Bottom line: AI is not a fad, it is not a bubble waiting to pop, and it is not something you can wait out. The strongest signal across labor markets, capital investment, infrastructure buildout, and adoption curves is that AI is becoming core business infrastructure, much like the internet and the smartphone. That does not make every use case smart, every vendor credible, or every criticism wrong. It does mean business leaders have a practical choice: build fluency inside the business now or compete against organizations that already did. That is not hype. It is the reasonable conclusion from the data, including the parts of the data that should make us uncomfortable.
I am not going to sell a clean story. The resistance to AI is real, and in some cases, it is grounded in legitimate harm. A serious advisor does not wave that away to close a deal. So, let’s walk through it straight: jobs, intellectual property, data centers, and then the practical question that matters most for business leaders: where does AI create durable value?
The resistance isn’t irrational. Let’s look at it.
Jobs: the picture is uneven, not apocalyptic — but it’s not nothing either
The World Economic Forum’s Future of Jobs Report 2025 projects roughly 92 million roles displaced globally by 2030, against 170 million created. That’s a net gain, but a messy one, because the jobs created rarely match the skills of the jobs lost [1]. Stanford’s 2026 AI Index goes further, sharpening the picture: employment for software developers aged 22–25 has fallen nearly 20% since 2024, while employment for developers over 30 kept growing [2]. SHRM’s 2026 survey work echoes this, showing that displacement risk isn’t evenly distributed; it concentrates in roles with fewer “non-technical barriers” to automation, and it’s highest in food service and lowest in fields requiring judgment and physical dexterity [3].
Net: this is a real transition, concentrated at the entry level, not a mass unemployment event. Pretending otherwise, in either direction, is not honest analysis. As I wrote the other week, this looks less like a normal technology cycle and more like Industrial Revolution 2.0. The difference is speed: the first industrial revolution unfolded over roughly 120 years; this one may compress comparable business change into a decade or less.
Intellectual property: the courts are actively sorting this out, and it’s not going AI’s way by default
More than 70 copyright infringement suits are working through U.S. courts against AI companies right now [4]. The most honest headline is that results are mixed. Universal Music Group and Warner Music settled with Udio and Suno respectively in late 2025, trading litigation for licensing deals that put artists in control of how their voice and likeness get used [4]. But Thomson Reuters v. Ross Intelligence went the other way: a federal court found that training an AI tool on Westlaw’s headnotes wasn’t fair use, precisely because it competed directly with the original product [5]. Anthropic’s own $1.5 billion settlement in Bartz v. Anthropic over pirated training data is part of this same ledger [4].
Net: “AI is built on stolen work” is not a strawman, but it is also not a settled legal conclusion across every case. The honest position is narrower and more useful: some training practices created real legal exposure, the courts are drawing lines case by case, and the industry is being forced toward licensing, provenance, and compensation where the facts support it. For business leaders, that makes training-data provenance a real due-diligence item, not a philosophical footnote.
Data centers: this is the one worth slowing down on
The infrastructure argument matters because it tests whether the AI thesis can survive contact with real-world constraints. Jobs and intellectual property are transition risks. Power and water are operating constraints. If AI is becoming core business infrastructure, then the infrastructure behind AI must be part of the business conversation, not a side note for utilities and regulators.
This is where I want to push back hardest against both sides, because both sides are cherry-picking. The pro-AI side tends to treat power and water as engineering problems that will be solved later. The anti-AI side tends to treat every data center as proof of reckless extraction. Neither frame is good enough for business decisions.
The scale is not in dispute. U.S. data centers consumed about 183 terawatt-hours in 2024, roughly 4% of total U.S. electricity, comparable to the entire annual demand of Pakistan, and the IEA projects that climbs to 426 TWh by 2030 [6, 7]. Globally, data center electricity demand is set to more than double by 2030, with AI-focused facilities tripling their draw [7]. Water is the less-discussed number: the IEA estimates data centers consumed 560 billion liters of water in 2023, two-thirds of it indirectly through electricity generation rather than direct cooling [8].
