[E12.3] Amazon Web Services: monetizing information technology
Cloud computing so any company can pay as they go for the IT services they need
Dear Readers,
In this Profile for Product | Strategy | Innovation I will continue a 3-Part Series on Amazon with each part covering one of its 3 Pillars. These Pillars include the Amazon Marketplace, Amazon Prime and Amazon Web Services in the order that they were launched.
All 3 Pillars play a role.
Each Pillar serves a unique need,
Each Pillar reimagined a key expense to generate significant revenue, and
Cross-Pillar integration enhances the barriers for others to enter growing markets.
In this Part 3, I will cover the 3rd Pillar known as Amazon Web Services, or AWS, which is credited with commercializing cloud computing infrastructure and services.
AWS is a juggernaut that generated $80 billion in revenue for Amazon in 2022. If we valued the AWS business at 10x sales, it would be worth $800 billion based on its 2022 revenue.
I will cover this 3rd Pillar as follows:
AWS: High-Level Overview
High-Performance Computing (HPC)
AI: Generative Models
Amazon / AWS: Response to ChatGPT
AWS: Long-Term Strategy
Conclusion
The original cloud computing concept dates back to the early 1960s when MIT computer scientist, cognitive scientist and artificial intelligence pioneer John McCarthy, PhD stated computer time-sharing technology could be organized with computing power and even specific applications as a public utility like a telephone system or electricity.
The following information was previously shared in the Part 1 Background section on the AWS orgin story.
Amazon was incurring significant expenses maintaining its information technology infrastructure to support the rapid growth of its eCommerce business. Agile teams were also asking to spin up servers and core infrastructure to launch new projects. But these were early ideas that could fail. This created a need within Amazon for on-demand computing resources that were easy to turn on, scale and turn off. Cloud computing and cloud services were an innovation within Amazon to address their own needs. But as Amazon built out what eventually became Amazon Web Services (AWS), they realized agile teams outside of Amazon driving innovation within other corporations and new ventures would eagerly adopt these cloud services.
The actual genesis of AWS came in the early 2000s after Amazon built its eCommerce-as-a-service platform Merchant.com to offer third-party retailers a way to build their own web-stores. Amazon pursued service-oriented architecture to scale its engineering operations. AWS launched the first web service in 2002. This opened up the Amazon.com platform to all developers. Andy Jassy who is the current CEO for Amazon took over AWS in July 2002 and mapped out the vision for an “Internet Operating System (or OS)”. AWS is a key driver of profitable growth and makes up the third Pillar for Amazon.
One of the early implementations of cloud computing was in 1999 when Salesforce.com introduced deploying enterprise applications through a simple website as a service.
Amazon launched AWS in 2002 and Google Docs emerged 4 years later in 2006 to simplify document creation and sharing.
Microsoft entered cloud computing in 2009 with the launch of Windows Azure.
AWS had a substantial lead by the time Google and Microsoft got serious about cloud computing. This lead gave AWS substantial momentum to scale adoption for its unique services.
AWS also benefited from a wide range of internal Amazon uses at scale including streaming Prime Video and eCommerce operations with significant growth to challenge any IT deployment with rapid, ongoing change and chaos. These challenges align with the advantages and flexibility of cloud services.
AWS: High-Level Overview
Cloud computing is based on 3 fundamental concepts.
Provide a service like computing or storage (this is a public utility)
Multiple people share the same computer resource (this is time-sharing)
Access these resources through networking (this is a network)
Cloud architecture has also evolved into 3 major model segments. Each model has its own benefits and key features. AWS supports all 3 of these models.
Software-as-a-Service (SaaS)
Service provider delivers and maintains applications and software to organizations over the Internet.
This eliminates the need for end users to deploy the software locally.
Platform-as-a-Service (PaaS)
Service provider offers a computing platform and solution stack, often including middleware, as a service. Provider delivers the networks, servers and storage required to host an application.
End user organizations can build on that platform to create applications or services. End user oversees software deployment and configuration settings.
Infrastructure-as-a-Service (IaaS)
Service provider eliminates the need for organizations to purchase servers, networks or storage devices by providing the necessary infrastructure.
Organizations manage their software and applications, and only pay for the capacity they need at any given time.
Cloud architecture covers 4 different deployment models. AWS started commercially as cloud services available to multiple third party organizations as a public cloud. Over time they have added other deployment models to support customer needs to scale.
