This is the second article in a series. The first article can be found here: “Although AI is a Bubble, It’s Still Investible.“
Generative artificial intelligence (Generative AI) has been in the works for several years, but the general public and the stock markets didn’t pay much attention until ChatGPT came along, demonstrating the immense opportunities of the new technology. Within just five days, the generative AI language model from OpenAI, which produces original material in response to user requests, has attracted one million users. Now, it seems that we cannot get enough of AI — it’s all over the headlines. It takes first place in Google searches, and it’s garnering investor attention and, well, money.
Let’s Get Generative
Long before the Generative AI advance, many economic sectors had been exploiting Predictive AI – models that analyze existing massive amounts of data to detect patterns, provide the basis for making decisions, produce analytics, classify data, and detect discrepancies, problems, or fraud.
Predictive AI has been providing massive amounts of economic value by cutting costs, optimizing processes, and, in general, making use of Big Data possible as well as effective. Predictive, or “traditional”, AI has been extensively used in autonomous vehicles, medical diagnosis, stock trading, video games, and automating numerous routine knowledge-based tasks.
Now, we are at the dawn of Generative AI. As its name suggests, it can generate new and original content based on its learning of large datasets, just like humans do — but at a vast scale. Generative AI has captured the general public’s attention by its ability to create art, compose music, and write texts (and even pass Wharton School’s MBA exams); its future applications may exceed entertainment and education uses.
Futurology aside, one of the most important tasks that Generative AI has performed so far – through its missionaries ChatGPT, DALL-E, and Midjourney – was to draw attention, and, consequently, money, to the AI technology and its vast capabilities.
The cutting-edge AI technology can be used in virtually everything, from providing full-scale customer support to from-scratch software generation to the creation of 3D worlds models for simulations for car development to developing new protein sequences to aid in drug discovery to helping with accurate weather forecasting and natural disaster prediction. With these capabilities, the economic added value of AI is potentially unlimited.
According to McKinsey’s latest research, AI can add up to $4.4 trillion (more than the UK’s GDP) annually to the global economy. The technology can automate up to 70% of work activities, with the largest contribution in the near term seen in banking, tech, and life sciences. In the near future, we will see the emergence of new business models and applications and even brand-new industries. The AI era is in its infancy, but the potential of AI cannot be overestimated.
The Hard Base of AI
Artificial Intelligence is a very complex technology, requiring significant research and development as well as vast investments. Besides the cutting-edge software models, it cannot develop without appropriate hardware, infrastructure, and data storage. Because of that, it’s not only the companies developing Generative AI models that are benefiting from the vast influx of investments.
While companies like ChatGPT-developer OpenAI draw most of the public’s attention, the whole AI ecosystem is developing around the technology, with many more companies expected to find their place in the AI value chain.
The footing for all things AI is hardware – the chips that form the computational basis of the technology. AI-computing workload has been doubling every three to four months up to now and will undoubtedly accelerate further; AI-spurred demand for essential GPUs (graphic processing units) will bring immense benefits to chipmakers. That is why Nvidia (NVDA) has seen its stock surge by over 200% year-to-date.
Although Nvidia’s hardware underpins the lion’s share of AI applications, it is not the only producer of chips that are designed to handle the intensive computations required for training AI models. Such a lucrative field already draws competition from existing leaders, and many new ones will come along.
One of the main competitors to NVIDIA in the GPU space is Broadcom (AVGO), a major supplier of chips for Ethernet switches in data centers. In April, the company revealed its new chip for supercomputer connection that will add more capacity to handle AI-driven network load.
Alphabet (GOOGL) has been designing and deploying Tensor Processing Units (TPUs) since 2016 for its own use; in April, the company revealed its new TPU-based AI supercomputer, saying that it’s faster and more efficient than Nvidia’s systems. Meta Platforms (META) has entered the hardware race as well, scooping up the whole AI-chip team from a British startup Graphcore to design and develop supercomputing systems to support AI in its data centers. Meta was slow to adopt expensive AI-supporting hardware, but now it is all in the game, boosting investment in everything AI-related.
Two previously-sleepy giants, Advanced Micro Devices (AMD) and Intel (INTC), are also stepping up the AI chip race. AMD has just revealed its new chip, which it described as “the world’s most advanced accelerator for generative AI,” representing a challenge to NVIDIA’s dominance in the field. Intel, one of the largest players in the market with a long history of technology development, is working together with the AI leader Microsoft (MSFT) to enable AI applications for PCs; Intel is also shifting its chip-developing strategy to compete with NVIDIA and AMD.
AI Back Office
Besides the fastest and the most potent chips, AI needs infrastructure, including servers, data centers, and cloud systems that will house and run the AI. This infrastructure provides the necessary storage and processing capabilities, whether those are based on in-house servers or on cloud-based systems that provide storage, computing power, and various other services.
Major players include Amazon’s (AMZN) Web Services (AWS), Google Cloud, and Microsoft Azure. Infrastructure as a service (IaaS) allows users to access computational resources on demand, which is vital for the large-scale computations needed for AI.