Here is where the argument gets genuinely contested. Basic supply-and-demand economics explains why the fight breaks differently by region, even though that rarely gets said out loud. Senator Elizabeth Warren claimed residents near data centers have seen electricity bills rise as much as 267% in five years. PolitiFact checked that number and found it referred to wholesale grid capacity prices, not residential bills, a critically important distinction [9]. A Rutgers state policy lab study went further: a panel covering 22,834 utility-rate observations across 24 states found no robust, detectable link yet between a nearby data center and higher residential bills [10]. That “not yet” is doing real work. In power-abundant grids, much of the rural South and Midwest, a new data center can behave like another large industrial customer, spreading fixed grid costs across more usage and, in some documented cases, lowering residential rates [11]. In grids already running tight, the opposite can happen. PJM’s mid-Atlantic corridor saw its capacity auction price jump from $28.92 to $269.92 per megawatt-day in a single cycle, and Virginia residential rates are up more than 13% in a year [12, 13]. Net: this is not only a data center problem. It is a supply-curve problem: fixed, slow-to-build generation meeting a sudden surge in demand. That is the strongest practical argument for steering new capacity toward power-abundant rural regions instead of grids already running hot, and it is an argument the industry rarely makes out loud because it implies some markets should simply say no.
Water generates more real anger than any electricity bill, and it deserves the same rigor rather than one scary national number. Zoomed out, the national picture can look almost reassuring: U.S. data centers directly consumed roughly 449 million gallons of water a day in 2021, about 0.14% of total U.S. water consumption [14]. But that framing is exactly the kind of sleight of hand that erodes trust, because the draw is not spread evenly. It concentrates in specific watersheds, often dry ones. A single large AI-focused facility can use up to 5 million gallons a day on its own, roughly the water footprint of a town of 10,000 to 50,000 people, and the electricity that powers it adds an estimated 211 billion gallons of indirect consumption nationally [15].
The regional flashpoints make this concrete. In the West, Lake Mead sits at 32% of capacity and Lake Powell at 24%; the Bureau of Reclamation has declared a Level 1 shortage for 2026, cutting Arizona’s allocation by 512,000 acre-feet, about 18% of its annual apportionment, while a Western Resource Advocates analysis projects data centers across five Southwestern states could consume 7 billion gallons a year by 2035 [16]. In Georgia, the country’s second-largest data center market and a state currently under drought, residents in one Fayetteville neighborhood traced their falling water pressure to a nearby campus that had drawn nearly 30 million gallons through two industrial connections the county utility didn’t even know existed (E&E News/Politico, 2026). Texas tells the same story at bigger scale: the EPA projects U.S. direct data center water use could reach 38–73 billion gallons by 2028, and one Texas-specific estimate puts that state’s data centers alone at up to 399 billion gallons by 2030, comparable to draining Lake Mead [12].
The fair caveat, because this argument deserves one too: a Georgia Tech researcher reviewing the Fayetteville numbers noted the raw water figures alone don’t alarm him, there are plenty of farmland and other industrial users drawing similar volumes, and the county itself later walked back some of its initial accusations against the developer [18]. There are real fixes on the table, too: closed-loop and immersion cooling can cut a facility’s direct water draw by 70–90%, and states from California to the Great Lakes basin are starting to require water-efficiency review or consumptive-use permits before a new facility can break ground [19].
Net: on power, the data center fight is really a siting fight. Put the capacity where the grid has room, and the economics favor everyone. On water, it’s a genuine resource conflict in the places that can least afford one, and it won’t be solved by a national percentage that looks small on paper. This is exactly the kind of nuance that gets lost in a LinkedIn hot take, and it’s exactly the kind of nuance a business decision-maker needs before signing off on any AI-adjacent capital commitment. Any AI infrastructure conversation that skips the watershed map isn’t a complete conversation.
Where AI actually earns its keep: complement, not replacement
Here’s the data that should be driving your decision, not the headlines.
The most rigorous study we have on this, Brynjolfsson, Li, and Raymond’s Generative AI at Work, published in the Quarterly Journal of Economics, tracked 5,179 real customer support agents at a Fortune 500 software company through a staggered AI rollout. Average productivity rose 15%, but that average hides the real story: novice and low-skilled workers productivity improved by 34%, while the most experienced agents saw almost no gain, and, in the case of some measures, experienced a slight dip (Brynjolfsson, Li, & Raymond, 2025). The AI tool worked by capturing what your best people already know and putting it in front of everyone else, faster.
That is the complement thesis in one data point. AI does not replace your top performer. It cannot, because they are already near the ceiling. It compresses the time it takes your newest hire to become useful. Brynjolfsson has said this plainly: he does not think generative AI replaces workers, but workers who use it will replace those who do not [21]. I would sharpen that for the SMB owner reading this: the real competition is not worker versus worker. It is whether your business can absorb the learning curve of the whole team faster than your competitor can.
Where this matters for you: if you are a 25–250 person company in the Carolinas, you probably do not have a deep training bench, and you do not have six months to wait for every new hire to get fully productive. That is exactly the gap this data says AI closes fastest. This is not a strategy for cutting headcount. The study showing the biggest gains for your least experienced people is not a layoff memo; it is an onboarding accelerant. Businesses that misread it as a labor-cutting play will misapply the tool and trigger the backlash the resistance argument warns about. Businesses that read it correctly get faster-ramping teams and a real cost advantage.