Public cloud architecture
Different organizations access the same infrastructure
Lowers cost through optimized utilization of shared resources
Private cloud architecture
Organizations access their own dedicated infrastructure
Higher cost due to more specialization and exclusive capacity
Hybrid cloud architecture
Organizations optimize cloud architecture with a mix of public and private clouds.
Edge computing extends the cloud closer to end use. This could be where AI has a big impact.
Cost can be optimized with more use of a public cloud when possible and reserving specialization for services requiring a private cloud and edge computing.
Multi-cloud architecture
Mixes cloud services from multiple providers for more selection and optimization of who provides each service
Optionality improves with edge computing and enterprise service providers like Palantir because they can commoditize many aspects of cloud computing
The Cloud Computing market is dominated by the top 3 providers with over 66% combined market share as of Q3 2022 for US$143 billion in combined revenue over the trailing twelve months.
AWS still has the largest market share by a wide margin, but AWS growth has slowed over recent years as MS Azure improves and leverages enterprise deals across many products and services. Think MS Teams vs. Slack, but with less impact so far for Microsoft.
The cloud computing market is expected to reach US$1.5 trillion in annual revenue by 2030 with a compound annual growth rate of 15.7% according to a February 24, 2022 report.
However, these projections do not reflect recent advances in generative AI including GPT-4, ChatGPT, Dall-E and Stable Diffusion and their impact on the cloud computing market in the future. If generative AI tools enable a worker to 3x their productivity, then the value these tools provide to enterprise deployments is substantial.
The IaaS model segment is expected to grow the fastest with end users hiring skilled workers to manage IT infrastructures and reduce deployment costs. However, the SaaS model segment will remain the largest through 2030.
Small and medium-sized enterprises are expected to grow cloud services faster than large enterprise companies.
The hybrid cloud deployment model is expected to grow faster than other deployment models.
Manufacturing is expected to be the fastest growing industrial segment due to the wide range of end use needs. These range from production systems, High Performance Computing (HPC), 3D printing, AI/machine learning, Internet of Things (IoT) and industrial robotics.
Asia Pacific is expected to be the fastest growing global region for cloud computing services.
The 4 foundational services provided by AWS include:
Compute by Amazon Elastic Compute Cloud (EC2). No up-front investment is required. Instead you pay-as-you-go by hours of usage as you scale up or down.
Network by Amazon Route 53 is a scalable and highly available Domain Name System (DNS) service. This provides secure and reliable routing to your resources like websites and web applications.
Storage with Amazon Simple Storage Service (S3) for industry-leading security, scalability, availability, and performance.
Database by Amazon DynamoDB with built in security, backup and restore for an unlimited amount of data while only paying for what you use.
Other AWS cloud services include Sagemaker for machine learning, Amplify to build mobile apps, RoboMaker for robotics, Lambda for services without a server, and Braket for quantum computing research.
AWS provides an exhaustive range of more than 200 cloud services for almost any computing need.
Deployments can start small with only a few core services to manage cost, but then scale to include as many services as needed to support your requirements.
You only pay for what you use.
High-Performance Computing (HPC)
Amazon works with the major chip suppliers like Nvidia, Intel and AMD to keep its server technology on the leading edge. As Amazon deploys more high-performance computing (HPC) infrastructure, AWS is also moving towards using its own in-house chip designs.
The AWS in-house ARM-based chip for HPC is called Graviton3E and will enable Amazon to make HPC more readily available by taking more control over performance, cost and scalability.
A key feature of HPC beyond specialized chips is parallel processing of data from the chip to the server and to cluster of servers. This enables a more flexible supercomputing architecture to meet ongoing computational needs. HPC is well-suited for cloud computing to take advantage of optimization to accelerate the work.
The demand for these services within one company may vary based on the project flow, but is more steady across an ecosystem of multiple companies. As the queue builds and forecasting verifies sustained demand, more hardware is deployed towards these services to scale the base infrastructure.
One key industry for HPC is pre-clinical research in the life science industry for biotech and pharmaceutical companies. Molecular dynamic simulations allow laboratory workflows to transfer research to computers. This allows researchers to investigate how libraries of small molecules interact with a target protein.
This in silico research allows reducing the number of small molecules of interest from tens of thousands to hundreds before entering preclinical studies. This saves time and money when entering these preclinical studies even if it takes some additional time and money to do the in silico research.
