Will Americans Control AI, or Will AI Control Us?
Controlling the Emerging AI Oligopoly Will Require a Substantially Larger House of Representatives
A hammer is neither benevolent nor malevolent; it simply amplifies the intent of the hand that swings it. It can be used to build a house or assault someone. Artificial intelligence is the same: It is a powerful but agnostic tool. What matters is who wields it—and, increasingly, who regulates the industry that develops, deploys, and profits from it. And that industry is already coalescing into an oligopoly powerful enough to dominate those elected to regulate it.
The AI Industry’s Path to Oligopoly
Today the AI sector appears to be fragmented with thousands of startups. Yet its economic power is already becoming hyper-concentrated, mirroring the classic U.S. industry pattern seen in automobiles and personal computers: Explosive early entry of numerous competitors, followed by rapid consolidation driven by economies of scale, network effects, and capital intensity.
In Q1 2026 alone, global venture capital hit a record ≈$300–330 billion, with AI capturing roughly 80–87% ($242 billion+) of that total. However, a tiny elite comprises the market’s center of gravity: The staggering concentration of the industry is evidenced by just four dominant players—OpenAI ($852 billion), Anthropic ($380 billion)1, xAI ($250 billion), and Waymo ($126 billion)—whose massive market valuations allow them to command a disproportionate share of the world’s computing resources and talent. Massive cloud-computing providers (hyperscalers) and these “frontier labs” dominate the data, models, and distribution channels. This apparent fragmentation is the prelude to oligopoly.
As the AI sector consolidates into a powerful oligopoly, it will increasingly be able to shape not only technologies and markets, but also the very rules that govern it. Recent analyses warn that an unregulated AI oligopoly threatens economic harms (chilled competition and investment), national-security concentration, social harms (misinformation, surveillance, and bias at scale), and political power imbalances.2
Because AI will evolve into such a formidable tool, we can be certain that if the American people do not control the AI industry, it will eventually control us. Ensuring vigilant congressional oversight therefore requires a Congress structurally resistant to industry capture. Traditional tech regulation will inevitably fail if the regulators themselves are captured. What is the solution? As explained in this essay, a substantially larger House of Representatives—restoring the Founders’ design of thousands of members serving small, community-sized districts—is the strongest available institutional safeguard against this eventuality.
The Distinctive Risks of an AI Oligopoly
An AI oligopoly is uniquely dangerous because its technology amplifies power at unprecedented speed and scale. Without countervailing democratic pressure, the industry’s policy preferences—light-touch regulation, expansive data access, liability shields, and continued data center subsidies—become self-reinforcing. Lobbying data illustrates the trend: In Q1 2026, just 11 major tech and AI companies spent $20 million on federal lobbying—an average of $226,000 per day. Six of the largest among them (Alphabet, Meta, Microsoft, Nvidia, Anthropic, and OpenAI) hired 307 lobbyists; that is roughly one for every two members of Congress (across both chambers).3
The consequences of an AI oligopoly are easy to foresee: Innovation is chilled as smaller players are squeezed out, concentrated infrastructure creates national-security vulnerabilities, and algorithmic amplification of bias or misinformation produces social harms. A further, often-overlooked risk is the economics of information asymmetry: Those who control the most powerful AI systems will gain superior insight into macroeconomic and global trends—commodity flows, supply-chain disruptions, financial-market signals, and geopolitical developments—creating unfair speculative advantages in stocks, commodities, currencies, and other markets.4 This is how the powerful can grow even more powerful.
In the longer term, oligopolistic control also raises serious affordability and accessibility risks. Even without overt price hikes, incremental “improvements” can quietly inflate costs. For example, recent testing of Anthropic’s Claude Opus 4.7 showed meaningfully better performance but dramatically raised the effective cost per task even though the listed per-token rate remained unchanged.5 Over time, such cost creep could push ordinary users and businesses toward cheaper alternatives, such as powerful, low-cost models developed under Chinese Communist Party oversight. Those foreign systems often embed content moderation aligned with their political priorities and pose documented risks of data exfiltration and influence operations. Americans turning to “free or cheap” adversary AI would erode U.S. technological sovereignty while ceding a powerful tool for shaping public discourse to a strategic rival.
