Public Cloud Economics: Lessons for Enterprise IT (Part 1)

The basic economics of public cloud infrastructure services, like Amazon Web Services (AWS) or Google Cloud Platform (GCP), largely rests with effective capture of Moore’s Law, combined with efficient capacity management and sales growth.  Scale plays a role in cloud economics, but not in the manner often trumpeted in the marketplace.  In the case of infrastructure or platform services, capturing Moore’s Law is the actual coin of the realm, and Google essentially suggests as much in this NY Times article.  Here's an overview of how it works, the enormous amount of value it can create, and some lessons for enterprise IT shops.

When scale matters

There is little doubt that the likes of AWS or GCP acquire equipment at lower prices than enterprise IT shops, or can bleed losses longer than many of their competitors.  But for whom does that create a barrier to entry?

In other industries, the initial capital outlay for an initial "unit of service" often far exceeds actual usage requirements.  An electric utility invests in power generation, airlines require aircraft, lodging – a hotel, even ground transportation needs a vehicle.   In information technology, capital investment scale provides little barrier to entry for an end-user.  The initial capital investment is trivial because it is easy to acquire equipment proportional to their usage requirement.  

Scale in "physical" businesses like power, airlines or lodging also have elements of product placement scale — presence in many locations.   For example, an airline that offers service to only two cities would address a very limited market.  High speed internet just about kills distance, so where a data center is located is hardy a scale barrier, although sovereign regulations about data storage may change that dynamic.  

Another scale argument is public cloud providers achieve higher utilization rates than individual firms.  That is likely true, but is it necessarily relevant to customers’ buying decision?  Does a service consumer really care about others' consumption rates if their own needs are being met?

Lower utilization is a generalized “outcome” measure and often reflects both high business opportunity cost (service capacity at the ready) and IT inefficiency.  For enterprise shops, the former is a legitimate service level challenge that, unfortunately, often masks the latter.  The result is a perception among stakeholders that internal IT squanders valuable resources entrusted to them.

How might enterprise IT shops change this perception?  A good place to start is with top-flight competitors.  What are they doing to increase utilization? 

The value of price signals

In the case of a public cloud provider, what is their opportunity cost if capacity is subscribed and paid for – but not consumed?  Are they willing to risk poor service to a paid-up customer and “overbook” inventory (i.e., oversubscription) to increase their utilization rate?  Unlikely.  Nonetheless, there still exists unused capacity, in the aggregate, just like enterprise shops.  The public cloud leader, AWS, was not content and pushed forward.

What if customers are incented to "turn-off" their reserved instances (by avoiding hourly fees) while waiting for demand to arrive, as AWS does?  AWS then prioritizes the queue of instances, and choreographs their placement, as they are “turned-on.”  Is oversubscription assumed, within AWS's algorithms, to manage requests?  Likely.  Oversubscription is simply a practice of dealing with observed versus real demand.   It's the use of price signals (i.e., customer choices) that enables AWS to do more than just fill an order.  AWS is orchestrating and prioritizing how resources are consumed to maximize their value to AWS by understanding how customers perceive value.       

What AWS's revenue management clearly teaches is price signals matter a lot - regardless of scale.  “Cost transparency” and “cost allocation models” are politically safe tactics for enterprise shops, but poor substitutes for sound price signals (rates) that smartly increase under-utilized assets.   

If organizations are not willing to price their internal economy differently, why expect internal customers to behave differently or change their perceptions?

How Moore's Law is captured

Capturing the value of Moore’s Law starts with acquiring new and better servers.  For most public cloud providers, the motivation to do so is sales growth.  For enterprise shops, the motivation is both demand growth and high inventory holding costs.


When a public cloud provider smartly adds new servers, it reduces the average server fleet age.  The younger the fleet, the higher the fleet’s overall performance.  The gap between the fleet's average cost/performance relative to the current market offering is the key metric.  The smaller the gap, the more pricing power a cloud provider enjoys in the marketplace.

Public cloud providers leverage pricing power by converting Moore's Law into price decreases, which is accomplished by holding constant “instance” performance specifications and slicing new, more performant servers into additional instance inventory.  Customers that need additional performance purchase more, larger or different types of instances to increase capacity or speed.

What's the capacity of your fleet?

Both AWS and GCP use their own standard capacity measure (respectively, EC2 Compute Unit and Google Compute Engine Unit) to define a service unit for instances.  From a technical perspective, these standard capacity measures provide a scalar value to compare relative performance among different instance type offerings, and do so for the entire portfolio regardless of the host server or its location.  From a business model perspective, these service units provide a framework that enable AWS and Google to:

  • Define and sell instances with clearly differentiated performance attributes.
  • Consistently offer those services across different data centers and regions.
  • Manage their fleet in terms of capacity, cost/capacity, and price/capacity.
  • Measure, capture and convert Moore’s Law into cost savings while controlling inventory risk

In contrast, enterprise shops often define a “service catalog” and pursue “cost transparency” for their services, but do so without the foundational building block of measuring and normalizing capacity across their fleet and services (e.g., the unit "sold" is a virtual cpu instead of a server, and like servers, neither represents a stable measure of capacity).  If IT isn't defining capacity, how are customers expected to rationalize use or cost with a moving target?

The grocer's dilemma

AWS is a classic example of a "virtuous cycle of growth."  Growth leads to a younger fleet, which leads to pricing power, which leads to price reductions, which promotes more growth.  So is public cloud dominance simply inevitable?  Maybe.  But what happens if growth slows?  Even at scale, the server fleet age is likely to drift older and fleet cost/performance will increasingly lag current market offerings and pricing power diminishes.  (For example, read why Rackspace recently hired Morgan Stanley.   Compare Rackspace's forecast revenue growth of 16% to competitors.)


As time evolves, server inventory is both highly perishable and quickly decays in value. Compare that to passenger air travel where seats are highly perishable (nobody pays to buy yesterday’s unsold seat), but aircraft do not decay in value.  If Boeing could deliver aircraft with 40% more fuel efficiency every year, aircraft value likely would decay.  So "scale" does poise some real inventory risk - as Rackspace experienced.

In fact, maybe servers should no longer be considered assets at all, but perishable consumable inventory?  Their economic behavior looks a lot like selling expensive produce in a bumper crop world.  Today’s vegetables may be highly appealing, but they will perish if not soon eaten.  And holding inventory today, to sell tomorrow, runs the high risk that tomorrow’s market price is likely to be cheaper and your vegetables are now a day riper.

The mindset of retailer

AWS seeks to minimize those risks by capturing demand and inventory positions in real-time and translate those signals throughout their operations and supply-chain to shorten the physical hardware order-to- virtual instance sale cycle, and ensure the freshest inventory while reducing over-capacity risk.  And because an EC2 instance is ephemeral, instance life is short-lived compared to a physical server (e.g., 200 days is cited here) and therefore the bond between a virtual instance and a physical server is very weak.

Selling ephemeral instances means physical server inventory turnover is not encumbered by instance inventory and the value of Moore’s Law can be captured with reduced  transactional friction.  It appears it took a retailer to fully grasp its market potential. 

In contrast, many enterprise shops have a very foggy picture of their current inventory position, rarely manage those assets to expressly capture the value of Moore’s Law, and then over-compensate for poor asset tracking with even poorer procurement tactics.

In part 2, I will review how enterprise IT shops can co-opt the practices of public cloud providers, like AWS and GCP, and adapt them to their own environment to reduce costs.