Report given by Kei Kebreau to a meeting of the CPUSA National Board on September 3, 2025.

Artificial intelligence (AI) is a computer system programmed to complete some task that up until then required a human to do it. As such, there are several different types of AI.

Artificial general intelligence (AGI) is an AI program that can do anything a human could do. It would be able to complete all mental work, and if given a body, all physical work. It’s a utopian technology that is hypothetical as of today.

Under capitalism, AGI would sharpen labor-capital relations across all sectors of industry starting with white-collar work.

The specific type of AI that is causing all the media and market noise right now is called generative AI. It is built to generate data — text, images, video, and so on.

AI’s ingredients

Today’s generative AI models rely on a few ingredients for their production. This includes basically all the data on the Internet, both fairly obtained and ill-gotten such as works for which the creator should be compensated. This is constant capital for big AI.

A worldwide supply of laborers to help manufacture the data into a form usable by the generative AI model and then to teach the AI model how to properly interact with humans. This is variable capital for big AI.

An enormous number of a type of computer chip called Graphics Processing Units, or GPUs. This too is constant capital for big AI, which the capitalist actually pays for.

Harms to people and planet

There are issues for big AI that correspond with the three sub-points above.

The uncompensated appropriation of intellectual products made by individuals, small capitalists, and big capitalists to build these models makes the offending AI companies enemies of everyone from your neighborhood painter to Disney.

Brutal labor exploitation on an international scale makes these companies enemies of the working class the world over.

GPUs are incredibly expensive to purchase, thanks to the extremely high demand driven by the surge in AI. This is excerbated by a near monopoly on the GPU market for AI. These chips are also extremely expensive to run and will require new, essentially dedicated, energy and water infrastructure to support the current explosion of data center building.

Big money for AI

Total U.S. market capitalization is about $63 trillion as of July 1, 2025. Market capitalization means share price times the total number of shares on the stock market. About $18 trillion — over a quarter of that — belongs to the so-called “Magnificent Seven.”  These include Alphabet (Google’s parent company), Amazon, Apple, Meta (which owns Facebook, Instagram, WhatsApp, and Threads), Microsoft, Nvidia, and Tesla.

NVIDIA is the best investment out of the these seven corporations because their business is selling GPUs which are fundamental to all AI. They are a U.S. company with a market capitalization of over $4 trillion. Their share price is up 25% since January 2025, 50% since last August, and 1000% since the release of ChatGPT.

There is a massive problem; generative AI is so expensive to produce that it isn’t yet profitable to do so. The high velocity of investment creates the conditions of a speculative bubble that could take a significant section of the market with it. This is a win-win situation for big AI, as such conditions in the current political environment would give them a chance to subsume competitors while getting bailed out by the government. Keep in mind that the government is already looking for ways to pour more money into big AI companies in the U.S.

Contradictions

The central contradiction of the AI industry is the immense socialization of its costs versus the privatization of its anticipated profits. The final AI product may seem omnipresent and digital, but it is inseparable from a deeply physical, geographically-concentrated production process.

Socialized costs is an issue because the industry relies on public resources. Data centers demand enormous quantities of public energy and water, placing a strain on local infrastructure. They benefit from local and state tax breaks. Their training data is a product of our collective social labor.

Privatized profit occurs because the goal is to translate the AI infrastructure, currently subsidized by the public and finance capital, into privately held monopoly power. The “winner-take-all” dynamic of the tech industry means that a few firms stand to capture immense value if the technology becomes profitable.

This contradiction becomes most visible at the site of the data center. Take the “Stargate” project, a reported $100 billion data center initiative by Microsoft and OpenAI. Its proposed location in Abilene, Texas, was secured with an 85% property tax abatement — a massive public subsidy for a highly automated facility that will generate few long-term local jobs.

This dynamic is creating a new front in the class struggle. Communities are beginning to recognize that they are being asked to sacrifice public resources for private gain. Recent successful campaigns to block or place moratoriums on data center construction in places like Tucson, Ariz.; Prince William County, Va; and St. Charles, Mo., demonstrate a growing awareness and willingness to resist the encroachment of big AI.

