Video Summary
Data centers are the physical backbone of the internet and AI; their growth drives massive energy and water demand.
Communities often oppose new data centers because of grid strain, water use, noise, and broken economic promises.
AI scaling (OpenAI's law) fuels a hyperscaler arms race that concentrates data centers and raises infrastructure costs.
‘Data gravity’ causes data centers to cluster, amplifying local socioeconomic and environmental impacts.
Much of internet work is hidden labor (content moderation, microtasks) often outsourced to vulnerable populations in the Global South or conflict zones.
The video highlights high electricity consumption that stresses local grids, extensive water use for evaporative cooling (often using potable water), diesel backup emissions, and noise pollution.
Site selection favors abundant land, tax incentives, existing electricity and water infrastructure, and strategic co-location advantages—cooler climate helps but is less decisive than power and policy.
Modern AI (especially large models) requires rapidly increasing compute and GPU capacity; OpenAI-style scaling means compute demand doubles quickly, prompting hyperscalers to build more facilities to stay competitive.
Data gravity describes how one data center attracts others (economies of scale, shared infrastructure), creating clusters that raise local utility costs, strain resources, and can displace other industries or residents.
It emphasizes microtask workers and content moderators—often low-paid freelancers or vulnerable populations—who perform data-cleaning and moderation work essential to AI and the internet but are rarely visible.
"The creator of this work denounces violence and the destruction of property in the strongest possible terms."
The video emphasizes that it is an educational and artistic exploration focused on the internet and its infrastructure.
It clearly states that any suggestions to the contrary are satirical, ensuring viewers understand the intent behind the content.
"Local officials often see them as an opportunity for economic development."
Data centers are increasingly being proposed in rural areas due to the availability of land, raising concerns among local communities about their impact.
There is anxiety regarding how these data centers will affect farmland, irrigation, and livestock watering needs for ranchers.
The community's perception is mixed, with some believing it's essential to adapt and support these developments for economic growth.
"A recent Gallup poll found that seven out of ten Americans oppose their construction."
Despite the economic arguments made by proponents, there is significant grassroots opposition against new data centers largely due to concerns about their resource consumption.
The opposition stems from worries about high energy and water requirements, which are often downplayed in discussions about these facilities.
Additionally, the noise pollution produced by data centers is noted as a problem.
"Data centers need to draw massive amounts of energy that are only going to go up as they continue to grow."
The operational demands of data centers are placing a strain on local power grids, leading to increased utility bills for residents.
They typically function with N1 redundancy, allowing them to avoid outages, but in case of a significant regional blackout, diesel backup generators are employed despite their considerable failure rate.
The implications of energy consumption raise concerns about long-term sustainability and local infrastructure resilience.
"Most data centers dump that heat into water for what's called evaporative cooling."
Data centers require substantial amounts of water for cooling purposes, contributing to local water table depletion through evaporation.
They often utilize pristine drinking water for their operations, adding to the complexities surrounding water resource management.
The truth about water usage in relation to data centers is nuanced; while some companies downplay their needs, critics often exaggerate the actual impact.
"The owners of tech companies have amassed money and power that's completely unprecedented."
The increasing presence of data centers reflects deeper socioeconomic issues, including how tech industry leaders wield power over local environments and communities.
The narrative surrounding data centers as manifestations of AI raises questions about their ethical implications and overall societal value.
The portrayal of data centers evokes feelings of them being a blight, representative of broader environmental exploitation by powerful corporations.
"When you actually dig into the purpose of AI, you quickly realize that the tech oligarchs making this technology aren't particularly subtle about their motivations."
The speaker expresses skepticism towards both the mainstream view of AI and the idea of rejecting it entirely, highlighting that the discussion often focuses on individual consumer usage while overlooking the broader implications of AI technology.
He suggests that the nature of AI development is not benign, indicating a close relationship between technological progress and economic inequality, predicting a world with high GDP growth but equally high unemployment and disparity.
"I think all the white-collar jobs are going to get reordered to May, and I think we're going to see massive hyperinflation in blue-collar salaries."
The speaker predicts a shift in the job market due to automation, where many white-collar jobs will become obsolete, and blue-collar workers may see substantial increases in wages, driven by a surge in demand for skilled trades such as welding.