An additional source of power for the AI oligopolists is their outsized influence over the dissemination of information. Corporate owners could subtly (or overtly) steer AI outputs—news summaries, policy analyses, debate prep, or voter research—to favor one congressional candidate over another, generating hyper-personalized political content at massive scale. The 2020 Twitter suppression of the New York Post's Hunter Biden story demonstrated how a single platform could suppress or amplify politically sensitive stories at scale.6 An AI oligopoly would possess far greater power through personalized, conversational influence that could be very difficult to detect or attribute. In a 435-member House, as currently exists, modest influence over even a dozen races could dramatically shift the balance of power; however, effecting a comparable outcome becomes nearly impossible in a much larger chamber.
Regulatory Capture Is Easier with a 435-Member House
A small legislative body is a concentrated interest’s dream. With only 435 Representatives, a few AI giants can monitor, persuade, and reward targets efficiently through lobbying, campaign finance, revolving-door relationships, and by shaping public opinion at scale. Economic theory (Mancur Olson’s logic of collective action) predicts exactly this: The per-member payoff for influence is high when the number of decision-makers is low. Oversized congressional districts (≈780,000 constituents each) further increase Representatives’ dependence on big donors and consultants, making them “ripe targets” for special interests.
This reality directly contradicts the Founders’ design. Madison in Federalist 10 warned that factions—including economic ones—thrive in small assemblies. Hamilton and Madison explicitly expected the House to grow with the total population (aiming for roughly one Representative per 30,000 citizens) to diffuse power and guard against “the intrigues of the ambitious, or the bribes of the rich.” The fact that the size of the House has been frozen at 435 Representatives since 1913 directly sabotages that constitutional logic.
A Substantially Larger House is the Best Available Structural Reform
Thirty-Thousand.org advocates for enlarging the House to several thousand members by establishing a minimum size of one Representative for every 50,000 residents. That exact ratio was affirmatively passed by two-thirds of both chambers of the first Congress in 1789, only to be derailed through an inexplicable single-word substitution in the amendment's final enrolled text.
As explained more fully on Thirty-Thousand.org, a vastly expanded House fundamentally alters the mechanics of political influence by raising the coordination costs and greatly diluting the per-Representative payoff for capture. Wholesale influence becomes far more expensive and logistically difficult, if not impossible. By reducing congressional district sizes to approximately 50,000 people, the cost of winning an election drops dramatically because, in such a district, a candidate can literally knock on the door of nearly every likely voter. That will shift the focus from “wholesale” media buys funded by corporate PACs to “retail” politics rooted in direct constituent engagement. Capture becomes structurally difficult not merely because there are many more Representatives to “buy”, but because the Representatives were never financially dependent on concentrated interests to begin with. That is, the Special Interests cannot capture someone who never needed their money to win in small congressional districts. Furthermore, the sheer logistical complexity and financial cost of attempting to capture a majority of 3,000 or more Representatives—each with highly localized interests—makes the coordinated industry lobbying (that currently paralyzes a 435-member House) structurally and economically unfeasible, especially if most of those Representatives are working from their home districts.
Skeptics may argue that a massive legislative body would become unwieldy, forcing power to centralize even further into the hands of a few party bosses. But this ignores the reality that power in our current 435-member House is already hyper-centralized. A substantially expanded House would simply function as the Founders intended: As a People’s House that is not dominated by a political duopoly. Because the vast majority of legislative drafting occurs in their committees, a larger House, which would be even less unwieldy, would serve primarily to vote up or down on the committee’s proposals.
Modern IT infrastructure, secure remote voting, and distributed deliberation would allow Representatives to remain anchored in their home districts, avoiding the need for a larger Capitol Building, and thereby breaking the “D.C. bubble”. Instead of being wrangled in backrooms by party whips and lobbyists, thousands of localized Representatives would act as an independent, decentralized jury on AI and other legislation. Modern technology makes this scale practical, as there are solutions to the logistical challenges that the Founders could not possibly have imagined. Moreover, Representatives anchored in small districts will also personally witness AI’s effects on their communities — job displacement, algorithmic bias, or surveillance — making their oversight instinctively grounded rather than abstractly theoretical.
It is worth noting that the Senate’s own small size and six-year terms create additional vulnerabilities to Special Interest capture of Congress, especially since they are elected from massive statewide districts (because of the 17th Amendment) rather than being appointed by the states’ legislatures as per the Constitution. This places even more emphasis on the House — explicitly designed as the people’s immediate check on concentrated power — to become an indispensable firewall. A hyper-representative, un-captured House can block corrupted legislation, generate public transparency, and force even the upper chamber to act.
