Fact or Fiction: Job Losses + Unregulated Monopolies

Losing your job as unemployment skyrockets and unregulated monopolies build robots that will end the world.

More often than not, these are the two “armageddon” scenarios people will play out for you if you ask them their deepest concerns with Artificial Intelligence. And although they appear only as jokes at first glance, they reveal themselves to be more valid than you would think.

As students at Harvard Westlake, we are often exposed to leaders in different fields displaying how Artificial Intelligence may shape our world. While most of us don’t genuinely worry about AI companies making robots that could end the world, as students soon to enter the workforce, we have a lot of real questions about AI.

How will AI impact the career I want to pursue? What role is it going to play in my life in the future? Are these actually concerns I should be worried about?

With the rapid changes in technology we see everyday, we must ask ourselves a simple question: are these concerns, shared by many people around the globe, actually supported by the consensus of the current economic community?

Although many economists are still divided over the impacts of AI and how it should be addressed, there are a few key points we should take into account to help us answer this question. Is there a way AI can lead to a market failure through job loss and monopolization, and how can our government help prevent this without hampering progress?

AI-Enabled Automation & US Jobs

As AI enters the workforce, it is clear that many preexisting jobs that previously required human effort will become either easier or fully automated. This is a correlation that many figures in the economic community, whether from Goldman Sachs or McKinsey & Company, agree upon. This consensus, however, does not extend to unemployment, as division remains on how AI will affect jobs in different industries. Despite this, MIT Nobel Laureate Daron Acemoglu, in his 2024 research paper with Professor Pascual Restrepo, sheds insight on this issue by showing a direct correlation between automation and unemployment across industries. By using data since 1980, Acemoglu showed that as firms across industries adopted automation, they based it solely on their own private cost savings. This meant that when they adopted automation, they often ignored the social cost it may have. By displaying this cross-industry adoption of automating preexisting jobs without taking into account the social costs of their actions, Acemoglu’s findings show that rapid automation due to the rise of AI could lead to specific harm to the US job market as well as have other adverse externalities on other parts of the US economy.¹

Acemoglu further expounds on this by also identifying why automation, once discovered, is adopted so rapidly by US businesses: tax favorability. Current US tax codes inherently encourage automation over manual labor, with effective labor taxes sitting between 25.5% and 33.5% while taxes on capital, software, and equipment sit at roughly 5%. This means that for firms throughout the United States, it costs them less in taxes to have machines and equipment producing their product in comparison to paying wages for human labor. While rapid automation at first seems reasonable from the side of the firms, when considering how they don’t take into consideration the social and thus total cost of their actions, we see that this rapid automation due to the inadvertent subsidization of US taxes could have long-term consequences for these firms and the U.S. economy. Even if certain industries' jobs became fully automated by AI, according to Acemoglu, AI would only increase U.S. Total Factor Productivity (TFP) by less than 0.55% over the next 10 years.² Although this only takes into account current US jobs, and not the ones they create, rapid automation due to AI could have a certain effect not only on the employment of US jobs, but also on the U.S. economy as a whole.

AI-Monopolization of Physical Infrastructure

What about AI and monopolies? Is there significant consensus in the economic community that unregulated monopolies will rise with AI, and what their impact will be on the US economy?

Much like the issue of how AI can impact US unemployment, economists are not sure whether, as AI continues, that monopolies will persist. There is, however, significant consensus that mass monopolization already exists in the upstream production of AI.

When thinking of AI, most of us look at the many different apps utilizing AI made by different companies. This, however, is only the downstream distribution of AI. The actual production of AI is a multilayered process requiring many different physical resources and components. In this light, US economists have observed that the upstream production of AI’s components is concentrated in monopolistic companies that asserted themselves early on.

For example, one key component in constructing AI is GPUs, pivotal for the physical infrastructure that processes AI’s complex operations. According to research compiled by economics professor at the University of Virginia Anton Korinek and Jai Vipra, as of 2024, Nvidia controls roughly 90% to 98% of the data center AI GPU market.³

---

Figure 1: Market share of leading vendors in the market for GPUs (2023)

---

In addition, Goldman Sachs has found that out of all the companies building massive data centers needed to power AI’s computing activity, the top five U.S. cloud giants (Amazon, Microsoft, Google, Oracle, and IBM) spent 736 billion dollars in 2025–2026 alone to build AI data centers, on track to control 70% of all global data center capacity by 2030.⁴

In a world where the specialized microchips you need to build AI are produced mostly by one company, and startups who cannot afford to build their own server farms rent data center space from cloud conglomerates, it's hard not to see the monopolization of AI production. Still though, do economists actually see a tangible issue with this?

Prominent economists, despite having different views on the actual scale AI will take in the future, agree that when analyzing their trends there are several issues that emerge. Professor Korinek and Ms. Vipra in their 2024 paper also remarked that each year, the computing power needed to train top-tier AI models grows 4.1 times.⁵ This means that as AI becomes more widespread and adopted, the companies that already have the technical lead in computing models will continuously have a greater lead. This would cement their monopolization and allow these companies to hike prices that could hurt the entire economy at the expense of their profit. Even in a world where AI adoption is not so rapid, monopolization of physical infrastructure presents a different issue that ties into Acemoglu’s assertion on automation. Acemoglu, when talking about monopolization of AI in 2024, argued that when big tech firms control the direction of AI, it will mostly be used to replace workers rather than help them. This compounding of the automation issue of AI would shift national income away from worker wages and into corporate profits. This would mean that the money saved from AI would not go into wages or lower prices benefiting society as a whole, but solely to the shareholders of tech giants.

While economists agree on the concentrated monopolization of AI production today and are concerned about certain issues, they remain split on whether this will persist or that competition will catch up. Regardless, both sides prove that the monopolization of physical AI production remains a serious threat.

