Value Monetization in the Age of AI
Part I: Value creation and distribution in the Generative AI Ecosystem
Author: Sam Lee, Abde Tambawala
Introduction
Recent breakthroughs in Generative AI (“GenAI”) have captured people’s imagination and opened our eyes to the potential of this new technology. Many business leaders immediately saw the immense opportunity for GenAI to create new value and its equally tremendous potential to disrupt every part of their business and beyond. Amidst all the euphoria and skepticism in the ongoing debate about GenAI, everyone seems to agree that, regardless of one’s view of this technology, no one can afford to “wait and see” or pass on it. In short — embrace and learn to use AI quickly or risk obsolescence.
As businesses across different industries rush to incorporate the latest AI/ML innovations into their products and workflows, leaders are increasingly asking more questions about GenAI in their companies. One of the most frequently asked questions we have encountered is this: “How is GenAI going to change my business model, and how should I think about monetizing GenAI in my business?”
As we grappled with this question, we have started penning down what we have learned so far — sharing concepts and frameworks that were helpful to us and, hopefully, to you. Please share your learnings and points of view — we look forward to learning from you!
In this first blog, we will explore the AI Value Chain, why every leader needs to understand the ecosystem that powers GenAI today, and what part of the value chain stack we believe will generate the most value in the future.
GenAI and a new(ish) technology ecosystem
Generative AI exploded in the market less than 18 months ago and has already become the fastest-adopted technology in history [1], easily surpassing other transformative technology in the recent past, like the smartphone, the internet, and the personal computer. Although all the hype and commotion make GenAI appear special and unique, the current adoption trend shares many common social, economic, and technological characteristics as other transformative technology adoption cycles we have seen in the past, such as:
- Spearheaded by highly disruptive technology with broad applications across a wide swath of the economy and people, leading to rapid market adoption
- Spawn new business ecosystems supply chains and create new market
- Spur and accelerate the next wave of productivity, innovation, and technology development
Collectively, these characteristics create a phenomenon that is sometimes referred to as a technology supercycle.
A technology supercycle refers to an extended period of rapid technological innovation and adoption that drives significant societal, economic, and cultural transformation. During a supercycle, breakthrough advancements in technology emerge across multiple sectors, leading to widespread disruption, new market opportunities, and shifts in consumer behavior. These innovations often build upon each other, creating a momentum that propels technological progress forward at an accelerated pace.
While still early, GenAI exhibits all the above, transforming markets and society with even more velocity and speed. As a result, business leaders will have even less time to grasp this impact and adapt their business model to the world of AI.
Similar to other technology supercycles in the past, GenAI is rapidly creating a new business ecosystem and supply chain. This new AI Value Chain will eventually encompass every company in every industry — from the chip makers making processors that power AI models to businesses using AI to automate workflows and enhance services. Every business leader should understand this new AI value chain because it underpins the strategic relationships, power dynamics, and business strategy for every participant in a business ecosystem that will eventually include their own.
The AI Value Chain Today
We see the AI value chain consisting of at least seven layers in three distinct but (slightly) overlapping stacks. The first stack refers to the foundational technology and compute infrastructure required to support the creation and operations of GenAI workloads. These markets are more established with a clear product-market-pricing fit. The second stack encompasses core technology providers in AI — basically companies that are (or will be) offering tools and services for customers to build, train, fine-tune, and operate their AI models and AI-enabled applications and services. Data platforms and providers will play a massive role because the need for data to train models and to enable advanced GenAI use cases will create an explosion of demand and new business opportunities for everyone. These are emerging markets where rapid innovations are creating new use cases. Some early leaders are emerging across the ecosystem, but there have yet to be clear winners in the market. Finally, the third stack refers to businesses that will utilize GenAI in their goods and services and companies that will participate commercially in the exchange of data that will power this AI ecosystem. These are existing markets that GenAI will impact, but (with few exceptions) how GenAI will impact these markets is still being defined (i.e., no clear product-market-pricing-fit), and we think there will be a lot of experimentation and opportunities for innovation in the coming years.
The first stack: Foundational Infrastructure
Chips & Hardware underpin the entire ecosystem and refers to companies designing and manufacturing the specialized microprocessors and servers that power AI Models. Reducing Chips and Hardware to a single layer is a gross simplification, as semiconductor design and manufacturing is an entire complex ecosystem in and of itself, with many companies participating and competing in different parts of the market. For this blog, we will focus on fabless chip designers like Nvidia and AMD because we believe the competitive dynamic in this market will evolve quickly and significantly impact the rest of the value chain.
