For Every Public Purpose There are Corresponding Private Interests: AI Data Centers are No Exception

Many discussions of large artificial intelligence data center investments focus on the expected benefits of job creation. But there also are costs, which might include the lost tax revenue for states which offer tax incentives to data center operators; higher consumer power bills; higher water consumption; stranded utility assets and even far fewer new jobs than anticipated. 

In that regard, large AI data centers remind me of large sports stadiums: the argument is that government subsidies will promote economic growth. In fact, such investments mostly shift consumer spending from one category to another, with possible net zero gains. 

Study / Source	Main Finding on Net Economic Benefits
Baade and Dye (1990), Growth & Change	No consistent evidence that stadiums boost metropolitan income or employment.
Noll & Zimbalist (eds.), Sports, Jobs & Taxes (1997)	Broad survey finds subsidies rarely generate promised jobs or growth; most benefits overstated.
Coates & Humphreys (2008 review)	Majority of peer-reviewed studies show little to no positive impact on city-wide income/employment.
Baade and Matheson (2000, 2004) – Super Bowl studies	Event impacts overstated; spending largely shifted from other activities, little net new growth.
Matheson (2018, J. Policy Analysis & Management)	Citywide gains minimal; limited neighborhood or intangible benefits possible, but not enough to justify heavy subsidies.
Mercatus Center  (2015 working paper)	Finds no evidence subsidies increase growth; in some cases correlate with lower growth.
Bradbury, Coates, Humphreys (2022/2023 survey)	Comprehensive literature survey: consensus remains that stadium subsidies almost never deliver large net benefits.
Policy briefs (Journalists’ Resource, CBCNY)	Ex-ante booster studies overstate benefits; opportunity costs of public spending usually higher than stadium returns.

That is not to deny the need for large AI data centers, simply to point out that the local economic benefits might not be as often touted.

Study/Report	Year	Promised/Expected Jobs	Actual/Direct Jobs	Key Findings
Upwind Industry Research	2025	N/A (hype around economic catalysts)	1,688 (construction, temporary); 157 (permanent operations)	Construction jobs are short-term and high-volume, but permanent roles are minimal relative to facility size (~$243M economic add during build vs. $32.5M annually ongoing); underscores automation's role in limiting ops staffing.
University of Michigan / STPP Report ("What Happens When Data Centers Come to Town?")	2025	High-paying jobs as primary benefit (per subsidy pitches)	Few permanent positions (not quantified)	Jobs generated are disproportionately low compared to subsidies (e.g., billions in tax breaks); recommends redirecting funds to education or renewables for better ROI; no broad economic uplift observed.
Wall Street Journal (on Stargate AI Project)	2025	Employment "bonanza" (tech/political hype)	~100 full-time	Lowest jobs per square foot of any major facility (1M sq ft campus); contrasts with 500 jobs at a nearby 286k sq ft cheese plant—exposes AI boom as labor-light post-construction.
Good Jobs First (analysis of subsidies) - Google Indiana	2025	200 jobs (governor's announcement)	30 (local tax abatement minimum; no state mandate)	Massive gap between promises and enforceable commitments; subsidies flow without job guarantees, prioritizing corporate gains over local employment.
Good Jobs First (analysis of subsidies) - Amazon Indiana	2025	>1,000 jobs	400 direct (plus 600 subcontractors)	Subcontractor roles often lack benefits/wage parity; highlights "illusory" totals and long timelines (e.g., spread over years), diluting immediate impact.
Oxford Economics (for Google Data Centers)	2018	N/A (focus on multipliers)	1,900 direct (national); ~700-1,700 total per state	Positive multiplier (3.3-4.6x, national 5.9x) boosts indirect jobs to 11,000 total, but direct ops roles remain low (~100-200 per campus); early study predates AI surge but shows pattern of limited core employment.
CBRE Economic Ripple Effects	2024	N/A	3.5M direct-related (up 20% from 2017)	Strong growth but concentrated (e.g., 490 ops + 1,500 construction in NE, 2022); 7.4x multiplier inflates totals, yet critics note it masks low direct density in AI era.

Then there are the other issues such as higher consumer electricity bills and impact on water usage. 

McKinsey estimates suggest that by 2030, data centers globally will  require $6.7 trillion in investment for compute operations, of which $5.2 trillion in capital expenditures will support artificial intelligence operations. 

If correct, that represents nearly $7 trillion in capital outlays by 2030, and would be an increase of 3.5 times the capacity of data centers from 2025 to 2030 alone. 

source: McKinsey 

source: McKinsey

Data center power needs in the United States alone are expected to add about 460 terawatt-hours of demand from 2023 to 2030, three times the current level of consumption, McKinsey estimates. At the same time, data center water demand could rise about 170 percent by 2030, according to analysts at WestWater Research. 

Those forecasts might be wrong on the high side, but even so, much to all of that capacity will mostly have to be built, somewhere.

And the point is that the benefits and costs will accrue to different participants in the information technology value chain, in different quantities. There will probably be less benefit for local economies, taxpayers and electricity and water ratepayers than often is assumed.

70% of IT and AI Projects Fail for Simple Reasons

Information technology investments are often treated as purely technical endeavors rather than organizational transformations that require changes in processes, culture, and human behavior, which possibly explains the gap between IT investments and observed results. 

source: McKinsey 

Some of the exceptions might be retail; communications and media, where productivity seems higher. For communications and media, technology often creates the platform for services, delivering a higher degree of observable value. 

source: McKinsey 

Still, many studies suggest that IT projects have a high failure rate overall. 

