The Oligopoly Solidifies

For nearly two years, the narrative around artificial intelligence has centered on a simple question: who will dominate? The answer, it turns out, is more complicated and more troubling than the binary competition between OpenAI and Google that dominated 2024's headlines.

The market for general-purpose large language models has crystallized into an effective oligopoly. Three companies—Anthropic, OpenAI, and Google—control nearly 90 percent of the $37 billion enterprise market for AI models. But the composition of this triumvirate has shifted dramatically. Anthropic now commands 40 percent market share, followed by OpenAI at 27 percent and Google at 21 percent. This represents a seismic realignment from just two years ago, when OpenAI stood unchallenged as the market leader. The narrative of inevitable dominance has given way to something far more fragile: a precarious three-way balance that could tip at any moment.

The depth of capital commitment underscores just how serious these companies are about maintaining their positions. Microsoft, Meta, Alphabet, and Amazon alone plan to spend $610 billion on capital expenditures in 2026—nearly triple what they invested just two years prior. This isn't merely aggressive competition; it's a form of economic warfare waged through data center construction and computational horsepower. Only a handful of corporations on Earth possess the financial resources to participate in this arms race. Apple, despite its unparalleled resources and brand power, made a strategic choice to partner with Google rather than spend the hundreds of billions required to develop a competitive foundation model from scratch.

The Application Layer Invasion

But the oligopoly's most troubling feature isn't its concentration—it's the way these model providers are leveraging their control to colonize markets downstream. Anthropic has begun competing directly in application spaces where its own customers operate. The company has ventured into financial and legal services, directly invading the territory of firms like Thomson Reuters that built their business around accessing general-purpose AI models from providers like Anthropic itself.

This dynamic creates a fundamental conflict of interest that regulators have barely begun to address. When the company controlling the infrastructure also competes in the application layer, it possesses an asymmetric advantage that no competitor can match. An AI application company dependent on Anthropic's models faces an uncomfortable reality: the company powering its technology is also its potential rival. And if Anthropic decided that an application company's success threatened its own ambitions, it could theoretically degrade or eliminate that company's model access entirely.

The threat isn't merely theoretical. Developers across the industry—large enterprises and scrappy startups alike—now operate under a shadow of existential uncertainty. Any software company building applications on top of OpenAI, Google, or Anthropic models must confront an uncomfortable question: what happens if the model provider decides to enter my market?

The Consumer Battle Lines

While enterprise customers navigate these treacherous waters, the consumer market tells a different story—one where incumbency still matters enormously. ChatGPT dominates consumer AI with approximately 400 million weekly active users, a commanding position that reflects both first-mover advantage and persistent brand recognition. Yet even this fortress shows signs of vulnerability. Meta has claimed that 500 million people have tried its AI features, suggesting that while Meta lags behind OpenAI in mind-share, it possesses the distribution networks and user bases to pose a genuine long-term threat.

OpenAI has simultaneously begun a slow but deliberate push into Google's search territory, challenging what was once considered an unassailable moat. The search market, dominated by Google for nearly three decades, suddenly looks permeable to attack from an AI company with billions in funding and a product that fundamentally reimagines how users access information.

The three-way battle shaping consumer AI—ChatGPT's dominance, Meta's aggressive expansion, and OpenAI's lateral move into search—suggests that the consumer market will develop differently from the enterprise space. Where enterprise customers face an oligopoly with limited alternatives, consumers will likely enjoy genuine choice. But this bifurcation creates its own problems: while large technology platforms compete ferociously for consumer attention, they simultaneously collaborate on infrastructure investments and maintain comfortable market share in enterprise applications.

The Forgotten Players

Scattered across the AI landscape are companies trying to survive in spaces carved out by the titans. Meta and xAI maintain their own models. Open-source projects from China and France exist as alternatives. Yet none of these constitute meaningful competition at the frontier. When Meta—the company that once challenged Google's dominance and invented mobile advertising—must resort to building its own AI infrastructure rather than competing through applications and services, the market's structural constraints become apparent.

Salesforce, recognizing that AI competence would determine its survival, spent $8 billion to acquire Informatica. Other enterprise software providers have made similar bets, spending heavily to integrate AI capabilities into customer management, financial analysis, and operational tools. Yet these companies remain fundamentally dependent on access to frontier models developed by Anthropic, OpenAI, or Google. They have become sharecroppers on land owned by others.

X and its AI model Grok represent perhaps the most uncertain position. The model has achieved technical parity with established competitors, but its owner's unpredictable leadership and the platform's ongoing uncertainty about its business model create an aura of instability that talented researchers and enterprise customers avoid.

The Regulatory Void

Remarkably, this landscape is emerging almost entirely without coherent regulatory frameworks. The concentration of market power in three companies, the vertical integration of model providers into applications, the ability to cut off access to critical infrastructure—none of this has triggered the antitrust actions that would have been routine in previous technology eras.

The geopolitical dimension adds another layer of complexity. While American and European companies dominate the frontier, open-source models from China and France represent alternatives that don't require dependency on Silicon Valley. This creates a peculiar dynamic: as American regulators remain passive, Chinese projects like open-source initiatives could eventually provide genuine alternatives to the American oligopoly. The irony is that American companies' ability to lock out competition might be undermined not by antitrust action but by geopolitical competition and the natural economics of open-source development.

The Data Question

Underlying the entire competitive structure is a question that few in the industry acknowledge directly: who will foot the bill for the data centers and computational infrastructure required to train the next generation of models? The answer shapes everything. If training frontier models requires hundreds of billions of dollars in capital investment, only companies with deep financial resources and willingness to accept years of losses can participate. This naturally creates oligopoly. Anthropic, despite its recent market gains, survives as an independent company only because of massive venture capital backing and relationships with major cloud providers.

Data privacy, meanwhile, remains almost entirely unaddressed. These models train on vast corpuses of human-generated content, much of it created without informed consent. The regulatory frameworks governing this data use remain embryonic. Governments are beginning to grapple with these questions, but the pace of regulation will almost certainly lag the pace of technological and commercial change.

The Inflection Point

We stand at an inflection point where the outcome remains genuinely uncertain. The three-company oligopoly in enterprise models could consolidate further through acquisition or collapse through disruption. The consumer market could fragment into multiple competitors or concentrate around ChatGPT's dominance. The application layer could be progressively colonized by model providers, reducing software companies to mere feature builders on top of someone else's infrastructure.

What seems virtually certain is that the current trajectory is unsustainable without intervention. The capital expenditures required to maintain frontier model competency grow larger each year. The concentration of market power continues to deepen. And the regulatory frameworks necessary to address concerns about access, competition, and data privacy remain underdeveloped.

The winners in this competition will be determined not merely by technical superiority or capital advantage, but by strategic choices about how aggressively to expand beyond core competencies. Anthropic's invasion of applications suggests that the most successful AI companies will be those comfortable competing across the entire value chain. Yet this same strategy accelerates the very dynamics that concentrate power and raise existential questions about whether meaningful competition can survive in an AI-dominated economy.

For developers, enterprises, and consumers, the question isn't simply which company will win the AI race. It's whether there will be any meaningful choice left in the marketplace when someone inevitably does.