This layer is currently dominated by Nvidia, with over 80% of the GPU market share today [2]. The barrier to entry into this market is very high due to high R&D costs, knowledge barriers, and limited manufacturing capacity available from cutting-edge foundries like TSMC. However, as contract manufacturing capacity improves, this part of the AI Value Chain is poise to become more competitive over the next few years, primarily from three types of new potential market entrants:
- Established chip makers like AMD and Intel will seek to gain market share in this new compute category.
- New startups (ex., Groq) designing specialized chips for specific types of AI processing.
- Hyperscalers seeking to gain efficiency through vertical integration. [3, 4]
Due to its similar economic characteristics, we believe this market will gravitate to 2–3 players in the long run, not unlike what happened for personal computing, servers, mobiles, etc. We also expect the Hyperscalers to pursue vertical integration aggressively for financial reasons and will design, build, and deploy their accelerated computing hardware in their data centers.
Cloud infrastructure providers provide core computing infrastructure (compute, storage, networking) to train, host, and run inference, applications, and services. The existing hyperscalers — Amazon (AWS), Microsoft (Azure), and Google (GCP)- currently dominate this layer. This market features extreme capital costs and economies of scale, making it almost impossible for new entrants to compete. However, new technical requirements such as power, networking, and cooling, as well as emerging regulations concerning security, privacy (e.g., GDPR), and data + AI sovereignty, mean that the market may fracture along both technological and geopolitical fault lines, potentially allowing new entities to gain a market foothold. New market entrants may emerge from three vectors:
- New data center startups (Anyscale, Mangoboost, Lamdalabs, etc.) specializing in accelerated computing infrastructure.
- Companies participating in other parts of the AI value chain (e.g., Nvidia, Meta) seeking efficiency or new market opportunities through vertical integration.
- National cloud providers in support of data sovereignty / sovereign AI initiatives in various countries.
The second stack: Core (AI) technology stack
Foundational Models refer to companies and research organizations that train and serve foundational AI models and are the subject of most of the market hype today. It can be thought of as belonging to both the first and the second stack, as one can argue that it is both the core AI technology and a foundational component for companies participating higher up in the value chain. Participants in this layer require no introductions, with companies like OpenAI, Anthropic, and MistralAI seemingly achieving universal brand recognition overnight. This market is incredibly diverse and energetic, with new entrants, new models, and new research papers dropping almost every week.
Like the Cloud Infrastructure Provider layer beneath it, this market is highly capital-intensive and has very high technical barriers. The cost to train a new frontier class foundational model continues to increase and requires access to hundreds of thousands of GPUs and many exabytes of data. The computation cost alone to train GPT-4 was estimated to exceed $100m. Meta’s massive investment in H100 GPUs worth billions last year further showcases the escalating cost required to innovate at the cutting edge in this market. Furthermore, regulatory scrutiny seems to be increasing, with concerns regarding data privacy, security, safety, and sovereign AI quickly attracting attention from regulators and politicians worldwide.
Given the market dynamics involved, although the market is currently very energetic and highly fragmented, we expect the market to eventually stabilize and consolidate into several vital players (similar to the OS layer). However, the market and the technology are still in their infancy, and we are seeing rapid innovations along many different technical, architectural, and business vectors. (ex. Multimodality, “Agentic” workflows, “Generalist” vs. “Specialist” models, Closed vs. open source, large vs. small models, etc., to name a few). We expect the period of explosive growth will continue at least in the foreseeable future until the market begins to anoint winners. At that point, we expect this market to consolidate rapidly.
From a monetization perspective, one of the most critical business dynamics currently being played out in this layer is the battle between the closed-source models like OpenAI’s GPT or Anthropic’s Claude vs. open-sourced models like Meta’s Llama or Mistral 8x22b. While we don’t believe open-sourced models will entirely displace ChatGPT — just as Linux didn’t kill Windows — how quickly these open-sourced models will improve and be adopted by businesses to create new values will be a significant factor in determining how GenAI will be monetized and how these market will form.
Data + AI Orchestration refers to companies providing the platform and tools enabling organizations to manage the end-to-end data and AI deployment process effectively, from data cleanup and augmentation (e.g., labeling & embedding) to model training to fine-tuning to deployment and running inferences. By leveraging these platforms, organizations can streamline workflows, ensure data quality, and accelerate the development of accurate and reliable inference operations in GenAI applications. Like the first layer (Chips & Hardware), reducing this market to a single layer is a gross simplification as this market overlaps with many adjacent technology segments with rich and complex business ecosystems (ex. Data Analytics, Data Engineering, Security, Observability, DevOps, etc.). This market is still nascent and formative, with some strong product-market-pricing-fit in some cases. However, because both foundational models and implementation techniques are still rapidly evolving, there are potentially many undiscovered killer use cases that have yet to surface in this market.