There is a humorous commentary on the potential future job landscape, suggesting a dichotomy of employment in sectors heavily influenced by AI, where jobs may become scarce for the majority while opportunities remain for a select few.
"The promise of AI is that it can be used to usher in a world of total surveillance, privatization, and complete capital capture in the hands of a few tech oligarchs."
The speaker discusses the darker potential of AI, asserting that it may facilitate a nefarious societal structure characterized by extensive surveillance and privatization, which could consolidate power among a small elite.
He underscores the fundamental intent of AI investments, which, despite popular narratives, are centered around military and industrial applications rather than benign consumer-focused technologies.
"Data centers are, for the foreseeable future, the one choke point still vulnerable enough that we regular people can still wrap our hands around it and squeeze."
The speaker emphasizes the importance of targeting data centers as a focal point for activism against AI, arguing that local opposition can exert political pressure on their construction and operation.
He mentions that whether people are inclined toward radical actions or favor regulatory frameworks, the community still has leverage to influence AI development through scrutiny of data centers.
"Historically, there have been four main criteria for choosing locations for data centers."
The speaker introduces the key factors that influence where data centers are constructed, starting with the idea that they are often located in cooler climates to reduce cooling costs, although this isn't the most decisive criterion.
Tax incentives and infrastructure, particularly access to electricity and water, are outlined as more significant factors in determining data center locations, with a particular note on the strategic co-location of data centers with Amazon fulfillment centers for operational efficiencies.
"In critical geography called code space, the computer code that runs them is just as, if not more important than the physical infrastructure."
The concept of "code space" illustrates the significance of the software that drives data centers, emphasizing that their functionality relies not only on physical components but also heavily on the underlying code.
An incident involving a software deletion that led to a failure in an important air traffic system serves as an example of how crucial this code is, reinforcing the idea that the virtual aspects of data infrastructures merit as much attention as their tangible counterparts.
"Essentially, without those lines of code, every airport in the country stopped being airports and were instead transformed into massive buildings filled with angry people."
"Amazon uses a lot of advanced proprietary techniques to ensure their warehouses are running at peak efficiency."
"Despite proving to be effective at shaving precious seconds off the time it takes to get each item, the layout is so confusing that not even seasoned workers are able to make sense of it."
"Once data centers come to an area, they quickly start to multiply... this process of techno-forming an area creates a vicious cycle known throughout the industry as data gravity."
"Most data centers aren't actually owned by tech companies themselves, but instead by private firms called real estate investment trusts."
"Data Center Alley lives up to its name as the largest data center market on Earth."
"While you probably won't be surprised to hear that the negative effects of data centers disproportionately affect poorer racialized areas, they are actually likely to be found in more affluent areas too."
"Connectionism has two major advantages: it is better at producing commercially viable products and it requires massive amounts of data for training."
The rivalry between symbolists and connectionists in AI research peaked during the mid to late 1960s, with symbolists initially gaining the upper hand.
By leveraging substantial financial resources, connectionists ultimately succeeded, largely due to their capability for commercial applications and access to vast datasets from wealthy companies.
Notably, the development of generative AI as we know it today stemmed not from Dartmouth but from tech giants like Google and newer entrants such as OpenAI.
"OpenAI began as a nonprofit founded on fears that unchecked AI could lead to global disaster."
OpenAI was established in 2015 by a team that included Sam Altman and Elon Musk, motivated by Musk's concern over Google's AI advancements.
Musk's apprehensions centered on the potential emergence of superintelligent AI, which he believed could pose an existential threat if not controlled.
To counter this, OpenAI was initiated with an impressive $30 million in funding.
"OpenAI's law indicates that AI compute usage doubles every 3.4 months, posing significant scaling challenges."
A pivotal moment for OpenAI came in 2017 when the research community, including Google, published the transformer paper, detailing a novel approach for training neural networks.
Recognizing the scalability potential of this architecture, OpenAI aimed to exponentially increase its compute resources, which led to the creation of "OpenAI's law"—a stark contrast to Moore's law.
Despite existing challenges and the need for faster chips, OpenAI opted to procure more graphics processing units (GPUs) to meet the demands of scaling AI technologies.
"The revelation of OpenAI's scaling strategy ignited an arms race among major tech companies to establish data centers."
