Moving from artificial intelligence (AI) as a shiny new object to AI as a mature driver of economic value requires a corresponding shift from simply embedding AI permanently into existing processes to leveraging it as a catalyst for discovering innovations that lead to subsequent products and solutions. The goal is not just to use AI. The goal is to use AI as a tool to achieve specific outcomes. This forward-looking perspective allows us to extrapolate the future AI economy by considering the types of AI business models that are likely to be more profitable over the long term and adjusting policy priorities to get there. Against a backdrop of geopolitical competition, this is a critical moment to shift attention from simply protecting the basic technology infrastructure to supporting rational subsequent use.
Defining the next stage of the AI economy
Both AI optimists and AI skeptics can agree on one thing. That means the state of the AI economy is not yet mature. There will be ups and downs, but there will be plenty of opportunity for both sides to claim victory. Along the path to maturity, AI technology development (both chip design and model training) will continue to improve, but ultimately it will be the use cases and business models that will determine what the AI economy will look like. When it comes to business models, profit margins matter. A crucial reason why the software-as-a-service (SaaS) market originally led to the explosive growth of Silicon Valley venture capitalist (VC) firms was the lure of high profit margins. SaaS startups have figured out how to design one piece of software and sell it to everyone with a relatively low upfront investment. On top of that, the ubiquitous subscription pricing model and digital advertising ecosystem have steadily increased revenue and lowered marginal costs.
The rise of AI will distort and disrupt the lucrative business models of previous internet booms. Currently, the AI costs charged to users are artificially low, but the rampant expansion of data centers makes the use of AI seem limitless. As the real costs of AI tokens, API calls, and cloud computing resources become visible and passed on to organizations, the way they are calculated will change. Current enterprise AI use cases are primarily attempts to improve existing processes through the use of persistent agent AI. Given the expected increased cost of large-scale language models, in the long run, such an approach may lead to diminishing returns, regardless of revenue. Even though tiered pricing based on model size and model capabilities allows you to better manage token allocation and AI costs, focusing on process improvements is only the first step.
From improving processes to driving innovation
This does not mean that AI-powered process improvement is not valuable in itself. Speeding up existing processes frees up time for employees to focus on more value-added activities, highlighting the link between process improvement and social innovation (though outside of coding tasks, the evidence for AI productivity gains remains unclear). In any case, for the economy as a whole, true transformation will occur when the use of AI brings new innovations, new companies, new products, new industries, and enables Schumpeterian creative destruction. Business models that can define such a transformation must involve thoughtful and targeted use of AI, with long-term benefits in mind, and allow for an innovative complete redesign of workflows that can potentially render many existing processes obsolete. In other words, AI doesn’t just speed up processes; it requires fundamental transformation. Organizations need to fully assess where AI can actually support operations by changing jobs, and where it should not, and change their business models accordingly.
It is important here to distinguish between AI used within IT software, specifically agent AI, and AI applied to downstream use cases. The core activity of software design has a history of gaining speed and ease through progressive layers of abstraction (i.e., coding languages abstract the need to understand machine language, code libraries provide prewritten code for common tasks), and in this sense, AI for software is inherently modern and may be the most influential abstraction layer in its evolution. This makes the productivity and efficiency gains more obvious when writing and testing code. But again, the single-minded urge to spend more tokens is itself a sign that you’re measuring the wrong thing.
For the rest of the economy, the real value of downstream applications of AI is currently less clear. Projects generated autonomously by inferential models, even with better and faster results than entry-level employees, can be more costly than context-savvy humans, both to avoid AI-generated “work delays” and to internalize the insights of the project and its inputs for use in subsequent negotiations. This is further exacerbated by the fact that, unlike other utility-like inputs such as electricity, water, and broadband internet, the outputs from using AI are unpredictable and non-standardized. Although AI is currently driving growth in the VC market, the most exciting startups in the US are those building frontier models, indicating that uncertainty remains about the long-term profitability of using AI in downstream applications.