And that is just the life science industry. AWS services HPC needs across many industries. And because it already has other cloud computing relationships across these industries, it is much easier for AWS to penetrate the HPC market with additional services beyond its core cloud computing services.
Another domain that benefits from HPC at scale is not just machine learning but deep learning using supercomputer clusters to develop AI-based models and applications. Deep learning is characterized by neural networks with much greater complexity using more discrete node layers to form the network. Tesla has designed its own chips and hardware infrastructure to build one of the most powerful supercomputers in the world called Dojo.
Tesla’s Dojo supercomputer is optimized to train computer vision neural network models for autonomous navigation on electric vehicles and humanoid robots. Tesla may also offer this specialized HPC service to other companies, but AWS can build out and offer more generalized HPC services with on-demand cloud computing to a broader ecosystem of end users.
AI: Generative Models
OpenAI has gotten a lot of attention with their GPT-4 (multimodal input to text output), ChatGPT (conversational text) and Dall-E (text to image) apps based on Generative Pre-trained Transformer (GPT) technology. The mission of OpenAI is to build safe and beneficial artificial general intelligence.
GPT was inspired by the original 2017 research paper on a Transformer machine learning model that connects an encoder and decoder with an attention mechanism. The paper is “Attention Is All You Need” by Vaswani et al. and was published by a team at Google Brain with colleagues from Google Research.
At a very high level, the Transformer encoder processes an input and the decoder generates an output. “je suit etudiant” as an input phrase is transformed into “I am a student” as the output for a French-to-English translation. This mode is text-to-text or chat plus language translation, but audio would enable voice translation either as voice-to-text, text-to-voice or voice-to-voice between supported languages.
The recently launched GPT-4 model is a multimodal transformer that will eventually support both text and image inputs and is the successor to GPT-3.5 that powered the release of ChatGPT. More details will come out about this GPT-4 model, but the earlier GPT-3 model includes 175 billion parameters and was trained with 570GB of data obtained from books, Wikipedia, articles and other forms of text available on the internet.
That translates into about 300 billion words to train the GPT-3 model. These are also called large language models (LLMs) based on the size of the neural network model itself and the volume of data required to train the model. The quality of the output these models generate will be an important factor as the technology evolves.
ChatGPT is the fastest application to reach 100 million unique users (just 2 months) following its launch in late 2022. For comparison, the popular apps TikTok (9 months) and Instagram (2 1/2 years) took longer to reach the same milestone.
Cloud computing providers benefit from generative AI models through the demanding computational requirements to process such vast amounts of data for the original training as multiple versions of a model are developed in parallel. This is an ongoing body of work to advance the technology as fast as possible.
End user applications are then connected to a released version of a model (like GPT-4) to input the prompts that generate the model outputs for use by these apps.
This requires scalable compute services like AWS EC2 to operate this service. GPT-4 is not perfect but these large language models are a proxy for what is likely a key advance for AI.
Microsoft partnered with OpenAI in 2019 leading to a $1 billion investment and then an expanded partnership in 2021. Under these partnerships both companies are jointly building new Microsoft Azure computing capabilities with OpenAI creating new AI technologies to port its services to run on Azure.
Microsoft is OpenAI’s preferred partner for commercializing new Generative AI technologies. Microsoft announced in January 2023 another $10 billion multi-year investment into OpenAI to host GPT technology on the Azure Cloud.
The outcome of this latest partnership is an GPT-powered Bing search engine and Edge browser that will integrate a new sidebar for chat and compose features to enhance the Bing search experience.
Microsoft will offer a new and improved MS Office package that includes the GPT-powered Bing and Edge products and integration within other MS Office applications like Word and Teams. Ads can also be used within the chat and compose sidebar to help monetize this capability.
Microsoft is an important distribution partner for OpenAI. Slack created a new category for messaging across teams and launched in 2013. Daily Active Users (DAUs) for Slack in 2022 reached 18 million. Microsoft launched their Teams product in 2017 to compete with Slack. Teams reached 20 million DAUs in 2019, but scaled to 270 million DAUs in 2022 by bundling Teams in the MS Office package.
Generative AI-powered search by Microsoft and other companies will likely disrupt Google’s dominant search strategy that uses Adwords to monetize clicks to other sites.
Google is responding to ChatGPT with it’s own AI-powered Bard product using chat. Reviews are not favorable so far, so Bard may be rushed prematurely to market just to counter ChatGPT. Google has significant AI resources, but they may not be hungry enough to close the gap.
