A couple of weeks ago, at an event hosted by the United States Department of Justice (DOJ) and Stanford Business School, policymakers and industry stakeholders convened to discuss the competitive landscape of artificial intelligence (AI) industries. The discussion ranged from the promise of large language models (LLMs) and foundation models (FMs) to the regulatory mechanisms for this emerging technology. It was clear that regulatory and enforcement actions were being considered even as the regulators were trying to fully comprehend the technology they were tasked with overseeing. It is also clear that the assumptions they’re making are flawed and that skepticism over cloud service companies entering the LLM market are both unfounded and at odds with antitrust law’s mission to protect competition and consumers.
LLM AI technology holds tremendous promise. It is poised to revolutionize various sectors, such as healthcare and manufacturing. For the LLM industry to thrive and fulfill its potential, it needs reasonable regulatory frameworks, substantial capital access, and supportive policies. Bringing these advanced AI systems to market requires investments that are often beyond the reach of smaller companies innovating in this space. They demand extensive research and development (R&D) efforts, including creating new architectures and optimizing algorithms. All of this requires significant computational power, often involving thousands of graphical processing units (GPUs) or tensor processing units (TPUs) over prolonged periods. To attract top AI researchers and engineers, companies must offer competitive salaries and benefits. Additionally, maintaining the infrastructure for LLM deployment, such as data centers and cloud services, requires ongoing substantial investments. FMs, a class of general-purpose LLMs that provide a base layer for LLMs with more specific purposes, are especially costly to create and maintain. Small companies have and will invest in creating LLMs, but asking them to take on the more foundational layers of the LLM and FM stacks—the ones that necessarily require huge investments—is not fair to the small businesses that need those services to be robust. Thus, the government should not prevent or discourage larger companies with existing physical assets or know-how from moving vertically into FM and LLM services.
Consider a parallel example. As the latest model cars take on more autonomous features, the industry is undergoing major shifts. New entrants are taking advantage of these disruptions to make credible inroads, while incumbent auto manufacturers are poised to leverage their existing vertical supply chains and manufacturing bases to make autonomous cars. Imagine if policymakers decided incumbent auto manufacturers should either be barred from making autonomous vehicles or regulated differently from market entrants just because they already have a vertically integrated advantage and compete successfully in the market for non-autonomous cars. Such a decision would deprive consumers of options from some of the strongest competitors.
Similarly, companies like Amazon, Microsoft, and Google have a sprawling network of existing assets in upstream industries such as cloud computing, which will already serve as part of the vertical stack for LLMs and FMs. Just like building a car requires auto manufacturing capacity and a supply chain, LLMs and FMs require lots of computing power, and this is one of the main inputs for LLM and FM services. In turn, cloud services are the main providers of this capacity. So, government intervention to prevent cloud companies from providing cloud capacity for their own FM efforts—or even to prevent them from providing compute capacity to other FM or LLM projects—just because they are large incumbents in cloud computing appears to defeat the purpose of antitrust law.
In this context, misguided enforcement actions driven by unwarranted bias against established companies amounts to essentially using governance power to stifle the nascent LLM industry. Real-world competition is not about numerous companies providing nearly identical services with minimal margins but about market-driven investments fostering innovation. Restricting large companies from investing in LLMs could raise costs and lower service quality for small businesses. Allowing timely, market-driven investments instead will drive innovation and ensure affordable, high-quality services.
Members of ACT | The App Association, as demanding consumers of LLM and FM services, oppose government interference that tends to lower service quality, reduce real choices, and raise input costs. Market-driven investments from large companies are crucial to meet consumer demands. Government restrictions could stifle this emerging industry by limiting capital flow.
Regardless of one’s stance on the issue, it is crucial to recognize that regulations and enforcement actions should only be implemented with a clear understanding of their downstream impacts. Innovation works best when driven by small, agile companies in the app economy that leverage reliable, capital-intensive foundations that established players are in the best position to provide. Proper regulatory frameworks can encourage such investment while promoting healthy competition. Just as investments in the telecom industry expanded internet access, LLM investments can democratize access to advanced AI tools.
The markets for LLM and FM services are at a crucial crossroads, with significant investments poised to drive innovation and economic growth. Policymakers must recognize the importance of these investments and avoid stifling innovation with overly restrictive regulations and misguided enforcement actions. This will help unlock the full potential of FMs and LLMs, benefiting society and small business-driven problem solving.