[U11] Update: Machine Learning Business Models
Vertical, Horizontal & Hybrid business models to advance this key technology
Dear Reader,
This week in Product | Strategy | Innovation is an update on machine learning. This is a rapidly advancing branch within the artificial intelligence (AI) field that uses big data and algorithms to imitate the way humans learn to gradually improve accuracy. These algorithms are trained to classify data or make predictions to uncover key insights. The business opportunity is to monetize the associated data science through products and subscription services.
My objective for this update is to add to the due diligence to inform profiles on the AI service company Palantir and edge computing company Cloudflare. The update is also relevant regarding Google, Amazon, Tesla, Nvidia, Taiwan Semiconductor, IBM, Salesforce, Facebook, Square and other companies that leverage machine learning to advance their innovation programs. The banking, energy, transportation and healthcare industries lag. These are the target for disruptive AI ventures.
“Hey Siri, play ZZ Top La Grange.”
The origins of machine learning date back to about 1959 when Arthur Samuel at IBM coined the term based on his work in the field of computer gaming and artificial intelligence. IBM clearly saw the business opportunity for computer systems providing automation and decision support.
Models for machine learning include artificial neural networks, decision trees, bayesian networks, genetic algorithms and many other methods that have improved over time. But 2012 was a key milestone in the evolution of machine learning as discussed in a prior update on autonomous navigation with computer vision enabled by deep learning.
Neural networks have been used for decades to mimic multiple connected neurons in the human brain. Deep learning as demonstrated in the 2012 ImageNet competition with AlexNet takes neural networks even further with about 10 network layers, inexpensive computer hardware optimized to process image data fast and over 1 million labelled images to train models.
Open source software is also important to make the tools readily available to experiment with deep learning.
Computer vision and natural language processing have emerged as two key use cases for machine learning. Pattern recognition would be a more generalized application. Easy access to billions of publicly available photos and audio files have also contributed to these advances in machine learning. Open source software is also important to make the tools readily available to experiment with deep learning.
Many tech companies are now racing to build out the machine learning platforms to dominate business opportunities that will increasingly use these capabilities in the future. This update will explore the primary business models used by companies pursuing machine learning. Three business models seem relevant at this early stage:
Vertical Integration - Objective: Build competitive advantage using existing big data within the company to scale the data science team, tools and machine learning platform to drive adjacent value for core products and services. Example: Google Photos automatically organizes all the people & pet photos on your phone.
Horizontal Integration - Objective: Build competitive advantage using other companies data science teams and big data to drive scale for a machine learning platform. Example: NVIDIA Tensor Core GPUs sold to data centers and edge computing businesses.
Hybrid - Objective: Leverage capabilities built through vertical integration to offer products and services to other companies to further enhance competitive advantage for a machine learning platform. Example: Amazon Prime improving service to customers with demand prediction and then offering these same machine learning tools to customers through Amazon Web Services (AWS).
Commercial products and services using machine learning require a viable ecosystem with abundant talent, tools, traction and capital. Beyond the business models discussed, a critical component for machine learning success is building out the talent base to advance and scale the technology. With deep learning really taking off just in the last decade, it means new and retooled computer scientists and engineers are needed. For this reason, established tech companies pursuing the vertical integration business model are really critical to accelerate the build out of the talent and tools. Google and Facebook are great examples.
Google has an AI initiative that can be found at ai.google.com for access to open source tools and online learning. Machine learning is a key focus of the content. Google is known for TensorFlow to build, train and deploy machine learning models. Google Research also includes the Google Brain Team to focus on machine learning and People + AI Research (PAIR).
Facebook AI Research (FAIR) also plays a significant role contributing to open source tools and online learning for machine learning. In fact, PyTorch is one of these open source tools. Tesla has mentioned at conferences that it is using PyTorch to build its Dojo AI training platform. Top talent will work at companies like Google and Facebook to build out the machine learning platforms to advance the products and services of these companies, but these companies also realize the network effect of creating the open source tools and online learning content for everyone. Open source strategies allow hiring to focus on talent that is already trained and ready to contribute day 1 versus hiring and on-the-job training over time to build out competencies on proprietary systems.