No institutional fix provides a 100% solution. A larger body might initially slow complex legislation. Yet these are manageable risks compared to the status quo of a 435-member House whose massive congressional districts make it extremely vulnerable to capture by concentrated interests. Antitrust enforcement, open-source diffusion, and public-benefit corporate structures are valuable complements—but they operate downstream of legislation. (Though open-source AI exerts a healthy competitive pressure and keeps innovation decentralized, it cannot substitute for democratic oversight of the frontier labs that set the pace for the entire AI ecosystem.) A Congress that is structurally immune to capture is the foundational safeguard that makes those other tools more effective. Moreover, a larger House makes it likely that many members will bring deep technology and AI expertise to the committees’ deliberations. For the specific threat of AI-oligopoly regulatory capture, House enlargement is the most direct, constitutionally grounded, and historically rooted lever available.
A Larger House Also Benefits the AI Industry Itself
Properly structured oversight does not mean stifling innovation—it means channeling it toward high-trust outcomes. When regulation reflects broad public interests rather than narrow corporate ones, AI deployment fosters genuine collaboration between people and technology. Trust accelerates adoption: Individuals and enterprises integrate AI more deeply when they believe systems are agnostic, safe, transparent, and accountable. This virtuous cycle—better products, broader use, richer feedback—drives faster progress than extractive, low-trust alternatives.
We have historical precedents for how oversight can stabilize concentrated markets, which today serve as cautionary tales as much as success stories. While the early twentieth-century FDA successfully built the public trust necessary for the food and pharmaceutical industries to flourish, recent history highlights the devastating cost of its gradual regulatory capture. From the institutionalized promotion of unhealthy, ultra-processed food options to the approval of questionable medical treatments, the modern agency often reflects the priorities of corporate giants rather than the public interest. These failures illustrate that even the most essential agencies cannot fulfill their mission if the Congress overseeing them is too small and too easily influenced to provide vigilant oversight.
Moreover, if Congress is captured by the AI industry, the resulting lack of oversight risks a hostile public disposition towards the AI: Backlashes, boycotts, or erratic punitive rules born of distrust. History shows industries that help shape balanced governance thrive; those that fight accountability often face sharper corrections. A substantially larger House, by making capture extremely unlikely, tilts the AI industry toward the high-trust path that sustains long-term investment, legitimacy, and societal partnership. In short, vigilant democratic oversight is pro-AI.
Conclusion
The AI hammer will be swung by whoever controls the industry. If we allow regulatory capture in a 435-member House, that control will increasingly rest with a handful of corporations rather than with the American people. Substantially enlarging the House is not a radical idea, as it simply restores the Founders’ vision. And it equips Congress to provide the vigilant oversight the AI era demands: Ensuring that AI serves the many, not the few. The Founders understood that a republic scaled to its population is the best remedy for the “diseases most incident to republican government.” In the age of AI, their wisdom has never been more relevant.
Technical advisor for this essay: Jonathan Strength.
FOOTNOTES:
There are reports of a pending round that could value Anthropic at $900 billion or more.
Narechania, Tejas. “An Antimonopoly Approach to Governing Artificial Intelligence”. Yale Law & Policy Review. Volume 43, Issue 1. Fall 2024.
Sullivan, Bela. Reeling From Major Lawsuit Losses, Big Tech Injects Huge Sums Into Influence Operations. Issue One. April 21, 2026. And, Munis, Jacqueline. Big Tech is spending $226,000 a day on lobbying Congress. Fortune. April 23, 2026.
Narechania, supra note 2.
In head-to-head tests on identical coding tasks, Opus 4.7 consumed approximately 3.6× more tokens (2.9× more output tokens plus 4.8× more cache tokens) than its predecessor, resulting in a 3.6× higher effective cost ($1.38 vs. $0.38) despite unchanged official pricing. See Robert Matsuoka, “Opus 4.6 vs 4.7: The Real Cost of Incremental AI Improvements,” From The Trenches, April 22, 2026.
Internal Twitter documents released after the 2022 acquisition (collectively known as the “Twitter Files”) revealed high-level decisions to block links to the New York Post’s Hunter Biden laptop story in October 2020 and to limit its visibility. See the House Judiciary Committee Select Subcommittee on the Weaponization of the Federal Government, interim report, February 2023, and contemporaneous reporting by various reporters (e.g. Matt Taibbi).