Meta (formerly known as Facebook) also has significant resources focused on AI to optimize its social media and over 3.5 billion users worldwide across all of its products and services. Meta seems to be pivoting away from the “metaverse” towards generative AI as the next big thing.
Amazon / AWS: response to ChatGPT
How Amazon responds to generative AI may be different from other cloud providers. Amazon already leverages AI-based technologies at scale throughout all its operations with recommendation engines, autonomous navigation for warehouse robots, forecasting, and many other use cases.
Amazon has 200 million Amazon Prime members (consumer end-users) searching for products that could expand into more generalized search for experiences and fact finding with generative AI and chat.
Amazon Marketplace has many third-party sellers (business end-users) who might be interested in adding more cost-effective creative resources with generative AI to help promote their products and services.
Early adoption of generative AI at scale outside of Amazon’s core operations will likely be business-to-business applications. Amazon can build new generative AI-enabled creative services for third-party sellers in its Marketplace who promote their products and brands.
However, AWS as a cloud computing platform provider plays a different role regarding generative AI when compared to the rest of Amazon. AWS works with multiple tech ventures who are advancing the leading edge of any trending technology with the flexible cloud services they need.
This yields multiple AI-based models (1st order) to enable the AWS ecosystem of developers and companies to integrate these models into their products and services (2nd order). End-use by consumers and businesses becomes the 3rd order.
AWS has multiple partnerships with generative AI leaders like Hugging Face with its leading multi-lingual language model technology and AI21 Labs with its OpenAI GPT rival Jurassic. AWS can continue to deepen its collaborations with these and other companies to accelerate the adoption of their technology.
U.K.-based Stability AI is another competitor to OpenAI.
Stability AI built its generative AI model using an AWS cluster with 4,000+ Nvidia A100 GPUs and multiple AWS cloud services like Sagemaker for machine learning.
Stable Diffusion is Stability AI’s technology focused on text prompts as input to generate images as output.
StabilityAI announced in November 2022 that it was doubling down on AWS as its preferred cloud provider to build and train leading AI models for image, language, audio, video and 3D content generation. These could be separate models or multi-modal models for more flexibility.
Stability AI’s partnership with AWS now provides one of the world’s largest supercomputers: the Ezra-1 UltraCluster.
StabilityAI has also discussed positioning products and services tailored for unique industries like health care and creative studios. Generative AI in the future could create personalized treatment plans for individual patients and movies to augment written work.
These partnerships might add more value to AWS with many narrowly focused AI-based tools across specific industries than a more generalized audience with Microsoft’s Office products. However, Microsoft is better positioned for a first wave of large deployments using enterprise-wide licenses for generative AI-enabled MS Office products.
Amazon and AWS seem very well positioned to capitalize on generative AI internally and across their ecosystems.
StabilityAI is training and running multiple generative AI models at any time. One area of R&D is model compression.
A 2GB model of Stable Diffusion can run locally on computers and even an iPhone. Model training and compression would still require cloud computing, but operating the model locally would offload the compute services from the cloud.
Google Search and its AdWords platform may have the most to lose if Microsoft is successful with its AI-equipped Bing and Edge products . Google recently announced Bard to pursue its own version of generative AI for chat-based search.
Amazon will focus on the needs of its partners and customers and innovate solutions to wow them. If they do that exceptionally well, then it doesn’t really matter what Microsoft and Google offer. That has always been the Amazon Playbook. Focus on the customer and their needs.
AWS: long-term strategy
HPC and AI-powered services will boost growth for a cloud provider if they can pull away from competitors with unique and proprietary offerings.
AWS can use generative AI-based models as higher order “servers” to drive intelligence across end-user applications that use the other core AWS cloud services to deliver all the required capabilities.
Without unique, proprietary HPC and AI-powered cloud offerings, core cloud services could become less differentiated over time. Growth would slow and margins would compress with more competition.
AWS is still the market leader for cloud services, but as Microsoft potentially closes the gap with generative AI technology incorporated into MS Azure and Office products, the question is what is the long-term strategy for AWS in a competitive marketplace when it no longer has a 7 year lead.
