloud is abstract, the implementation is manufactured across a complex supply chain with long lead times. And with the long lead times, we needed to have enough inventory, also known as safety stock, across the supply chain to allow for the on-demand cloud promise.

To make things more complex, we had profitability targets to meet. If you build too much, you hurt margins. If you build too little, you stifle revenue and likely send customers to competitors. The optimal plan was found by balancing these two forces, ensuring we had just enough to provide the customer a promise of limitless capacity.

Fortunately, economies of scale and technological efficiencies drove a lot of the margin gains with the business. As the cloud gets bigger, complex technical projects that extract out percentage point gains start to have a positive return on investment. As a software company, Microsoft is great at solving those problems. Prior to the LRP, I was the product manager of two of these services that reduced downtime and enabled a denser packing of tenants on our hardware.

Scale challenges with forecast uncertainty

However, intrinsic to the rapid growth of the cloud is the forecast uncertainty associated with anticipating customer need. When we are dealing with construction timelines on physical assets like datacenters, there is a large cone of uncertainty with how the business will look in a few years from now. At current scale, this problem costs around $20B annually for hyperscale cloud providers.

And while software can help make the cloud more efficient, the nature of the cloud business makes it difficult to address this problem from strictly a technological perspective. The best solution towards addressing the forecast uncertainty is an economic one, where you enable customers to help drive the uncertainty down. Reservations are an example of the changing cloud business model.

While I’ll dive deeper into the cloud business model in another post, cloud customers have more data and capabilities to reduce uncertainty in the ecosystem. Thus, they can create value in how they purchase their cloud resources. And that’s the core of what we’re doing at Optrilo. We’re helping customers take advantage of the shift in the cloud business model.