This focus on scaling has turned AI research into a competitive landscape, characterized by the dominance of hyperscalers like Amazon, Microsoft, Google, Meta, and Alibaba.
The rapid investment in data centers not only aims to capture market share but is also crucial for fueling AI advancements in a race against time and resources.
OpenAI's ambitious plans for a significant data center were compared to the Manhattan Project, reflecting the escalating competition among tech giants.
"In 2026, the big five hyperscalers are projected to generate $320 billion but will spend $800 billion, necessitating a trillion-dollar revenue by year's end."
Despite significant revenue projections, hyperscalers face dire financial realities, as they often overspend in pursuit of rapid growth.
A striking statistic indicates that only about 3% of individual AI users actually pay for AI services, and many businesses have reported no real improvements from AI adoption.
Companies are navigating cash flow issues by building even more data centers, adopting creative financing strategies such as long-term loans that delay repayments.
"There's a physical limitation on the pace at which new data centers can be established and integrated into existing grids."
The process of constructing data centers and connecting them to the power grid can take several years, creating a bottleneck that threatens the industry’s ability to meet demand.
This limitation raises concerns about the sustainability of AI growth, with predictions that advancements will stall without sufficient infrastructure to support them.
The situation underscores a broader issue where the rapid expansion of AI technology could outpace the reality of physical and resource constraints.
"AI will just fix this... the thing is, while training and usage become more efficient over time, all of that efficiency gets eaten up by higher demands."
The argument that artificial intelligence (AI) will naturally resolve issues related to efficiency is both misleading and overly optimistic. Although AI does improve training and usage over time, the increased efficiency is often negated by the growing demands placed on these systems.
The speaker challenges advocates of AI by requesting evidence of any occasion where AI power consumption was lower than in the previous year, indicating dissatisfaction with their claims.
It is emphasized that skepticism towards AI does not equate to a dismissal of the technology itself. There is an acknowledgment that AI could potentially contribute positively in some domains.
"While the big tech companies will be fine, for the rest of us, things probably aren't going to turn out as well."
The transition of tech companies towards decentralized data centers and self-sustaining power sources, predominantly nuclear, indicates a strategic shift. This shift aims to mitigate risks associated with over-reliance on traditional energy and infrastructure.
The speaker expresses concern that any setbacks facing AI companies could lead to broader economic stagnation, particularly affecting countries heavily invested in AI technologies, including the US and Japan.
A potential financial crisis is noted, where a drop in stock prices could threaten pensions and broader market stability, leaving ordinary citizens in a precarious situation.
"The metaphor of the cloud is a vague world of make-believe used to handwave away technical concerns."
The concept of cloud computing is rooted in much older ideas than most realize, tracing back to AT&T's conceptual illustrations as early as 1951. This illustrates that the challenges faced today are not novel but have historical parallels in technological evolution.
The speaker references an 1884 lecture series by John Ruskin, who identified a "new kind of cloud," metaphorically relating it to societal developments of his time, highlighting persistent public concern about atmospheric changes—both literal and metaphorical.
"Ruskin called his new cloud a storm or plague cloud, describing its malignant qualities."
Ruskin's lectures delivered in the 19th century foresaw a transformation attributed to industrialization and capitalism, which he believed was corrupting nature and society. His emphasis on the "plague cloud" signifies a prelude to environmental awareness.
He attributed darker atmospheric changes not merely to industry but speculated they might be linked to larger, supernatural forces stemming from human conflict, particularly from battlefields.
This nuanced view of clouds as indicators of societal malaise reaffirms that technological and environmental concerns have been deeply intertwined throughout history.
"Baron was worried about the sky... developing a communications network for the US military that could survive a full-scale nuclear attack."
In the era of escalating Cold War tensions, Paul Baron of the Rand Corporation sought to conceptualize networks capable of enduring catastrophic events, highlighting safety and resilience in military communications as paramount.
His exploration of communication structures revealed patterns critical to understanding how the military could adapt in the face of global conflict and technological change.
This historical perspective sets the stage for current discussions on the infrastructure behind AI and data centers, linking back to the pervasive influence of conflict on technological advancement.
"The network's true weakness wasn't their location, but their shape."
"Baron named these centralized systems and warned the military of the risk they posed, suggesting the construction of a new kind of network arranged in an interconnected web."
