Dynamics Of The Technology Ecosystem

Context 

The last two decades have completely changed the technology (tech) sector. Cloud computing, smartphones, apps, and high wireless data speeds are taken for granted. These developments have created a dynamic tech ecosystem. 

Figure 1: Technology Ecosystem abstraction with examples in each layer 

Having worked with some very smart people to develop and market new products in all parts of the tech ecosystem - Networks and mobile phones (SIEMENS), Network Operators (T-Mobile), Semiconductors (National Semiconductor), Algorithms/Artificial Intelligence (Audience), Cloud-based Applications (EVER, Moki), Blockchain and Consumer Hardware (Open Garden) - gives me a unique perspective on how the technology ecosystem works and how it might evolve. 

In this article, I share my view on what constitutes the technology ecosystem, what are the primary business drivers depending on a company's position in the ecosystem, and how the ecosystem might change with Artificial Intelligence (AI).

This technology ecosystem can be abstracted to four layers supported by semiconductors (tech stack) - Cloud (Storage and Compute), Network, Application Platforms, and Applications (Figure 1). Except for Applications, all the other layers are directly dependent on semiconductors for many technological improvements. Let’s consider how a half square inch icon on your phone is powered by the layers of this technology stack. Take Uber as an example. It operates in the Applications layer and on top of multiple Application platforms (based on iOS and Android). The distribution of the apps happens mainly through Apple’s Apple Store or Google’s Play Store. To use the Uber app, your phone has to be connected to a Network provided by wireless carriers like Verizon or AT&T (The network operators don’t develop any new technologies so I have kept them out of the stack. Examples of network players are network equipment manufacturers like Ericsson, Huawei, etc.) Uber uses Cloud services, provided by Amazon or others, to compute and tell you how far away the driver is and how much the fare will be. All your history is also stored in the cloud. The cloud, the network, and the applications platforms get better, cheaper, and faster with new semiconductors (provided by ARM, Broadcom, etc.). 

Google and Apple operate in all layers to various degrees. 

This abstraction can be applied to streaming platforms (SmartTVs, AppleTV, Roku, etc.) and personal computers (Mac, PC, etc.). Both streaming platforms and personal computers with underlying OSes and App Stores/internet browsers act as Application Platforms as shown in Figure 1. 

Competitive Advantages 

To win in any layer, the players have to offer something new that is better than what already exists in the market (Relative Advantage). However, each layer has its own unique drivers of success. Let’s see what they are. 

Figure 2: Primary business drivers for each layer 

Applications

All the apps you see on your smartphone, personal computer, or on your SmartTV are in the Applications layer. To win in this layer, the players have to focus heavily on User Experience (UX) and continually enhance the UX with the Application Platforms upgrade. Another unique business driver in this layer is the Customer Acquisition Cost (CAC). It has become very expensive to get customer attention, the players have to continuously find ways to lower it. You might have a good product that offers a relative advantage but if you can not sustain CAC then you are unlikely to succeed. This is the most dynamic place in the technology world where new companies go in and out all the time. However, this layer leverages the tech developed in the other layers and finds new ways to apply technologies. Except for big-tech like Facebook, Google, and Apple, most companies don’t develop new technology that becomes available  for use by other companies in the ecosystem. New ways of doing things still spread because employees move around from one company to another. Sometimes UX itself can be a relative advantage. For example, Slack and Zoom are successful mainly because of the UX they provide. 


A lot of what is happening in the Applications layer has become possible because of Application Programming Interfaces (API).

Application Platforms

It is tricky to have a definition for this layer that works for all applications.  In the mobile world, is it the smartphone, the Operating System (OS), or the App Store? I would define it as the combination of hardware that runs the OS that runs the application and the application store where apps are bought and sold (Figure 3). So both Google and Samsung are in this layer. Google provides the Android OS (technically it is open source) and the Play Store (where you can purchase apps) and Samsung provides the Smartphone hardware. Of course, Apple plays here with iOS (OS), App Store, and iPhone (hardware). This layer also provides the software tools to the Application layer to develop and test the application software. To be successful in this layer, the players have to develop a developer ecosystem or use an existing developer ecosystem. Scale and developer ecosystems are interconnected. Developers won’t be attracted to an ecosystem that does not have scale. Huawei and Xiaomi have been relatively successful in its own developer ecosystem in China. To do that they had to create their own OS (based on Android). 