Study/Source	Year	Key Finding	Sample Size/Scope	Failure Rate/Metric
Standish Group CHAOS Report	2020	Only 31% of IT projects are successful (on time, on budget, with required features)	50,000+ projects across multiple industries	69% challenged or failed
McKinsey Global Institute	2012	Large IT projects run 45% over budget and 7% over time, while delivering 56% less value than predicted	Analysis of 5,400 IT projects	17% of projects are "black swans" with cost overruns >200%
Harvard Business Review - Flyvbjerg & Budzier	2011	Average cost overrun for large IT projects is 27%, with one in six projects having cost overruns of 200%	Study of IT project performance patterns	16.7% massive overruns
PwC Global CEO Survey	2019	73% of CEOs believe their digital investments are not delivering expected returns	1,378 CEOs globally	73% not meeting ROI expectations
Deloitte Tech Trends	2021	70% of digital transformation initiatives fail to meet their goals	Survey of 1,000+ executives	70% failure to meet objectives
MIT Sloan - Brynjolfsson & Hitt	2003	IT productivity paradox: firms with higher IT spending don't always show proportional productivity gains	Longitudinal study of 527 large firms	Mixed correlation between IT spending and productivity
Gartner IT Spending Analysis	2019	85% of big data projects fail to deliver business value	Analysis of enterprise big data initiatives	85% failure rate
Accenture Technology Vision	2020	Only 37% of organizations successfully scale their digital pilots to enterprise-wide implementations	Survey of 4,000+ business and IT executives	63% fail to scale successfully
Boston Consulting Group	2018	70% of digital transformation efforts fall short of their goals	Analysis of transformation initiatives across industries	70% shortfall rate
KPMG Global CEO Outlook	2018	65% of CEOs question whether their technology investments create competitive advantage	Survey of 1,300 CEOs	65% uncertain about competitive value
IBM Institute for Business Value	2019	Organizations realize only 20% of anticipated benefits from AI investments	Study of AI implementation across enterprises	80% benefit shortfall
Forrester Research	2020	60% of customer experience technology investments fail to improve customer satisfaction scores	Analysis of CX technology implementations	60% fail to improve target metrics
EY Digital Transformation Study	2018	55% of digital transformation programs are abandoned before completion	Survey of 500+ executives across industries	55% abandonment rate
Capgemini Digital Transformation Institute	2017	Only 36% of organizations are digital transformation leaders achieving significant benefits	Study of 1,000+ organizations globally	64% are laggards or followers
McKinsey Technology Trends	2021	Cloud migration projects deliver only 65% of expected cost savings on average	Analysis of cloud transformation initiatives	35% savings shortfall

"Five Nines" is Gone and Almost Nobody Actually Cares

Our user experience of applications, devices and networks is far from the “five nines” standards (99.999 percent availability) telcos used to tout. And it probably doesn't matter as much as some would suggest, as we all now have protocols for dropped calls, lost emails, momentary glitches in streaming content.

If the value of an experience or app is high enough, we simply work around availability or "reliability" issues. "Best effort" means some amount of failure for any given instance of an app session.

So, as a practical matter, today’s heterogenous, edge-powered, internet transport fabric, IP-based application environment absolutely means user experience cannot approach 99.999-percent availability for any applications. And nobody cares much, most of the time.

To be sure, some might claim "core systems" in banking, financial trading or some security-critical use cases are "five nines" in terms of availability, but that only applies to the core systems, not the end user experience of those systems. 

The problem is that no matter what any single participant in the value chain might claim for its own availability, and even if that availability is between 99 percent and 99.99 percent, the entire end-to-end value chain depends on the sum total of availability across the whole value chain, and that math is challenging. 

Consider an example where contributor availabilities are:

• Device: 99%

• Home broadband access: 99.5%

• Internet backbone: 99.99%

• App server: 99.9%

• Local power: 99.5%

The end-to-end availability requires multiplying all those discrete availabilities. So the formula is 

0.99 × 0.995 × 0.9999 × 0.999 × 0.995 ≈ 97.4 percent. That means 229 hours of downtime per year, not the 5.26 minutes per year allowed by "five nines” standard.

The only reason end users seem unaware of the change is that much of the downtime happens when they are not actively using their connections (devices not present; devices in “do not disturb” mode; user is sleeping; apps not in immediate use). 

Value Chain Component	Typical Availability (%)	Major Downtime Factors
User Devices – Mobile	95%–99%	Battery loss, OS/software issues, dropped connections
User Devices – Fixed	96%–99.5%	Power outages, device crashes, local network (Wi-Fi) issues
Access Network – Mobile	97%–99.9%	Tower outages, congestion, interference, maintenance
Access Network – Fixed	98%–99.9%	Fiber/cable cuts, power issues, last-mile failures
Global Internet Backbone	99.99%+	Rare fiber cuts, DDoS attacks, routing errors
Application Servers (Cloud)	99.5%–99.99%	Cloud region outages, software bugs, maintenance, cyberattacks
Local Power Supply	99.0%–99.9% (urban)	Grid instability, storms, infrastructure failures
End-to-End Availability	Often < 95%–98%	Cumulative failures across components

The point is that people have learned to adapt to services, networks, applications and experiences that are never "five nines" reliable. We have work-around solutions that funciton so well nobody really cares about the lack of "five nines" reliability.