"Two years after Baron's unsuccessful meeting with the military and telecom brass, another team of government scientists began to work on what would be the most important network of the modern age."
"In short, Paul Baron had unknowingly designed the internet."
"Modern-day submarine cable landing points are the not-so-distant descendants of the telegraph stations."
"Just as the Chinese laborers who broke and poisoned their bodies constructing the railroads were systematically erased from all photographs, so too is the labor performed by people in the Global South that keeps the internet working hidden from view."
"It's worth asking why this narrative of the internet as a weightless distributed cloud has stuck around for so long."
"The zeitgeist of the 60s was dominated by the fear of an apocalyptic nuclear confrontation as well as an obsession with interconnected networks."
"Not everything is actually connected, and in fact, thinking that it is is itself a symptom of paranoia."
"In the early hours of May 28th, 1961, three AT&T microwave receiver stations exploded one after another."
The explosions occurred in the Great Salt Lake Desert, near the Utah-Nevada border, due to dynamite that was strategically placed at each station and detonated using a homemade detonator.
These stations were targeted because of their remote locations, which resulted in minimal security and their crucial role as a choke point connecting California to the rest of the nation.
The destruction caused disruption of all calls, telegraphs, and television signals for over 20,000 people; however, services were quickly restored.
"This incident ignited a wave of paranoia, leading to rampant speculation about possible attackers."
The bombing triggered widespread fear and conspiracy theories, suggesting involvement from the Kremlin or even the Pentagon.
In reality, the attackers were a small group known as the American Republican Army, comprised of just two individuals—Jerome Bruce and Dale Jensen—motivated by a shared disdain for AT&T and their customer service practices.
"Three weeks later, the FBI apprehended Bruce and Jensen while they were on a yacht in Mexico."
Upon their arrest and extradition to the U.S., fears about the attack began to diminish as the true culprits were identified.
Despite Bruce's efforts to claim responsibility by mailing letters to AT&T, it was evident his messages went unnoticed, which he humorously attributed to the company's poor customer service.
"Despite the attackers being caught, paranoia still spread among officials."
"Less than two weeks after the incident, the Senate proposed a bill to enhance communication infrastructure security."
This bill aimed to extend protections against malicious damage to privately owned communication facilities that were utilized by the military.
Complications arose concerning which circuits should be classified for military versus civilian purposes, as many military operations depended on secret infrastructure.
"The entire network fell under military jurisdiction, a situation complicated by the dynamic nature of communication circuits."
This highlights the evolving intersection between civilian technology and military needs, as it's difficult to pinpoint which parts of the network are being used by the military at any given time.
The vagueness of the definitions led to convoluted legal situations surrounding cyber-security and communication infrastructure.
"Historically, governments leveraging extensive communication networks often hinder their citizens' access."
The video references incidents such as the Egyptian government's shutdown of internet access during the Arab Spring, showing that oppressive measures can be taken against communication networks in times of unrest.
Similarly, in San Francisco, in reaction to public protests, authorities preemptively disabled wireless communication to disrupt organizing efforts.
"The cloud conceals the true nature of the internet as a vital resource composed of physical infrastructure and human labor."
This emphasizes the contradiction that greater connectivity often corresponds to increased vulnerability to government intervention, as observed in the suppression of online communication during protests.
Such actions illuminate the tension between a desire for freedom of communication and the mechanisms that enable governance and control.
"Undersea wires have always had significant military applications, particularly in espionage."
Undersea cables have played a crucial role in global communication networks, tracing back to the telegraph era when they served both colonial and military purposes.
An example of this is John Pender, a British merchant who in 1871 covertly laid underwater telegraph lines along the west coast of Hong Kong Island, linking China to Europe, defying the Qing government's wishes.
In more modern times, undersea cables have caught the attention of organizations like the CIA, where director Allan Dulles saw their potential for control and surveillance, indicating a continuous thread of military interest in these infrastructures.
"The cloud hides the labor that goes into running the internet."
A significant aspect of the internet's functionality is supported by micro work, which includes unskilled computer tasks that are just beyond what can be automated, such as content moderation and AI training.
Amazon’s Mechanical Turk, inspired by a historical chess-playing automaton, is one of the largest platforms for hiring people to perform these tasks.
The laborers typically involved in this work are often underpaid freelancers, who manage the digital operations that sustain the cloud, similar to how the Mechanical Turk was an illusion disguised with intricate human effort.