Figure 3: Parts of an Application Platform

There are billions of devices which constitute Application Platforms in this layer, so the semiconductor companies like ARM, Broadcom, and Qualcomm do well as suppliers to this layer. The semiconductor companies invest in R&D to make their products better so that the platform companies would pay them for the upgrade.  For example, Apple produces a new iPhone every year. A semiconductor product x might be included in iPhone 11 but if x does not improve its performance for iPhone 12 then it would not be included in iPhone 12 or the price of x would have to be reduced significantly or Apple would give that business to another semiconductor company with a product similar to x. Since it is Apple, they might develop their own product similar to x, and, if the product x is not ready in time when Apple is making the design decisions then that semiconductor company loses revenue for 200M+ units (number of iPhones sold every year). So, the two key things a company needs to be successful in this layer 1) Continuous technology improvements and Intellectual Property (IP) development. Both are the result of research and development (R&D) 2) Time to market (how long it takes to develop a new product). 

Apple realized the importance of semiconductors to its business and shortly after launching the iPhone moved to developing its own application processor (by licensing technology from ARM), a type of  semiconductor. Last year, Apple bought Intel's modem chips business. Qualcomm is the dominant supplier to all other high-end smartphone manufacturers. 

Semiconductors (chips) is a diverse layer and has different types of chips for different functions. Examples of type of semiconductors include application processors, baseband processors, graphics processors, microcontrollers, field-programmable gate arrays (FPGAs), memory chips, analog chips, digital signal processors (DSPs), etc. Many of the big semiconductor companies offer their own hardware development kits and tools like Nvidia CUDA, Qualcomm Snapdragon, and ARM Pelion. Raspberry Pi sold more than 30M units last year.

. 

Network 

This is the layer that makes up most of the the actual physical Internet. A good way to think about the network is in three concentric circles (Figure 4) - Access, Edge, and Core. Access is also called the “last mile” where the technology used to transmit/receive information from the Application Platform might vary. For example, your mobile phone uses wireless technologies to transmit or receive information and your SmartTV might use cable broadband or DSL to transfer information. Information goes to the Edge after Access.  Edge is where a lot of information is stored and a lot of cybersecurity is applied. Information that is frequently accessed is stored on the Edge so that it can be accessed faster. For example, when you are streaming a Netflix movie, most likely it is stored on the Edge of the network (See content delivery networks for details). After the Edge, many things go to the Core where the network function is to route this information cheaper and faster. The Internet is a network of networks. It works because players in this layer follow standards. How do players differentiate if every player in this layer has to follow standards? Mainly with cost and time to market. Huawei and ZTE have come out of nowhere in the last two decades and now dominate this layer because of cost. Lately, Huawei is believed to be ahead of other players for 5G (fifth generation)  wireless networks. Of all the value created in the tech stack, the network players get the smallest share of the value created. 

Figure 4: Network abstraction 

Google has tried playing in this layer, mainly as a network operator with some software based network elements, with Google Fiber, but gave up. Currently, it is running Google Fi which is a virtual network leased from T-Mobile, Sprint, and US Cellular. Large Internet companies like Google, Facebook, etc. run their own private networks, but  that does not mean that they are playing at the layer. Playing in a layer means they sell products or services to others. 

There are a lot of other players in this layer e.g. Juniper, Cisco, Ciena, etc. I find it fascinating that a lot of information moves through light with optical networking. 

Semiconductors play a big role in the network layer. Just like in the Application Platform layer their success depends on continuous technology improvements (R&D) and on time to market. 