IBM has Watson, Microsoft Research has its own AI initiatives and Amazon announced earlier this year that it is building a Machine Learning Research Center with USC. These initiatives are also critical to fund research and train talent. Governments can also play a key role. US President Joe Biden and his team launched the National AI Initiative Office earlier this year. AI touches national defense, cybersecurity, labor, higher education, policy and regulation, so coordination between the public and private sector is critical.
Vertical integration provides a first mover advantage when there are clear use cases and big data are available to train deep learning models
Netflix suffered from subscription churn with its movie streaming service. Netflix Research vertically integrated machine learning into its operations to improve the personalization of its video streaming service through its recommendation engine. This reduced churn and saves Netflix an estimated $1 billion a year.
Google introduced Google Photos in May 2015 to classify and organize photos. Users can search for anything in photos. The service has a free tier that will end in June 1, 2021. After that date, photos will contribute to the 15 GB free limit across associated Google products, but Google Pixel users will maintain the free tier. So machine learning applied to photos becomes a strategic asset to monetize services. Google already had access to billions of photos in the public domain and its services to train its deep learning model. Google reported in 2020 that 28 billion photos and videos are uploaded to the service every week with about 4 trillion photos stored at that time.
Amazon released Alexa as a virtual assistant in November 2014 using machine learning to understand human voice commands and requests. This enabled the chatbot concept to evolve into 2-way voice communication with a device. Amazon Alexa supported products like the Amazon Echo allow skills to be added to Alexa like language translation, to-do lists, order products, set up shortcuts and timers, control devices, etc. Some analysts estimate Amazon Alexa related revenue with device sales, paid skills, etc. could reach $19 billion in annual revenue by 2021.
Horizontal integration expands deep learning to more use cases when big data are still available to train models.
Nvidia develops and commercializes semiconductor chips for graphics, gaming, high performance, edge and data center computing applications. The company was a clear first mover for horizontal integration using graphics processing unit (GPU) chips to build hardware optimized to train deep learning models with big datasets. TensorFlow and PyTorch are GPU-accelerated so data scientists can just build their deep learning models without any GPU programming.
Nvidia provides deep learning software development kits (SDKs) to further integrate Nvidia GPUs into deep learning projects. As cloud services and data centers have evolved to support both vertical and horizontal deep learning business models, Nvidia is right there to seamlessly integrate their GPUs into the hardware. Nvidia is also leading edge computing applications where deep learning moves into a distributed network to reduce the latency of IoT applications. And then the edge can be taken all the way to devices with neural processors for real-time control. Nvidia and Cloudflare recently announced a partnership to integrate Nvidia GPUs into Cloudflare worldwide distributed edge computing hardware.
Box is primarily a business-to-business data and collaboration service provider. Box can offer HIPAA-compliant data storage solutions to companies in healthcare and document collaboration services across multiple stakeholders. The company also offers third-party application integrations and Box Skills to add machine learning services to customer data stored with Box. So customers who store data on Box can also add IBM Watson, Google Cloud Services, and other services to auto-label images, extract data from contracts, etc. So these are additional services leveraging machine learning for incremental revenue on top of the core data storage subscription.
The evolving opportunity for deep learning horizontal integration appears to be organizing huge customer datasets that span public and private cloud systems and then using a deep learning platform developed across multiple projects to build out a customs application for a specific use case. Government applications were the early adopters for machine learning and are a key source for long-term contracts. Banks, pharmaceutical companies, hospitals, insurance companies, airlines, etc. all have needs for deep learning to improve their operations, but likely lack the resources to tackle scaled vertical integration even with abundant open source tools. That is what I will explore further as I dive into Palantir and Cloudflare who sell their IT services to other companies.
A Hybrid business model was also mentioned to commercialize machine learning. This is more a combination of the first and second models versus something truly unique. It is assumed this would start with Vertical Integration to apply machine learning within an organization using the talent and tools developed to do so. But then when commercial opportunity is identified outside the company, these same services could also be offered to other companies. Amazon and Google are probably the best examples through their cloud services. It is straight forward for these companies to add machine learning capabilities developed for their own applications to the cloud services they offer elsewhere. They also already have teams in place to deliver business-to-business sales and services to other companies. So machine learning just becomes another offer.
Best,
Stephen
Nothing in this post is intended to serve as financial advice. Do your own research. I’m long PLTR, NET, TSLA, GOOG, AMZN, IBM, CRM, BOX and SQ mentioned in this update.