Therefore, business models that find ways to use AI to drive innovation and do not rely on the most resource-intensive versions of AI for ongoing activity may be more valuable and better suited to driving sustainable economic growth. AI should be seen as a catalyst, not a crutch. Knowing when to use AI will be just as important as knowing when to use it. What a relief from an environmental perspective if the most widespread use of AI were to facilitate carry-on impact rather than requiring continuous energy and token consumption.
This matters little in the short term, as investors and companies are more concerned about missing out than betting on the wrong use cases for AI. And you’ll probably need to overuse AI to some extent at first to start narrowing down the truly profitable opportunities. Therefore, China’s full-stack approach to AI adoption and experimentation could move the country further along in its AI journey, even though adoption remains largely inefficient. But as the AI journey progresses from cutting-edge AI model development, to widespread adoption of AI models focused on process and productivity, and ultimately to leveraging them to enable profit-driven innovation, organizations and countries that have thought deeply about applying AI in a scalable way will be better able to weather investment resets and bubble bursts. This will likely mean that some organizations will fail and new industries will develop. Similarly, countries that are able to first pass these steps toward profitable and effective AI adoption are likely to gain an advantage in the geopolitical AI race by ensuring more dynamic and vibrant economies.
Policy priorities for a mature AI economy
U.S. policymakers can accelerate advances in AI by creating an environment for its long-term, sustainable use.
- This means shifting policy focus from the “technology sector” to industry-specific practical AI applications. The potential uses of AI are as diverse as the risks, making industry-wide regulation and enablement overly general. The way the pharmaceutical industry enables or limits AI in drug production is very different from the way it creates opportunities for self-driving cars on the road, for example. A shift to industry-specific AI will also allow for more effective management of public skepticism about AI, as it will foster practical conversations focused on resolving concerns with concrete impacts and away from vague anxieties about AI in general terms. To drive innovation, it is especially important to maintain a vibrant space for startups within the industry. Sensible competition-focused legislation is key here. California’s proposed Strong Dominant Platforms Stopping Anti-Competitive Self-Preference (BASED) Act takes the lead in recognizing that while some rules are sensible and others are flawed, AI could lead to further platform lock-in and, in turn, limit innovation.
- Recognize that data privacy is more important than ever, as data is essential to improving AI models and designing new AI-enabled business models. While AI provides the tools, it is regulatory certainty around data privacy, as opposed to the current regulatory vacuum, that will enable the necessary experimentation. Regulations around what kind of information can be used to target individuals and groups, and how people can control the extent to which their data is used, will give people the confidence to trust the output of AI. This also pertains to industry-specific approaches, as data is used differently in each industry. For example, using AI to process large amounts of patient and clinical trial data to create personalized drug treatments could be a pharmaceutical innovation that makes it more logical to store health data in ways that are not acceptable as data about shopping habits, as another example.
- Invest in education and training in AI fundamentals. Policies that spread the wealth, so to speak, will further deepen the economic benefits to society, as greater investment in AI training is essential if AI is to become a universally available tool to solve real-world challenges. This will require both changing when and how AI and new coding methods are taught in schools and creating upskilling opportunities for existing employees. Large-scale workflow redesigns to enable innovation often fail due to inertia within the organization, which can only be managed when highly skilled employees are involved in the necessary changes. These structural parts create a long-term scaffolding for managing the maturation of the AI economy.
The AI economy is still in its infancy. Even greater value can be derived from this new technology from business models that focus on innovation, not just process efficiency. Policymakers need to put in place the foundations to foster the long-term use of AI to pursue higher returns through innovative redesign of entire workflows. In the long term, the focus of geopolitical competition should be on scalable AI, not just more AI.
Inuo Geng is an adjunct fellow in the Strategic Technology Program at the Center for Strategic and International Studies and a managing vice president at Gartner, Inc. in Washington, DC. The views expressed are the author’s own.