Product Positioning will be key to maneuver:
Google, Meta, Amazon and now Microsoft monetize products with advertising (we as end-users are actually the product sold to brands by opting in to free services). [this advertising model is most at-risk with generative AI]
Microsoft has partnered with OpenAI to incorporate its technology into core Microsoft products to license generative AI features to business and consumer end-users. [this “up-sell” model is about to explode by leveraging generative AI, but is a 100% bet right now on OpenAI]
AWS is the preferred cloud platform for an ecosystem of third-party developers and companies to develop products and services for end-users. [this model continuously optimizes for best-in-class and monetizes information technology infrastructure]
AWS generates significant value for Amazon as one of its 3 Pillars, but it may be better for AWS to eventually spin out of Amazon as a stand-alone, hyper-focused company on cloud computing for a worldwide ecosystem of third-party developers and companies reaching billions of end-users.
A spin-out is also strategic if AWS wanted to accelerate its pace of advancing HPC and AI-based models beyond what might be feasible within Amazon that has broader objectives, advantages and scope.
AWS already contributes over 2/3 of Amazon’s realized market cap. A good proxy for what is possible for AWS might be ASML. The Dutch tech companies ASML and NXP Semiconductors were previously separate business groups within Philips.
Both businesses spun out of Philips at different times but both scaled into transformational leaders in their respective domains as stand-alone companies in the semiconductor industry. ASML is now about 15x the market cap of Philips.
If GPT technology and Azure become burdened within Microsoft, an AWS spin-out could capitalize with a focused strategy if generative AI models are a key opportunity of the century vs. an over-hyped, near-term fad.
Amazon could still access contracted IT services through AWS and also benefit from the upside of its commercial success as an equity holder in the AWS future. But a minor divesture of 10% of AWS to the public markets could provide more optionality for both Amazon and AWS.
Amazon could then align more divesture of AWS over time through the public markets if initiatives to enhance the core Amazon businesses materialize.
But even if the probability of an AWS spin-out is low, what would Amazon do to fill the loss in revenue and free cash flow generated by AWS?
How about replace the estimated $1.5 trillion TAM for cloud services in 2030 with an even larger TAM like $28 trillion for global financial services or $12 trillion for global health care.
Amazon is already in health care following the Pill Pack and One Medical acquisitions, but it could take the capital raised by partially divesting some of AWS to accelerate further penetration into health care.
Health care as an end market could create or replace a Pillar. Otherwise, with less focus and investment, it likely becomes a significant business group within Amazon Prime to benefit 200 million members including some markets outside the U.S.
Fintech within financial services would provide a more focused IT function more aligned with commerce for businesses using Amazon Marketplace and personal finance for consumers using Amazon Prime.
If Block can build out the Bitcoin payment rails for software developers to build payment processing infrastructure, and Cash App could be rolled into Amazon Prime and Square could be rolled into the Amazon Marketplace, this could be a really attractive fit.
However, such a fintech/eCommerce combination may not pass regulatory scrutiny even if an AWS spin-out creates a huge void to fill.
Health care may be the more viable option just because of the deflationary force Amazon can provide to control costs while also hopefully improving the quality of care.
And health care would again take a key expense with employee health plans provided across all of Amazon, to turn that into a revenue generating business.
Conclusion
One of the defining attributes of Amazon has been building a strategy around 3 Pillars including Amazon Marketplace, Amazon Prime and Amazon Web Services.
Each Pillar provides unique value, but with cross-Pillar synergies and integration, Amazon builds an economic moat around these 3 Pillars.
Another feature of these Pillars is taking an expense, building services to create value, and generating revenue by selling this advantage to individuals, families and other organizations.
AWS was built to deliver more reliable, secure, scalable, on-demand IT services within Amazon.
AWS launched its original services in 2002 to sell to other organizations. AWS generated $21.4 billion in revenue in Q4 2022 with 20% year-over-year growth and $80.1 billion in 2022 revenue.
That is called monetizing information technology!
How Amazon responds to generative AI will be a key decision for its internal- and outward-facing operations and services, but could also have a big impact on AWS. Microsoft has taken decisive actions to take advantage of generate AI to close the gap it has with Google across search and advertising.
The most decisive strategy for Amazon would spin out AWS for a hyper-focus on information technology innovation. This will help accelerate creating and training the most advanced generative AI technologies for third-party developers and end-user applications.
Best,
Stephen
I’m long AMZN, PLTR, SQ and TSLA mentioned in this Profile. Nothing in this Profile is intended to serve as financial advice. Do your own research. The opinions and views expressed in this newsletter are those of the author. They do not purport to reflect the opinions, views or policies of any other organization, company or employer.