Experts Expect Limited Competition in Generative AI Industry

New research suggests tech giants may gain an edge in generative artificial intelligence (AI), raising questions about the industry’s competitive future.

According to a recent paper, massive computational requirements and network effects naturally lead to market concentration. This consolidation may result in a few key players wielding outsized influence over pricing, data control and AI capabilities. It’s a trend that worries many in the industry.

“We are likely to see decreasing prices for smaller models and continued differentiation across large models,” Alex Mashrabov, CEO of Higgsfield AI, told PYMNTS. He points to OpenAI’s GPT-4 for prosumer use cases and models like Flux and Llama for easy fine-tuning as examples of this differentiation.

Observers say that limited competition in the generative AI industry could lead to higher prices and fewer choices for businesses seeking to integrate AI tools into their operations. This concentration of power might also slow the pace of innovation, potentially hampering the development of new AI applications that could drive productivity gains across various sectors of the economy.

As PYMNTS has previously reported, in recent months, major tech companies have rapidly released iterations of large language models (LLMs) that power chatbots.

Pricing and Innovation

In the new analysis of the generative AI industry, researchers from MIT, Harvard and UC Berkeley warn that the sector may be headed toward oligopolistic control by a few tech giants. The paper argues that while intellectual property rights may not provide lasting advantages, large companies’ dominance over crucial complementary assets could lead to a highly concentrated market.

The authors state their central claim: “… While formal intellectual property and secrecy are unlikely to durably prevent innovative firm entry, incumbent firms’ tight control over key complementary assets will likely usher in a highly concentrated market structure.”

They identify six key assets: computing infrastructure, model deployment capabilities, safety protocols, performance metrics, access to training data, and potential data network effects. These factors, they argue, could allow incumbents to confine new entrants to the application layer of the AI stack, similar to patterns seen in the smartphone industry.

The researchers propose policy measures to foster competition, including government-led benchmarking efforts, expedited legal clarification on critical issues, and initiatives to encourage shared access to AI infrastructure. They emphasize the need to balance competition with maintaining incentives for innovation in this rapidly evolving field.

Philip Alves, founder and CEO of DevSquad, told PYMNTS that competition could impact AI pricing. 

“This could limit access to advanced AI tools, creating a gap between enterprises that can afford premium AI services and smaller businesses that can’t,” Alves said. He draws parallels to the cloud computing market, where major players set terms for the entire industry.

Standardization may accelerate with market concentration, potentially benefiting businesses seeking reliable AI deployment tools. But the trade-off could be a slowdown in breakthrough innovations. Alves noted that in the software-as-a-service world, competition drives creativity more effectively than oligopolies.

Data Privacy Concerns

Data privacy emerges as a critical issue in a concentrated AI market. Mashrabov warned of “a high risk of data privacy issues related to a monopoly similar to one Facebook has, when they acquire a lot of VPN and third party data.”

Alves elaborated on this concern: “A concentrated market puts vast amounts of data into the hands of a few companies … creating an environment where data privacy becomes vulnerable.” He argued that data ownership and transparency can become opaque when customers have fewer choices.

The limited variety of AI providers also raises questions about embedded biases. Mashrabov cautioned that “most of the models today have certain biases (political, etc.), and a limited variety of models leads to products inheriting those biases.”

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.