Two other major players in the infrastructure field are IBM (IBM) and Oracle (ORCL). IBM offers a suite of AI tools, pre-built models, and robust infrastructure services through Watson (IBM’s data analytics processor that uses natural language processing) and IBM Cloud. Oracle provides cloud infrastructure along with a suite of AI tools, databases, and machine learning services. Oracle Cloud Infrastructure (OCI) provides the foundation for cloud-based data management powered by AI and machine learning (ML).
It must be said that the above-mentioned giants not only provide the underlying hardware and infrastructure but also offer higher-level services like pre-built machine learning models, AI tools, and development environments. They effectively span multiple stages of the AI value chain.
In addition to the infrastructure providers, numerous companies are serving as infrastructure facilitators. For instance, cloud-networking company Arista Networks (ANET) provides solutions that are designed to improve scalability, performance, and reliability, which are key factors in maintaining the robust infrastructure necessary for AI and machine learning workloads.
Its competitor Cisco Systems (CSCO) has launched a new generation of processors that can support massive GPU (graphic processing unit) clusters for AI workloads. Another networking company chipping in is Juniper Networks (JNPR), with its AI-driven switches and software for AI-enabling networks.
Data is Power
AI requires large amounts of data to train and function effectively. Therefore, this stage involves collecting, storing, and processing this data. Companies gather data through various means, from user interactions on digital platforms to IoT devices. The data is then stored securely and in a way that it can be easily accessed for training AI models.
However, just collecting vast amounts of data is not enough. Raw data is often messy and contains lots of irrelevant information. Therefore, it’s necessary to clean the data, deal with missing values, and standardize it into a format that can be used to train AI models. Data processing might also involve anonymizing data to ensure privacy and comply with data protection regulations.
Many companies facilitate data collection and storage; many of them also provide data cleaning, processing, and management solutions. Most of the companies in this field are the same one-stop shops mentioned earlier: Alphabet, Microsoft, Amazon, Oracle, and IBM. These giants not only provide the infrastructure and services necessary for data collection and storage, but they also use AI to improve their own products and services, making them key players in the AI training landscape.
However, there are several companies whose main business is data. Palantir (PLTR) specializes in big data analytics; Splunk (SPLK) provides software for harnessing the value of big data; Teradata (TDC) specializes in data warehousing and analytic applications; Iron Mountain (IRM) engages in the provision of storage and information management solutions; Equinix (EQIX) provides collocation space and develops data center solutions; Informatica (INFA) supplies AI-powered platform that connects, manages, and unifies data; and NetApp (NTAP) provides cloud data services, data storage systems, and data management solutions.
While the main business of these companies is centered around data collection, storage, and analysis, it’s important to note that they usually serve a broader range of industries and are not solely focused on AI training. Nevertheless, their services form a critical part of the infrastructure needed to train AI models.
Putting AI to Work
After the data is collected, stored, cleaned, dissected, and analyzed, it must be put to work. At this stage, the data is used to train AI models. This involves selecting an appropriate model, defining a loss function, and training the model using machine-learning frameworks. This step also includes testing and validating the model to ensure it performs well.
The main players in this field are IBM, NVIDIA, and the creator of ChatGPT, OpenAI. IBM is a well-known player in the field of AI, with a history that dates to the 1950s; its AI platform, IBM Watson, is used across various industries for tasks like data analysis, automation, and predictive insights.
NVIDIA is primarily known for its graphics processing units (GPUs), but the company also provides a range of software and hardware tools that are widely used in AI model development. Microsoft-backed OpenAI is currently the most influential company in the AI sphere, as its highly-advanced models have a significant impact on the AI field. Another notable influencer is Google’s subsidiary DeepMind.
Once a model is trained, it needs to be integrated into an application that can use the AI’s capabilities. This could be anything from a voice assistant to a self-driving car. It also involves creating user interfaces and APIs so that the AI can interact with users and other systems.
As for AI application development, there are numerous companies in this less capital-intensive playground, ranging from mega caps to tiny startups. The most notable companies in this field are Salesforce (CRM), which develops cloud-based enterprise software for customer relationship management, and Adobe (ADBE), which integrates AI into its digital marketing and media solutions and provides an AI-powered platform to automate tasks and offer predictive analytics in its applications. There are many other companies involved in AI models and application development. As AI technology advances, we’ll likely see even more companies entering these fields.
The final stage in the AI value chain is the provision of AI as a service (AIaaS), a cloud-based service offering artificial intelligence outsourcing, making it available to end-users or businesses. This can involve providing cloud-based AI services or implementing AI directly into products like voice assistants, recommendation systems, or advanced analytics tools.
Most of the top players in the AIaaS sphere are the AI stalwarts and their divisions — Microsoft, Alphabet, IBM, Oracle, Salesforce, and Amazon. However, countless startups are coming down to play – maybe some of them will steal the crown from the giants of AI.
Investing in the Future
In the swiftly advancing landscape of AI, the entire value chain presents a cornucopia of investment opportunities — from firms fueling the engine of AI through hardware and infrastructure development to those that help manage and analyze data, and further down the value chain to companies at the forefront of AI model development and application, as well as to those that are democratizing access to AI and offering scalable solutions for businesses.
However, the stage is far from monopolized. Startups and innovators with breakthrough ideas are finding their niches, potentially offering high-return investment prospects. As we venture deeper into the era of AI, each link in the value chain represents a fertile ground for investment, promising lucrative returns and a stake in shaping the future of technology.