Where this leaves you
The bottom line has not moved: AI is not optional and pretending the criticisms do not exist is not a strategy. The businesses that win this decade will be the ones that took the resistance seriously, built AI deliberately and people-first, and used it to make their existing team better rather than smaller.
That is the whole premise LumenForge was built on. We are not selling hype, and we are not pretending the data-center debate or copyright mess is somebody else’s problem. We help business leaders cut through it, vendor-neutral and evidence-first, then build an AI transformation plan that fits a business their size, not a Fortune 500 case study.
If you want a straight, no-fluff read on where your business stands, book the 30-minute AI Reality Check Call. No pitch deck. No pressure. Just an honest read on your exposure, your opportunity, and the smartest next move.
References
[1] World Economic Forum. (2025). Future of Jobs Report 2025.
[2] Stanford Institute for Human-Centered AI. (2026, April 13). AI Index Report 2026.
[3] Society for Human Resource Management. (2026). Automation, AI, and job displacement risk in U.S. employment: 2026 data brief. https://www.shrm.org/topics-tools/research/automation-generative-ai-and-job-displacement-risk-in-u-s–employment/2026-full-report
[4] Copyright Alliance. (2026, January 8). AI copyright lawsuit developments in 2025: A year in review. https://copyrightalliance.org/ai-copyright-lawsuit-developments-2025/
[5] Norton Rose Fulbright. (2026). AI in litigation series: An update on AI copyright cases in 2026. https://www.nortonrosefulbright.com/en/knowledge/publications/ce8eaa5f/ai-in-litigation-series-an-update-on-ai-copyright-cases-in-2026
[6] Pew Research Center. (2025, October 24). U.S. data centers’ energy use amid the artificial intelligence boom. https://www.pewresearch.org/short-reads/2025/10/24/what-we-know-about-energy-use-at-us-data-centers-amid-the-ai-boom/
[7] International Energy Agency. (2026, July). Data centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions. https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions
[8] Li, P., Yang, J., Islam, M. A., & Ren, S. (2025). The carbon and water footprints of data centers and what this could mean for artificial intelligence. Patterns. https://www.cell.com/patterns/fulltext/S2666-3899(25)00278-8
[9] PolitiFact. (2026, June 12). How much have data centers increased electricity prices? https://politifact.com/factchecks/2026/jun/12/elizabeth-warren/data-centers-rising-electricity-costs/
[10] Rutgers State Policy Lab. (2026). Are data centers raising your electric bill? Mostly not. Yet. https://policylab.rutgers.edu/publication/are-data-centers-raising-your-electric-bill-mostly-not-yet/
[11] Marketplace. (2026, July 10). Data centers lowered electric bills in some places—for now. https://www.marketplace.org/story/2026/07/10/data-centers-lowered-electric-bills-in-some-places-for-now
[12] Fortune. (2026, July 26). Data centers were actually making electricity costs cheaper, but the $7 trillion buildout with no guaranteed AI demand is threatening the trend. https://fortune.com/2026/07/26/data-centers-electricity-costs-cheaper-7billion-buildout-ai-demand/
[13] Brookings. (2026, April). Global energy demands within the AI regulatory landscape. https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/
[14] University of Georgia CAES. (2026). U.S. data center water consumption estimate.
[15] Environmental and Energy Study Institute. (2026). Data center water consumption and indirect energy-water impacts.
[16] Quartz. (2026). Data center water-demand projections in the U.S. Southwest. https://qz.com/
[17] E&E News/Politico. (2026). Data centers and local water pressure concerns in Georgia. https://www.eenews.net/
[18] Atlanta News First. (2026). Georgia data center water-use reporting and Fayetteville utility concerns. https://www.atlantanewsfirst.com/
[19] Information Technology and Innovation Foundation. (2026). Data center cooling, water efficiency, and state permitting requirements. https://itif.org/
[20] Brynjolfsson, E., Li, D., & Raymond, L. (2025). Generative AI at work. The Quarterly Journal of Economics, 140(2), 889–942. https://doi.org/10.1093/qje/qjae044
[21] CNBC. (2023, April 25). Stanford and MIT study: A.I. boosted worker productivity by 14%—those who use it “will replace those who don’t.” https://www.cnbc.com/2023/04/25/stanford-and-mit-study-ai-boosted-worker-productivity-by-14percent.html
Leave a Reply
You must be logged in to post a comment.