Although still very early, several categories of workflows have emerged that appear to have staying power and are gaining importance. They are:
- Data Preparation and enhancement refer to workflows and operations related to preparing data for use by ML and LLMs in model training, model fine-tuning, and prompt engineering. They include traditional data engineering workflows like ETL and more AIML-specific operations such as data labeling, annotations, and vector embedding.
- Model management and fine-tuning refer to workflows and operations that fine-tune pre-trained LLMs and systems to test, customize, manage, and deploy these models for inference. Fine-tuning may include data pipelines to fine-tune models with curated datasets, human-in-the-loop feedback techniques like RLHF (Reinforced Learning from Human Feedback), or even more experimental techniques like RAFT (Retrieval Augmented Fine Tuning). Services to help companies discover, test, customize, manage, and deploy LLMs are called “AI Studio” and “Model Garden”. Several well-established services are already in the market, including Huggingface, Azure OpenAI Service, AWS Bedrock, and GCP VertexAI.
- Context Optimization refers to workflows and operations designed to augment input tokens (called “prompts”) to improve inference performance. It includes simple techniques like providing better (or multiple) prompts to the LLM (e.g., Prompt Engineering) to advanced workflows like RAG (Retrieval Augmented Generation). RAG is a preprocessing system designed to “augment” a user’s prompt with additional tokens from a database to improve the LLM inference response. Context Optimization is a rapidly evolving domain with a lot of active research, and we expect many different techniques and implementations in the future with applications in various use cases.
- AI Workflow Orchestration refers to platforms that manage and coordinate the workflows and operations mentioned in 1–3 above, manage the deployment of AI services, and integrate them into applications. Researchers and early adopters have shown that the operational and orchestration overhead required to serve high-quality, context-aware inference is very high. We believe the cost and complexity of deploying and maintaining high-quality models in production will create new opportunities for companies that can help customers solve this emerging problem, similar to how modern software development gave rise to CI/CD and DevOps.
Like the modern DevOps ecosystem, there will likely be dozens, if not hundreds of companies that will participate in this market in the future. This market is also shaping up to be an incredibly competitive space due to its close technical adjacency and dependency on established ecosystems like security, data platforms, and DevOps. We expect many existing companies in those markets today to expand into serving these new emerging use cases, competing alongside many new “GenAI Native” startups that have already gained incredible momentum in an incredibly short time. (ex: Huggingface, Humanloop, Cohere, Together.ai, Pinecone, ScaleAI, LabelBox, Glean, Appen, just to name a few)
Data Providers are companies that curate, generate, and monetize data for use in model training, finetuning, reinforced learning, and advanced context optimization techniques like RAG. They also include companies that provide core services to facilitate data production, cleanup, acquisition, and exchange. The demarcation line between this layer and the last is a bit blurry from a technical or workflow perspective because of the criticality of data in AI/ML — it’s almost impossible to talk about one without the other. However, these are distinct markets from a commercial perspective, as the market for Data differs from that for processing data and orchestrating AI workflows. The data market is already well established but is also evolving and gaining importance as demand for data explodes with AI/ML. “Data for AI” is best viewed as a massive use case expansion of the current data market. This expansion will likely bring many new entrants competing with well-established companies in new AI-centric data categories. Participants in this layer will fall into these four categories:
- Incumbent data providers and brokers — From finance data giants like Bloomberg and Revinitiv to data brokers like Liveramp and Experian.
- New AI-specific data creators/providers — e.g., scale AI and synthetic data companies creating data for AI and ML use cases.
- Companies providing data exchange services — Examples: Data cleanrooms, data repositories, data exchange services, and data marketplace
- “Non-traditional” data providers — AI/ML is created and powered by data, and every company can potentially be a consumer and a data provider for AI/ML use cases.
This layer is ripe for innovation and expansion. The last category mentioned above will potentially be a game changer as the push for AI/ML in enterprises will improve every company’s ability to capture and leverage its data. This growing data and AI sophistication at every enterprise will in turn create even more demand for data. We believe companies will consume third-party data at orders of magnitude higher rates than they do today to power enterprise AI. The data they collect and generate will, in turn, become an increasingly valuable commodity for other businesses, spurring new markets for different companies to share and monetize their data.
