Grounded in the Cloud
Showing results for 
Search instead for 
Did you mean: 

Does AWS Lose Money on Every Sale, But Make It Up in Volume?


Please note: This article is based on financial reporting for Q32015. A perusal of Q42015 results does not materially change the conclusions. I will update in the future. started reporting revenue and income from the web service (AWS) in Q12015. also provided the previous 12 months of revenue and income on quarterly calls. My sense was that most IT market analysts and IT companies were surprised by how large their business was ($1.566B revenue for the quarter, $265M income). While this sounds like a large number, I analyzed publicly available data, made a few assumptions that are included in this article, and firmly believe that AWS is spending more just on servers each quarter than their revenue. Let me state that again, “AWS is spending more on servers per quarter than their total revenue in the same quarter.” One could justify this if they were cornering the market, but that is a dubious strategy when AWS’s biggest competitors are Microsoft and Google, both of whom generate more cash and have much more cash available than In fact, if one were Microsoft or Google, a sound strategy could be to cut prices, grow the market, and keep doing that until AWS runs out of money. It appears that AWS is doing this for Microsoft and Google. The strategy does appear to be driving out weaker competitors (e.g., Hewlett Packard Enterprise closing their public cloud in 2016, CenturyLink and Verizon exploring selling their data centers).

Let’s step through the AWS reported numbers, server market information, assumptions I made to deduce how much AWS is spending per quarter on server purchases, and you can decide which estimate you think is closest. Even the best case estimate does not paint a financially sound picture for the AWS business. The AWS financials shown below are as reported by


AWS Server Revenue.jpg

Table 1: AWS by the Numbers (as reported by on their quarterly investor call, 22 October 2015)    

Let us now look at the size of the X86 server market. I used a blend of information from International Data Corporation ( and Gartner, Inc. ( IDC provides a worldwide view of the entire server market and breaks out the X86 server market separately. Gartner does not break out the X86 server market separately, but does break out the revenue for major manufacturers (e.g., Hewlett Packard Enterprise, Dell, Lenovo) and Others (Original Device Manufacturers (ODMs) including Foxconn, Quanta, ZT Systems, etc.). I verified that the total server market as reported by IDC and Gartner is within +/-4% of each other in the four quarters shown.






4Q Total

WW Server Revenue ($1000)






X86 Server Revenue ($1000)






ODM X86 Server Revenue ($1000)






ODM X86 Market Share






Table 2: ODM X86 Server Revenue

At this point, we have made no assumptions. and well-known research firms, IDC and Gartner, reported the above information. We will deduce how much AWS spends on servers quarterly using the three methodologies detailed below. The three methodologies are:

  • Percent of ODM market share represented by AWS
  • Total servers deployed by Amazon in November 2014 and using their reported growth
  • Estimated size of public cloud spending and percent represented by AWS


The table below uses the ODM X86 server revenue from above, and shows the AWS spend if they represent 25% and 50% of the total ODM or white box market. Why 25% and 50%? In May 2015, Gartner reported that AWS “is the overwhelming market share leader, with over 10 times more cloud IaaS compute capability in sum than the aggregate total of the other 14 providers in this Magic Quadrant.”[ix] If one assumes that these 15 largest public cloud providers make up almost the total market, then AWS is more than 50% (~90%) of total IaaS market. If we assume the fifteen largest public cloud providers make up only 50% of the total market, AWS is still at least 45% (half of~ 90%) of total market. My perspective is that 25% is too conservative and that 50% might also be conservative. We are also likely overestimating the AWS spend as some large corporations use ODM (Hyperscale) servers in addition to public cloud providers. We do not need to understand all the impacts yet as we will also use two other methodologies to get a range of estimates.