Cloud (Storage and Compute) 

Cloud has made life much easier for all companies. Software as a Service (SaaS) business model works really well for startups and for scaling data-heavy applications. I am not sure if any of the popular apps like Uber, Twitter, DoorDash, etc. would have started without the availability of data storage and computing power (Store and Compute) in the cloud that does not require huge Capital Expenditure (CapEx) upfront. There are three main global players in this space: Amazon, Google, and Microsoft. China has its own Cloud players like Alibaba and Baidu. Amazon started this category and now its cloud services (AWS) bring most of the earnings to the company. Ten years ago, it was unthinkable that big companies would willingly give their data to another company to manage. Startups mainly use AWS or Google. Microsoft has been more successful with bigger companies because in most US companies IT (Information Technology) infrastructure is provided by Microsoft. Facebook has its own cloud operations. To be successful in this layer, the players need scale, low cost, and developer tools;all three of these components are interconnected. Data storage and Compute capability in the cloud is helping artificial intelligence (AI) to be widely adopted. 

Cloud has changed how software is developed and deployed. These days, Microservices, Containerization, and CICD are common ways of developing and deploying software. NoSQL and In Memory databases are enabling a lot of real-time applications and big-data analytics we have become used to. This is a big shift from relational databases which was the most common way of storing data in the pre-cloud era.

For certain applications, Cloud players offer application platforms as a service (PaaS).

Semiconductors play a big role in the network layer. Technology improvements (R&D) matter more than time to market in this layer because clouds are ever-expanding and buying does not happen in cycles like in Application Platforms and Network layers. 

Where do we go from here? 

Following trends will continue: 

1. Businesses will continue to become digitized 

2. There will be more and more data generated, captured, and analyzed

3. There will be more and more machines that will communicate with other machines 

4. There will be more autonomous systems (AI)

5. There will be more new applications 

6. Cost of data transfer and storage will continue to decline 

7. Speed of data transfer will continue to increase 

8. Compute speed will continue to increase (with diluted version of Moore's law, parallel processing, ASICs, and many other ways)

9. Compute cost will continue to decline (with diluted version of Moore's law and many other ways)

Advances in Quantum Computing or some new technology might change the entire paradigm. How Blockchain and Crypto change the tech stack remains to be seen. Government regulations might derail the current evolutionary path of tech. 

If there is no disruption with new paradigm-shifting technologies or regulations, how would these trends change the tech stack? Nobody knows but the following are a few ideas on what can materialize:

Application Layer 

The dynamism in this layer will continue. We will progressively see more  applications. Some players might become big enough that they build their cloud infrastructure just like Dropbox did. We will start seeing applications that are for machines (IoT - Internet of Things, autonomous driving, etc.). We will also see more applications using AI. 

Privacy-related regulations might make targeted-advertising based business models less profitable. Companies like Facebook might be required to build and manage products that let people be responsible for their own data which incur additional costs. A lot of people aren’t tech savvy, most are already overwhelmed, and let’s face it most of us are lazy by nature. If you make people responsible for managing their own data, not much is going to change. 

Fintech and Healthtech are thriving. Apple and Google have initiatives to get into the Healthcare and Financial Services markets. The market sizes for these two industries  are in trillions of dollars. If the regulations don’t interfere, they have a good chance of succeeding given their Smartphone Application Platforms dominance and balance sheets.  

Wearable devices (Apple Watch/AirPods type devices and other new categories still in development) as accessories to smartphones would continue to expand.

Autonomous vehicles, Robotics and biotech are very interesting areas to watch for new applications. We will continue to see expansion of tech into new industries.

Application Platforms Layer 

It would be interesting to see if a new Application Platform develops outside of smartphones, streaming platforms (SmartTV, streaming boxes like Roku, AppleTV, etc.), and personal computers.

It is important to note that most apps are free and many make money with advertising (like Facebook). Application stores (Figure 3) make money from apps that are paid and the ones that sell digital goods. For example, a mobile app like Tinder that offers paid services has to give part of the revenue to the application stores. On the other hand, Amazon does not have to share revenue for any physical goods it sells through its mobile apps but for digital goods (like a movie), it has to. In case you were wondering why you can't rent movies on your Amazon Prime app, Amazon avoiding revenue share is the reason. There has been a backlash against Apple and Google charging 30% of revenue on all digital goods (15% on subscription services) on their application stores. We will see if anything changes.

For Smartphone Application Platforms, two players dominate - Apple and Google. Both operate on the Cloud and the Semiconductor layers as well. That gives them the ability to leverage AI capabilities. 