Where Korinek and Vipra argue that the continuing gap in technical capacity needed to power AI will lead to companies being unable to catch up to major monopolies, who as result have the technical power to build the best AI models, professors Simon Goldstein and Peter Salib disagree. They argue in their 2026 Minnesota Law Review Paper that although training the first premier AI model costs millions of dollars, rival companies quickly figure out how to copy and create more competitive, cheaper models.⁶ Although this is a valid point, even if these new models are cheap and competitive, they need to run on the physical infrastructure monopolized by major companies. In a world where a monopoly persists over the physical infrastructure of AI, only a few companies can continue building top-of-the-line models and control the rental cost for people to use them.

But if data shows that these problems could eventually become an issue, what steps can our government take to combat these possibilities?

The policy proposals are by no means foolproof, but leading economists who have brought up these issues have come up with solutions to potentially address them.

Tax Code Reform: Solves Social Cost or Causes New Externalities?

To address the idea that rapid automation in the United States due to AI can lead to increased unemployment, we must attack the root of the issue: the US government through its tax code incentivizes automation even when not productive. 

Acemoglu directly addresses this by focusing on how the US Tax Code currently taxes effective labor in the US much more than capital (25.5-33.4% vs 5%), encouraging US companies to automate so they can lower their private costs, though automation long term may not even raise productivity significantly. Taking this data, Acemoglu and other MIT professors propose that the US government adjust the tax code to fix this imbalance, setting labor taxes to 18.2% and capital taxes around 26%.⁷ This would cut excessive automation by 4.1% and increase total US employment by 4.02%. While these numbers may not seem significant, lowering payroll taxes on human workers and raising the tax rate on automated software could pave a way forward compared to our current US tax code. However, there still remains discontent and perceived weaknesses to this approach, with other economists refuting that raising taxes on capital and thus all kinds of software and machinery could accidentally slow normal business growth rather than optimizing them against overautomation.

NAIRR Extension: Lessens Physical Bottleneck or Weakens US Economy?

Even with the widespread debate over whether monopolies in the production of AI will persist as different companies build cheaper copycat models, the physical infrastructure and components of AI remain monopolized. As long as certain companies are able to absolutely control the cost and renting of these resources, competitors and customers become vulnerable to their actions. 

To combat this, leading economists in government such as Asad Ramzanali, the former White House Chief of Staff for the Office of Science and Technology, have proposed extending the National Artificial Intelligence Research Resource (NAIRR) into a full public cloud option. By teaming up across 13 federal agencies, this extension would give universities and small startups free or subsidized access to supercomputers and AI datasets, enabling them to have resources to build top models without selling out to big tech. 

In addition, economists like Korinek and Vipra have proposed enforcing “utility rules” by using regulators such as the Federal Trade Commission and Department of Justice to ban currently dominant cloud companies from hiking up rent prices for outside companies to use their data centers, just as they do with railroad and electric companies. However, these approaches still have many issues, with it being very difficult for subsidized supercomputers under the NAIRR plan to catch up to the caliber of current firms and the possibility of utility rules driving these companies away from still driving data center growth; all the while we see that these solutions merely lessen and do not fully solve the physical bottleneck these companies have in the production of AI.

Conclusion

There is no easy answer to these issues, and much division remains within the economic community over the validity of these points. However, they still remain crucial when thinking about the future of AI, and what it may mean in the next few years. By examining how each of these different economists have analyzed this data and identified these core issues we should consider when examining AI, we can, as Harvard Westlake students, better understand what these questions really mean, and hopefully ascertain a clearer picture of how to approach each of these issues when they appear in our everyday lives.

Notes

1. Mit, Daron, et al. Automation and Rent Dissipation: Implications for Wages, Inequality, and Productivity *. 2024, economics.mit.edu/sites/default/files/2024-05/Automation%20and%20Rent%20Dissipation%20-%20Implications%20for%20Wages%20Inequality%20and%20Productivity.pdf. Accessed 26 Oct. 2026. ↩
2. Acemoglu, Daron, Andrea Manera, Pascual Restrepo, et al. NBER WORKING PAPER SERIES DOES the US TAX CODE FAVOR AUTOMATION? Prepared for the Brookings Institution Spring Conference of 2020. We Are Grateful To. 2020, www.nber.org/system/files/working_papers/w27052/w27052.pdf. Accessed 26 Oct. 2026. ↩
3. Korinek, Anton, and Jai Vipra. Concentrating Intelligence: Scaling and Market Structure in Artificial Intelligence. Oct. 2024, https://doi.org/10.36687/inetwp228. Accessed 26 Oct. 2026. ↩
4. Lee, George, and Lucas Greenbaum. “Tracking Trillions: The Assumptions Shaping the Scale of the AI Build-Out.” Goldmansachs.Com, May 2026, www.goldmansachs.com/insights/articles/tracking-trillions-the-assumptions-shaping-scale-of-the-ai-build-out. Accessed 26 Oct. 2026. ↩
5. Goldstein, Simon, and Peter Salib. “AI Is Not a Natural Monopoly.” Philpapers.Org, 2026, philarchive.org/rec/GOLAIN. Accessed 26 Oct. 2026. ↩
6. Goldstein, Simon, and Peter Salib. “AI Is Not a Natural Monopoly.” Philpapers.Org, 2026, philarchive.org/rec/GOLAIN. Accessed 26 Oct. 2026. ↩
7. Acemoglu, Daron, Andrea Manera, and Pascual Restrepo. Taxes, Automation, and the Future of Labor Taxes, Automation, and the Future of Labor*. mitsloan.mit.edu/shared/ods/documents?PublicationDocumentID=7929. Accessed 26 Oct. 2026. ↩