On Aug. 6, community organizers in Tucson, Arizona stopped Amazon from building a data center in their city.

On Aug. 8 community pressure stopped the building of a data center in a rural part of Prince William County, Va. Two years earlier, the county Board of Supervisors approved a rezoning ordinance that allowed the project to move forward. The community struggled on multiple fronts, including taking the county to court — and won. A judge ruled against the construction of the data center near Manassas National Battlefield Park. The coalition behind the legal victory consisted of residents, environmental groups, and historical preservation societies. If completed, this would have been one of the largest data centers built to date.

In St. Charles, Mo., on Aug. 22, the city council voted unanimously to enact a one-year citywide moratorium on all new data center construction. This was a direct result of thousands of residents organizing against the secretive proposal known as “Project Cumulus.” It is considered the first citywide ban of its kind in the nation, despite expiring in a year. This sets a powerful precedent upon which other communities can build. Community activism was able to shut down the data center project, for now.

What might be done

To protect people and planet, it will not be enough to stop a few data centers. The tech oligarchy’s monopoly control must be broken. As the AI Now report lays bare, these firms operate in obscurity, shielded by proprietary claims and immense capital, while consuming public resources and reshaping our economy. Therefore, the strategic objective is a political one: to subject the development and deployment of AI infrastructure to democratic, public control. This transforms the fight from a local land-use dispute into a national struggle over who owns and benefits from the core technologies of the 21st century.

Mandate transparency

The industry’s opacity regarding its resource consumption is a political choice. Therefore, demanding transparency is a primary leverage point for the struggle for public control. For example, demanding a nationwide moratorium on new data center construction is a tool communities could use to assert control. With sufficient community pressure, data center permits could be halted until federal laws compel reporting of real time energy usage, including water, electricity, and carbon usage.

Fighting human exploitation

The fight against the physical footprint of AI must be linked to ending the exploitative labor practices that underpin it. Bringing organized labor into this fight is key. No public subsidies should be provided for an industry built on high-tech sweatshop labor. This unites local environmental concerns with the global struggle for workers’ rights, threatening the industry’s business model on two fronts.

A public alternative

A purely negative campaign is insufficient. The fight against corporate AI must be paired with a positive vision for a “People’s AI.” This means advancing a political program for public investment in, and ownership of, computing infrastructure. This infrastructure must be governed for the public good, not for profit. There must be democratic accountability and principles of data sovereignty.

AI in organizing

While fighting corporate AI, organizers can pragmatically use smaller, more accessible software models to train their own AI. The cost of running a capable, small language model (SLM) has plummeted. However, their limitations must be understood. SLMs are powerful text-manipulation tools, not strategic thinkers. Here are a few types of SLM:

• Tactical use: SLMs excel as powerful administrative assistants. They can reliably summarize long documents (e.g., meeting minutes), produce first drafts of communications from bullet points, and check for stylistic consistency. All output requires human review.
• Operational use: SLMs can assist with data triage, such as classifying canvassing notes into predefined categories. Using Retrieval-Augmented Generation (RAG), they can act as a “smart search” for an organization’s internal knowledge base. However, they are prone to errors and “hallucinations” and cannot replace human judgment.
• Strategic use: Strategy requires analysis, world knowledge, and an understanding of power — all of which are beyond the capabilities of current SLMs. Their use at this level is not recommended.

Here is a video presentation with discussion exploring the threats posed by modern technology and AI under capitalism, as well as the promise of modern technology and AI under socialism.

The opinions of the author do not necessarily reflect the positions of the CPUSA.

Images: Artificial Intelligence by Nicky Pe ( CC0 1.0 Universal), The Future of Graphic Design: AI, Automation and Creativity by Dazon Technologies (CC0 1.0 Universal), AI on the rise by Gerd Altmann (CC0 Public Domain).