Chinese Smartphone hardware manufacturers (Huawei, Xiaomi, Vevo, Oppo), which currently operate their own application stores, are coming together to create their version of Google Play Store. I hope that a new player emerges with a new OS. However, it is highly unlikely to happen in the near future. 

Artificial Reality (AR) might create new Application Platforms. There is a good chance that we will see new AR applications in the next few years. I think that the chances are low for a new player to do that because an AR product like Google Glass makes more sense as a companion to a Smartphone or another computing device in the vicinity. 

Virtual Reality (VR) might create new Application Platforms. However, applications of VR outside of gaming and entertainment are limited. There might be niche applications like remote surgery but I don’t see VR becoming widely adopted anytime soon. 

Miniaturization and cost reduction in sensors is driving a lot of devices to be connected to the internet. There are more connected IoT devices today than there are people in the world and the the difference will continue to get bigger. IoT and Machine to Machine applications might bypass the Application Platforms and connect directly to the network. With Smart Speakers like Amazon Alexa and Google Home it is happening already. Both Alexa and Home leverage the Cloud for third party application development and use free mobile apps purchased through the Application Platforms for the device setup and management.

Figure 5: Where will AI live?

Network Layer 

There might be new entrants in this space. Network layer used to be customized hardware, but now the hardware functions are moving to software  allowing the hardware  to be commoditized. There are a lot of IoT applications where companies can create their own private networks bypassing the wireless service providers.  In the US, wireless spectrum (CBRS) is easily available for these applications. Companies can use commoditized hardware for wireless communications and use software for network function. Amazon is already offering many network functions. 

Cybersecurity is the biggest growth opportunity in the network layer. 

Network Operators (AT&T, Verizon, etc.) will continue to operate dumb pipes but the pipes will be bigger and faster. There was an attempt by the operators to offer Edge Cloud services but it did not go anywhere. 

Artificial Intelligence

In essence, how most of the AI (supervised machine learning) works today is that you create a model (algorithm) and train the model with the historical data (labeled) to predict accurate outcomes.  For example, you can get data on historical house purchase prices for a zip code and develop a model (algorithm) that has house size, number of bedrooms, year of construction, number of bathrooms, etc. as inputs and house purchase price as output. After the model is ready, you back test it to see if it is actually giving you the accurate house purchase prices. If required, you make adjustments to the model. Furthermore, you continuously refine the model as new data becomes available. When a new house comes for sale in that zip code, the model would be able to predict the purchase price of that house.

Possibly, the players in the Cloud layer can move to the Edge of the network for certain AI applications (Figure 5). After the model is trained in the cloud, it can be deployed on the Edge of the network for applications where speed is critical.  There is a debate in the industry if the compute on the device (like an iPhone or an autonomous car) is Edge computing. In my view, on-device compute or storage is not Edge computing.

Things might evolve in a way that requires distributed deep neural networks (Figure 6). A challenge with Machine Learning and Deep Learning is that we know that they work but we don’t know exactly how. We don’t know how exactly our brains work either but we are good at using them. Let’s see what happens with AI. 

Figure 6: AI subsets from Andrew Ng

Cloud (Storage and Compute) Layer 

Many countries (Russia, India, China, etc.) are changing their laws and require that the data generated in their countries remain in their countries. This is not so bad for the big American Cloud players if the countries have scale. 

Microservices and Containerization will continue. Hybrid-Cloud and Multi-Cloud implementations might increase.

This layer might become an AI layer which includes Store and Compute (Figure 5). To do that fast, new semiconductors might be required. The challenge with AI is that it is domain-specific and narrow in scope. Separate models have to be created, tested, deployed, and refined for different predictions, and the data has to be made ready for the model which has a huge cost and it takes time. 

Advances in AI (Unsupervised learning), that does not need a lot of data, might move the entire AI function to the Applications itself. Unsupervised learning models learn by a game theory type incentive systems and can outperform supervised learning systems after a short period of time, as shown by Alpha Go Zero. 

Semiconductors Layer 

If China is able to develop its own semiconductors, that will change the dynamics of the industry. Otherwise, the current players in the semiconductor layer will continue to grow by technology improvements and by buying smaller players. There might be new players who develop AI chips and many of them would get acquired.

If you think that tech plays a big role in your life today, wait for a decade.