WW ODM Server Revenue2.jpg

Table 3: Method 1: AWS Server Spend Based on Capacity as a Percentage of Total Cloud Provider Market

The second methodology uses an estimate of the total number of servers deployed by Amazon. We use Timothy Prickett Morgan’s estimates from his article in EnterpriseTech.[x] His estimate is based on the number of AWS availability zones (AZs), numbers of data centers (DCs) per AZ, and servers per DC. His methodology is sound with his assumptions clearly stated. He came up with the low and high range shown below. Using AWS’s reported quarterly growth, we have approximately 70% YoY Growth. Ignoring replacement of older servers (reasonable assumption as rapid growth means most equipment was more recently deployed), we estimate that AWS deployed an additional 1.96M to 3.92M servers in the past year. Using IDCs average X86 server cost of $4758, we calculate the total server spend. We then discount the server costs by 20% representing ODM savings (20% is a reasonable discount as Hyperscale X86 servers are commoditized and use similar components from the same handfuls of suppliers (e.g., Intel, Samsung, Hynix, Micron, Toshiba, Seagate, Western Digital).

AWS Server Estimate.jpg

Table 4: Method 2: AWS Server Spend Based on Estimated Number of Servers in AWS

The third methodology uses the estimated size of public cloud spending on infrastructure. The information comes from the IDC Worldwide Quarterly Cloud IT Infrastructure Tracker and includes server, storage, and DC network. 25% and 50% of totals were used as in the second methodology to estimate AWS’s portion of total market spend. While these spend amounts include more than servers, we will consider the impact when looking at the ranges provided by the three methodologies. We use an average of the spend in 2014 and projected for 2015 as our period of interest is July 2014-June 2015.




2014Q3 thru 2015Q2

WW Public Cloud IT Infrastructure Spend ($1000)




AWS Spend if 50%




AWS Spend if 25%




Table 5: Method 3: AWS Infrastructure Spend Based on Total Public Cloud Infrastructure Spend

The table below combines the information from the three methodologies detailed above. Comments are added to provide my opinion on which methods may have returned high or accurate numbers. Keep in mind that for all methods, the annual expenditure is only for servers or infrastructure and does NOT include cost of the data centers, deploying the servers or infrastructure, and operating the servers. Servers depreciate on a three-year schedule and typically have a 3-4 year operational life. From a total cost of ownership (TCO) perspective, acquisition costs are generally in the 20-40% of TCO range. At 25%, the cost of purchasing servers is roughly equal to the annual operating costs for that server. Looking at the numbers in table 6 below, AWS total revenue over the four quarters is $5.98B and just acquisition costs range from $4.54B to $14.9B. No matter how one slices the information, one must conclude that this is a cash-negative business with negative margins.



Summary AWS Server Spend.jpg

Table 6: Summary AWS Server/Infrastructure Expense for All Three Methods

Now we will do some sleuthing. Much like detectives, once the clues start pointing to a suspect, focus shifts to collaborating evidence that supports those clues. When I first saw this data, my thought was that AWS must know this and their behavior would reflect it. AWS is known for aggressively cutting prices. Lo and behold, AWS announced only three price reductions this year (10+ in 2014, 50+ since inception). This year’s annual AWS ReInvent is also the first one ever where AWS did not announce any price reductions. SVP Andy Jassy commented on this, explaining that AWS is focusing on customer experiences. Was this not also the case in year’s past? Let’s also look at how other public cloud providers are doing. While AWS is the 800-pound gorilla, how much economy of scale is there once one goes beyond a few data centers and dozens or hundreds of customers? The cloud providers all buy infrastructure from the same vendors and margins are low which makes it unlikely any one provider is seeing significantly lower acquisition costs. CenturyLink has announced they are looking to sell their data centers, Verizon is considering the same steps, and the quarterly Rackspace revenue and income has not been good. With these companies obviously struggling in public cloud, why is AWS doing any better from a profitability standpoint? They bought more market share, but if one is losing money on each sale, one cannot make that up with high volume.

Why would someone continue to make this type of investment in a business that has such bad financials? I can come up with a few answers and will respond to each. I would be interested in additional reasons from readers.