The third stack: Applications & Services
Applications — This layer refers to companies building applications that leverage the rest of the AI value chain to provide value to end users/consumers/businesses. These could be existing companies injecting GenAI into their workflows to open new use cases or deliver new value to their customers or new companies looking to disrupt incumbents by “building a better tool” with AI as its anchor. Today, this market segment is characterized by intense competition and considerable hype, with many companies investing substantially in adding AI capabilities into their workflow. While some enterprises are positioned to offer long-term, defensible value through their AI strategy, others may struggle to find a sustainable competitive edge despite incurring substantial costs.
Although we are still at a very early stage of the innovation cycle, many early offerings have emerged within the application space. In the enterprise, these AI-powered services primarily manifest as intelligent chatbots, virtual assistants, or co-pilots, guiding users through their digital experiences and enhancing productivity. In the consumer space, we see similar chatbot/virtual assistant applications for text generation and knowledge retrieval (like ChatGPT) and more experimental content generation tools like DALL-E, Midjourney, and Suno. These early applications span incredibly diverse domains, from customer service and personal productivity to healthcare, media, entertainment, and advertising. It underscores the breadth of possibilities AI affords to revolutionize how individuals and organizations interact with technology. As GenAI technology improves and costs come down, new use cases will continue to emerge that will undoubtedly surprise everyone.
Services refers to professional services companies and systems integrators that help their clients plan, build, and maintain their AI infrastructure or applications. Existing global SIs and advisory firms like McKinsey, BCG, Deloitte, Accenture, and many others have already built substantial capabilities to help clients navigate this new technology and accelerate adoption. The expansion of the market opportunities in this layer will follow the expansion of AI use cases in the layers below. As the technology and unit economics of GenAI improve, it will create new use cases and more opportunities for existing and new businesses to participate in this market.
Parting Thoughts
Generative AI and other Artificial Intelligence technologies are shaping up to be among the most consequential technological innovations in the 21st century. They herald a new technology supercycle that will transform what we can do with software and computers, redefine how we interact and interface with technology, and potentially unleash a new wave of productivity gain for individuals and businesses. This transformation is happening at an incredible rate. It will disrupt every industry and enterprise and impact how companies create and deliver goods and services and their business model. AI will become essential to every company’s business strategy over time. Therefore, business leaders need to understand the AI Value Chain to understand the economic dynamics of this new technology, and every company will participate in the AI Value Chain as a supplier and consumer of this new ecosystem over time.
The AI Value Chain is still new, and the higher layers are emerging as the infrastructure and underlying technology foundation solidifies and becomes more established. While it is still early, we believe several emerging trends will play out over the coming years:
- There is tremendous pressure to integrate vertically in the lower layers due to high capital costs, technology synergy, strong network effects, and economies of scale. The Hyperscalers (AWS, Azure, GCP, and Meta?) will have an outsize presence in this new ecosystem as they have the capital, expertise, and incentives to integrate much of the value chain vertically. Many of these markets will likely be “winner takes most,” where dollars and market share will concentrate on 2 or 3 winners.
- AI will compel every company to upgrade its data infrastructure and capability. There is no AI strategy without a data strategy. Data is essential for every AI use case, from model training to finetuning to context optimizations. Companies at the forefront of data will have a headstart building and leveraging AI Systems, and companies with the expertise to help other companies plan, build, and operationalize their data strategy will have an opportunity to add (and monetize) a lot of value-adds.
- AI will increase the economic value of data and expand the market. As companies learn to incorporate more and better AI into their businesses, they will (by necessity) become better at capturing and using the data generated by their businesses. This explosion of data will be matched by an explosion of demand for data, as every company’s AI Systems will demand even more data for training and context optimization. This supply and demand dynamics will spur companies to buy more data and also allow more companies to monetize their data. Assuming we can reliably overcome security, privacy, and governance hurdles, the data market will have an opportunity to grow by leaps and bounds in the coming years, as every company can potentially be both a data buyer and a data provider in this market.
- It’s (still) about monetization and contribution margin at the end of the day. As AI technology matures and the market shifts its focus to operating and running inference at scale, business models, monetization strategy, and unit economics will become increasingly important considerations when considering AI deployment.
The views and opinions expressed in this post are solely those of the authors and do not necessarily reflect the official policy or position of their employers (its employees, or affiliates).
References:
[1] Suddenly AI — The Fastest Adopted Business Technology in history — Forbes
