Competition and innovation are inextricably linked, driving each other forward.

And without a doubt, one of the most significant innovations in recent memory is artificial intelligence — its development is ratcheting at an intense rate, with innovators racing to build the latest and greatest cutting-edge tools.

But because it is such a fledgling technology, it could very well be exploited by larger entities — or at the same time niche, specialized ones — whose technology got to the market first. For this reason, the Federal Trade Commission (FTC) has warned that it is keeping a close eye on AI when it comes to fair, competitive practices.

“Any scientific advancement benefits from having lots of people working on it,” Todd A. Jacobs, chief information technology officer at CodeGnome Consulting. “Without competition, technologies often stagnate for years, even decades, when they could otherwise advance even faster.”

Generative AI is set to become one of the world’s most dominant industries. One projection puts the market at $76.8 billion by 2030, up from a current valuation at $11.3 billion (registering a CAGR of 31.5%). Goldman Sachs, for its part, boldly says the technology could drive a 7% (or nearly $7 trillion) increase in global GDP.

Amidst all this, the FTC says issues could arise around control over one or more of the “key building blocks” of generative AI: data, talent and computational resources.

If a single company or handful of firms controlled one of these essential inputs, “they may be able to leverage their control to dampen or distort competition,” the agency asserts. “And if generative AI itself becomes an increasingly critical tool, then those who control its essential inputs could wield outsized influence over a significant swath of economic activity.”

In particular, the agency said firms could bundle and tie products — offering multiple products in a single package or conditioning the sale of one product on the purchase of another, respectively. There also may be issues with M&A, exclusive dealing or discriminatory behavior against new entrants in favor of partners.

These are not new issues, of course, or singular to genAI.

“Competition is generally considered to be beneficial to consumers for fair pricing and continued innovation, and for the workforce for fair wages and mobility,” said Adam Berry, a senior managing director on FTI Consulting’s data and analytics team. “This is not unique to emerging technologies, but is perhaps more complicated to attain with emerging technologies or products due to the R&D involved, or compute resources.”

Data — and exceptional amounts of it — is the foundation of any generative AI model.

The FTC cites this as a competition concern, pointing out that the volume and quality of data required to pretrain models may impact the ability for new players to enter the market.

“Fundamentally, it's not the technology that is really a competitive concern, it's the data,” said Jacobs, who is a contributor at the Theia Institute, a DC-based cybersecurity and AI ethics think tank.

He pointed out that companies including OpenAI have gathered massive amounts of publicly available data, and are holding it behind a paywall. This type of “financial walled garden” — which is offered in pay-as-you-go fashion — can make it difficult for other companies to get a foothold.

While there are open-source alternatives — including Colossal AI, which can build equivalent datasets to ChatGPT backend data at about a tenth of the computing storage — there are still issues with cost, compute, storage and talent.

The FTC agrees that open-source may widen and democratize the market, but also cautions against “open first, closed later” tactics. That is, when firms initially use open-source to draw in business, then later close off to lock-in customers and lock-out competition.

“Several elements in the lifecycle of a genAI solution might be exposed to some competition imbalance,” said Claudio Calvino, a senior managing director on the EMEA data and analytics team at FTI Consulting. “Data is one dimension. Companies that have access to large amounts of this commodity, maybe because of existing dominant positions, might, or will, have a significant advantage.”

As it stands now, companies heavily involved in generative AI are treating mathematical models as proprietary or trade secrets — even if they are based on publicly-available data, Jacobs pointed out. One pragmatic solution, he said, would be for companies to share datasets and raw data, but not the “secret sauce” of how they turn that into useful tools.

“The vast collection of information should not be permitted to be monopolized and monetized the way that it is,” said Jacobs.

Building, training and deploying AI requires massive, expensive computing power, either in the way of graphical processing units (GPUs) or cloud computing. Fine-tuning, on the other hand, requires far less compute than pretraining — but it still needs to be performed on existing pre-trained models, which means partnering or using open-source models.

This could lead to a market where the highest quality pretrained models are controlled by a small number of companies. Adding to this, markets for specialized chips — such as those offered by the trillion dollar-valued Nvidia — are highly concentrated and demand is quickly outpacing supply.

In fact, an FTC challenge to Nvidia led to the company to abandon a $40 billion purchase of Arm. The agency alleged that the merger would have stifled competition in multiple processor markets, including chips for cloud providers.

“GenAI models are extraordinarily computationally intensive and a very small number of chip manufacturers and the major cloud providers (Microsoft Azure, AWS and Google) control the majority of the GPUs needed for genAI,” said Kjell Carlsson, head of data science strategy and evangelism at Domino Data Lab.

Smaller enterprises can defend themselves against this market power, he said, by implementing hybrid cloud capabilities that enable their data scientists and developers to leverage any infrastructure, including central processing units (CPUs). This can help them negotiate the best rates and ensure reliable infrastructure access.

To avoid vendor lock-in and help future-proof in the rapidly innovating space, enterprises should “implement open platforms that can leverage any AI models regardless of their source, yet provide the governance and security capabilities to ensure that they are used reliably and responsibly,” Carlsson said.

Particular engineering and research skills are required for AI, and it could be difficult for enterprises to find, hire, retain and upskill workers. The FTC contends that with scarce requisite engineering, “powerful companies may be incentivized to lock-in workers and thereby stifle competition from actual or would-be rivals.”

A competitive and innovative marketplace demands that talented workers “move freely and, crucially, not be hindered by non-competes,” says the FTC, which also released in January the Non-Compete Clause Rule.

The agency cautions that going forward, the Bureau of Competition and the Office of Technology will use their “full range of tools” to identify and address unfair competitive practices.

Jacobs advised businesses to be “very careful” with the legal aspects of genAI.

Building something more complex on infrastructure owned by others essentially puts that technology into their hands, he said. Instead, organizations should consider building their own datasets or focus on areas that aren’t already being exploited by large companies.

“Make sure whatever you're doing is as portable as you can make it,” said Jacobs.

When it comes to agreements, he emphasized, know what you’re getting into and make sure you truly understand all the parameters and limitations.

“Some of it comes down to common sense business stuff,” said Jacobs, underscoring that “you want to make sure you’re not giving away the company store to someone else and then paying them for the privilege of using it.”