Google (NASDAQ:GOOG) (NASDAQ:GOOGL) recently began expanding its cloud service business. This has prompted various articles trying to estimate Google's chances of succeeding in this space. A particular piece by SA contributor Bert Hochfeld caught my eye due to the variety of its claims. I believe many of these claims to be wrong and based on a lack of understanding of the makeup of the cloud market. In this article, I illustrate by example what brings me to these conclusions.
What follows from nothing?
Let me briefly summarize the above article's claims. First, the author claims to know nothing about cloud computing.
But the issue is going to be one of scale for Google. The only real reason that I can see for Google to gain market share is based on costs. It does have lower and simpler prices than either Azure or AWS. It has huge disadvantages in terms of geographic coverage, its performance is probably the same as AWS and I really can't even begin to comment on whether it has better or more features than Amazon. Just reading about it is the same as reading about two kids playing or fighting in a sandbox.
I have a problem with this paragraph, especially the ending. It's okay to not know details about everything we discuss, but we should not say they do not matter or be derisive towards them. From not knowing something follows nothing. In the comments, the author continues to claim:
I would add, that when companies choose to compete on price it is usually because they have nothing else material to compete with. You can readily go through Google's marketing material to determine that at this time, price is their big selling point. Sometimes it is price/performance to be sure but that is a dangerous slope down which to go as most of the consultants felt that latency was an issue where distance was a factor.
I do not purport to understand some of what all of the consultants had to say when evaluating the positives and negatives of the services as they exist today. I really don't purport to have written anything close to a cloud buying guide. If a company cannot articulate a marketing message that can be understood by normal people, then they do not have a marketing message at all. If the message is that we perform better, than why put a 40% discount on it.
If normal people cannot understand a marketing message, there is no marketing message at all? A marketing message needs to reach the targeted customers. The author of the comment is presumably not the customer of Google's cloud services. Whether he understands the marketing is hence not really informative. What is relevant is whether customers understand a marketing message. CTOs. Software architects. Developers. Researchers.
Google's cloud advantage
The next thing other commenters have pointed out is lumping together all cloud services into one segment. The author goes on to say:
There's nothing particularly different about the services that Google supplies compared to Amazon or Microsoft.
This is rather inaccurate as well. Yes, whether your software containers run on Azure, Google or Amazon (NASDAQ:AMZN) is largely irrelevant. In fact, in the future, it is likely that your virtualized containers will move between clouds (multi-cloud management is already a growing market segment). This is also missing that Google is already innovating on the infrastructure level with custom machine types and minute level billing.
Saying that a late mover is hopeless in a market has been shown to be wrong many times in recent technology history. The second reason this is wrong is that what is going to matter moving forward is not the commoditized infrastructure level, but the software services running on it. Amazon for instance is not even a competitor on deep-learning driven machine learning services.
Microsoft (NASDAQ:MSFT) is doing okay here and has announced interesting initiatives, but Google is still releasing best-in-class tools. This issue is a bit harder to appreciate from the outside so let me illustrate it.
A few months ago, Google released TensorFlow, which is its internal machine learning execution engine. It allows you to define computation graphs and then takes care of executing them on arbitrary devices. That might sound trivial to you, but a large part of building modern machine learning driven applications is to implement a learning pipeline that best utilizes the available computation resources. Training a single model might take weeks. This is mainly driven by GPUs. Using TensorFlow, you just have to name the devices you have - which can be anything from a distributed cluster to a mobile phone - and TensorFlow will take care of executing your algorithms on these devices. This is huge because you would otherwise spend weeks writing custom code for your specific devices.
As a side note, the topic of deep learning on GPUs is also very relevant to the Nvidia (NASDAQ:NVDA) versus AMD (NASDAQ:AMD) discussion currently being fought out on Seeking Alpha, but I will weigh in on that on another article.
Just for illustration, TensorFlow came out last November. Other machine-learning libraries like Theano or Torch have been around for years. Yet, this happened:
Along with the release of TensorFlow 0.8, Google highlights how enthusiastic the public reaction to its software has been. TensorFlow is the most popular machine learning framework on code repository GitHub, and the most forked project on the site in 2015, despite only being released in November of that year.
Google recently announced to offer TensorFlow as a cloud service. This is just one example of a best-in-class product for which there is simply no competition. Neither Microsoft, Amazon nor IBM (NYSE:IBM) has anything even close.
The data center myth
Further, the issue of the number of data centers is hugely blown out of proportion on contributors here who do not seem to know much about distributed databases, yet use this number as a proxy for competitiveness. The fact that Amazon has many more data centers has no impact on the quality of service for a typical cloud customer. First, partitioning your service over many data centers does not make sense at all unless you are running them as completely separate operations with separate databases (e.g. due to local regulatory frameworks).
You do not magically get "copies" of your service in all these regions. Amazon's autoscaling feature only works within one region. Amazon's typical user will set up their service in their main customer region. They will replicate it locally in their database cluster and maybe in 2-3 other availability zones (or even regions with much more effort) to survive data center failure. Then they would use forward and reverse proxy caches, content delivery networks, ISP caches and client caches to deal with geographical roundtrip latency. Claiming that customers will always get better latency from Amazon just because AMZN has more regions is a rather naive view and only true for a few large customers who actually set up many dedicated versions of their service. In fact, even if latency is better for AWS depending on where your clients are, Google might have better throughput. This is not surprising if you have read a few of Google's papers on hyperscale distributed databases (Google Spanner/Megastore). I view this issue as minor going forward, especially since you can expect Google to aggressively add data centers in strategic regions.
Google does not have a scale problem. It is releasing best-in-class tools in the cloud. TensorFlow is illustrating the appeal of Google's products even as a late mover. One of Google's infrastructure leads has suggested cloud services will eventually make more money than ads for the company. I would tend to believe him.
Another issue completely ignored on Seeking Alpha thus far is the trend towards containerized services. In a nutshell, it means you define services as building blocks that live in virtual software containers, independent from the infrastructure, which is typically its own layer of virtualization. What this means is that running your application in containers makes it really easy to switch cloud service providers. You define your container once and can then launch it anywhere (this is what a company called Docker does). This trend should help Google gain market share since it makes the services on top of the infrastructure (e.g. Google's data, language and machine learning services) more important for new customers than the infrastructure. In short, from an insider's perspective (having built and used cloud infrastructures myself), I have every reason to believe Google will be able to aggressively take market share both on the infrastructure and the software-as-a-service level.
Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.
I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.