  1. AWS is establishing the public cloud market. Traditional IT and private cloud are still much larger markets than public cloud. The public cloud needs to be lower cost than private cloud for established companies to move to the public cloud for anything other than off-site back-up and disaster recovery. Once a company moves to the public cloud, there is a degree of vendor lock-in that makes a move back to private cloud difficult. This approach makes sense only if AWS expects public cloud prices to rise in the future or costs to fall quickly and they will see a positive ROI. With IT infrastructure costs continuing to drop, there will continue to be private cloud pressure as the private cloud gets cheaper and the footprint shrinks. One needs to look no further than the challenges being faced by the large IT equipment providers. The Total Addressable Market (TAM) is shrinking even when one adds the ODM revenue into total market revenue. Yes, cloud is growing and traditional IT market is shrinking at a faster rate. Moore’s law still holds and we continue to see prices dropping faster than the need for increasing capacity.
  2. AWS has established a dominant position in public cloud that is locked up. This would be a sound strategy if it were pursued by either Google or Microsoft. These two companies generate more cash from their non-cloud operations and have mcuh larger reserves of cash looking for places to invest. There are no indications that either Microsoft or Google will exit this market and both companies have stronger PaaS and SaaS plays (higher margin) than AWS’s IaaS play. One can deduce that Microsoft and Google are not using their superior cash positions and asserting pricing leadership by slashing prices to drive AWS from market because they are content to let AWS do it for them.
  3. Investment analysts reward companies for growth in the public cloud. stock price has increased ~45% since they started reporting AWS revenues separately. There is a lot of hype in this market, but the financials do not support the business case. The reported quarterly AWS revenue and income ($1.824B, $391M, 21.4% gross margin) look good, but the infrastructure hardware investment (before one considers related DC builds, deployments, and operations) do not support these rosy numbers. One can depreciate servers over three years to spread expenses, but one cannot continue to spend much more each quarter than one earns in revenue. One especially cannot do this when their two major competitors have much better cash positions, both from cash on-hand and cash generated from operations.

It is unsettling to have an opinion so far from the main stream. I talked with a number of people in the industry about my initial findings. I was surprised how many people had gut feelings that the financials do not work. Their gut feelings were based on information showing that the private cloud is cheaper to buy and deploy than the public cloud for medium and large companies as the cloud providers need to make money. The cloud providers’ costs might be lower, but their prices cannot be lower than a company’s cost if IT is a core competency. Public cloud providers are not making money and an industry consolidation is happening as IaaS is a money losing commodity business with too many players.

I welcome your comments and look forward to a healthy dialog on this topic. With many pairs of eyes on target, I have no doubt we will get to the bottom of this.


[i] IDC Press Release, 03 Dec 2014,

[ii] IDC Press Release, 03 Mar 2015,

[iii] IDC Press Release, 27 May 2015,

[iv] IDC Press Release, 25 Aug 2015,

[v] Gartner Press Release, December 3, 2014,

[vi] Gartner Press Release, March 3, 2015,

[vii] Gartner Press Release, May 28, 2015,

[viii] Gartner Press Release, August 26, 2015,

[ix] Gartner, Magic Quadrant for Cloud Infrastructure as a Service, Worldwide, 18 May 2015, ID:G00265139,

[x] EnterpriseTech, November 14, 2014, A Rare Peek Into The Massive Scale of AWS, Timothy Prickett Morgan,

[xi] IDC Press Release, 16 Apr 2015, Worldwide Cloud IT Infrastructure Market Grows by 14.4% in the Fourth Quarter as Service Providers Continue to Expand Their Datacenters, According to IDC,

[xii] IDC Press Release, 05 Oct 2015, IDC Forecasts Worldwide Cloud Infrastructure Market to Grow 24% Year Over Year in 2015, Driven By Public Cloud Datacenter Expansion,


Disclaimer:  I work for Hewlett Packard Enterprise. Opinions expressed here are my personal opinions, not those of HPE.

  • HPE Cloud
About the Author


Technology entrepreneur with 30+ years experience successfully launching new products and services for small-to-large companies. Now applying his knowledge and experience at Hewlett Packard Enterprise as a Chief Technologist. Also active in the Seattle start-up community. Educated in Physics at MIT and has an MBA from the Wharton School, University of Pennsylvania.

28-30 November
Madrid, Spain
Discover 2017 Madrid
Join us for Hewlett Packard Enterprise Discover 2017 Madrid, taking place 28-30 November at the Feria de Madrid Convention Center
Read more
See posts for dates
HPE Webinars - 2017
Find out about this year's live broadcasts and on-demand webinars.
Read more
View all