A good deal has been written about AI. In investment circles we spend a good deal of time trying to assess its costs, risks and benefits. Who will win this race of implementing LLMs and agentic AI? How will these technologies be commercialized? What type of infrastructure will they require? And, in climate related investment circles, how much power will it take to run more datacenters and how much water will be required to cool them? Unfortunately, we don’t really spend an inordinate amount of time on the human side of this equation, what will it do to us?
In 1952 Kurt Vonnegut wrote Player Piano. Vonnegut was a prolific novelist and this book, his first, was not his most famous work; he is better known for Slaughterhouse 9. But Player Piano may turn out to be his most lasting work.
Player Piano is set in a dystopian future in which much of humanity has lost its purpose due to mechanization. Written in a sardonic vein, those without Phd’s are considered nearly uneducated and people without bachelor degrees are supposedly under skilled and are forced to either join the army – itself highly robotic, or join the Wreaks and Wrecks – a civic corps that digs ditches, picks up garbage and performs a whole host of chores deemed too menial even to train the machines. Interestingly enough, the novel prophesied the professionalization of college football. In one scene, Cornell football players bemoaned their lower pay compared to rival schools.
Throughout the novel Vonnegut emphasized how people felt they had lost purpose in life. A subplot of his story was that humans worked in service of the machines as opposed to the other way around. Ultimately, this ended up in a violent confrontation that had populist strains that echo some of the things we are undergoing today.
We seem to be barreling towards the reality Vonnegut foresaw right now. A good deal of attention has been placed on the infrastructure demands of AI, but perhaps not enough on its potential human cost or impact on society.
AI is taking over many white-collar jobs while many blue-collar jobs are increasingly being replaced by robots. Mechanization of our economy is hardly a new phenomenon; but the recent speed of job replacement, particularly around knowledge workers, has been stunning. In the second quarter of 2025 the unemployment rate for recent college graduates was 5.3%[i] Media offers a host of anecdotes, many citing possible collateral damage to come in addition to the cost to individual lives[ii]. Perhaps some of the job losses could well be cyclical as many firms in tech, consulting and finance staffed up in recent years but they are substantial. In the fall of 2025 Amazon announced job cuts totaling 14,000; though many of these are likely white-collar jobs the company’s recent investments in robotics also suggest that blue collar jobs are also at risk.
Graduates in Computer Sciences appear to be doing even worse as AI is now replacing many coding roles. Computer science and computer engineering jobs, recently viewed as tickets to six figure starting salaries, are facing high unemployment rates of 6.1 and 7.5% respectively[iii]. A bitter irony is that a few short years ago students were encouraged to learn coding by billionaires and Presidents alike. Granted many college degrees offer wonderful, or not so wonderful educations, that turn out to not have much direct practical training for the workplace. But this phenomenon did not usually apply to fields outside of the humanities.
Over the past several months, as demand and forecast demand for AI has increased notably, spending has accelerated commensurately. Last fall Open AI entered a $38 Billion deal with AWS to accelerate its planned 30 gigawatt data center buildout, agreed to pay Oracle $30B annually for data center services as part of the Stargate project involving 4.5 gigawatts of capacity and entered a deal with Nvidia involving a $100 billion commitment[iv]. Amazon has added 3.8 GW in the past 12 months doubling capacity since 2022 and Microsoft projects over $90 billion in FY26 capex.
The AI space, or the consumer facing AI space is presently an oligopoly. Much like search engines or social networks it is hard to imagine many more entrants competing for our attention in consumer facing LLMs or even consumer generative AI. A useful analog may be search which is being replaced by LLMs; as of October 2025 over 90% of the search market was controlled by Google.[v] The late 90s saw a much more fragmented industry as Alta Vista, Lycos and Excite amongst others competed in this category but are long gone. It seems unlikely there will eventually be multiple players in consumer-facing AI.
The business AI space, as companies figure out how to deploy and utilize agentic AI may turn out to be a very different case. There may well be a large group of firms competing with each other over different market segments.
All of this course requires more servers though perhaps not as much as are currently forecast, Chinese models like DeepSeek and Qwen are built on different architectures that are cheaper and more energy efficient though there is reputational concerns around data.
Regardless of what unfolds we will be building more data centers. In the case of consumer demand this suggests a data center buildout. The electricity demand for a single new AI – training data center is 1 gigawatt or more, more than the entire state of Wyoming currently consumes. In a December 2024 report, Lawrence Berkeley National Laboratory estimated that data centers consumed about 4% of US electricity, that number could rise, the report estimated, to 12% by 2028[vi]. Much of this will be cooled by water, or quite possibly, not cooled by water but subsequently require even more electricity to run.
And of course there are widely varying estimates regarding water intensity for AI. A recent study by Global Water Intelligence and Xylem highlighted the fact that water use for cooling may not be as high as previously thought since many systems recycle their water[vii]. It can be a closed loop. Regardless, water is used in semiconductor fabrication and renewable energy so we will still have to use more water.
Of course we don’t really know what will happen but given prior examples of newly introduced technology a host of changes will be forthcoming. Given the power of AI in certain applications it is not hard to imagine a good deal of disruption to individual lives, career paths and society as a whole. We are right to worry about the material costs of the AI deployment; unfortunately there doesn’t seem to be much thought the accompanying human costs.
[i] Newsweek, It’ s the Worst Time to be a College Graduate in Years, September 04, 2025. Data from NY FED.
[ii] Wall Street Journal, Heard on the Street: The Kids Aren’t Alright, November 4, 2025
[iii] New York Times, Goodbye, $165,000 Tech Jobs. Student Coders Seek Work at Chipotle, August 10, 2025.
[iv] OpenAI was on the hook at the beginning of November for over $1 Trillion of spend assuming $30B per gigawatt compute – Author estimate.
[v] Statcounter Global Stats, found on Google Gemini.
[vi] Circle of Blue, The Year in Water, 2025 – Power Shift, December 9. 2025.
[vii] Global Water Intelligence and Xylem, Watering the New Economy, January 2026