"This work gets much darker than people solving captures for a living."
The hidden tasks within the internet often involve exposing workers to disturbing content that is deemed inappropriate for a Western audience, indicating a darker side of content moderation.
Gig workers, such as those in Kenya, have been tasked with filtering harmful material for AI training, enduring low wages while facing graphic images as part of their job description.
These roles are particularly prevalent among vulnerable populations, such as refugees who find themselves in dire economic situations and take on jobs that allow them to work from home, enabling them to avoid social restrictions while earning a living.
"The cloud not only emanates from sites of conflict, it depends on them."
Political unrest can create a swell of cheap labor, as seen in regions affected by violence, where the instability indirectly benefits tech companies seeking low-cost workers.
Reports have indicated how global conflicts, like those between Israel and Palestine, can lead to increased availability of skilled workers who are underemployed due to the turmoil in their regions.
In Northern Iraq, for example, Kurdish refugees have been recruited for tech jobs that involve tasks like tagging images for self-driving car technology, showcasing a trend where displaced populations are absorbed into the workforce of tech companies.
"For data to seem decentralized, it actually needs to be concentrated in one place."
Data centers are essential for the functioning of the internet and artificial intelligence, serving as the physical locations where data is managed and processed.
These centers have gained visibility due to their increase in numbers, yet the companies behind them often employ strategies to remain concealed, using shell companies and non-disclosure agreements to prevent transparency.
Google’s initial dealings for their first data center exemplified this approach, negotiating through an anonymous entity while requesting numerous benefits from local officials without revealing their identity.
"You couldn’t come up with a worse example of AI water usage if you tried."
The environmental impact of data centers is significant, particularly concerning their high consumption of water and energy, which raises concerns especially in areas facing drought.
Google faced scrutiny for planned projects that would exacerbate water shortages while simultaneously claiming eco-awareness during negotiations.
Despite increasing demand for data centers, they continue to operate under a veil of secrecy, attempting to hide their substantial resource consumption from the public eye.
"This is something I actually had a bit of personal experience with a few nights ago when I decided to pay a visit to one of my local data centers here in Toronto."
The narrator shares a recent attempt to film at a local data center in Toronto, named Surfer Farm. The name reflects a nondescript approach that helps it blend into the industrial setting around it.
Upon arrival, the narrator notices unusual security features, such as fences and painted windows, hinting that high security is a priority at this facility.
Initially, the narrator was able to film without interruption for about 15 minutes before security arrived to intervene.
"First, this guy came out and asked me what I was doing... I could see from his uniform that he worked for a big third-party security company."
The first security staff member who approached seemed more concerned with making the narrator leave than escalating the situation.
However, a second security guard, who appeared to be directly affiliated with the data center, had a significantly different demeanor. He demanded the narrator leave immediately and stated he had called the police.
This guard further insisted that the narrator show all footage before departing, which was not legally required.
"It seems a bit odd to me that the natural assumption would be that the public can't know or even film the exterior of this government-subsidized industry."
The narrator raises critical questions about the necessity for such extreme privacy and security around a government-funded industry, which contradicts public interest.
The narrator perceives the company's name, Server Farm, to be misleading in implying secrecy, especially since the government actively promotes the expansion of data centers.
This contradiction prompts a discussion on the broader implications of privacy and surveillance in the tech industry.
"Most data center briefs specified that they don't want their workers in the photographs."
Industry practices show that data center promotional materials are often devoid of human figures, emphasizing a clean, technological image.
This absence aligns with the perception that human error is the leading cause of downtime in data centers, suggesting companies prefer to present facilities as hyper-automated and uninterrupted by human presence.
Historical parallels are drawn to landscape photography that excludes human beings, reflecting a colonial mindset about uninhabited land.
"It turns out that very few data centers are making pictures of technotopian landscapes the old-fashioned way anymore."
The narrator notes a significant shift in the data center industry towards utilizing AI for promotional imagery, further distancing the depiction of these locations from human involvement.
This trend produces images that lack a sense of authenticity and connection to reality, as they are likely created for an audience that doesn't actively engage with the spaces depicted.
The narrator humorously declares their intention to boycott digital infrastructure utilizing blockchain until human photographers are re-established in this visual narrative.
"Data centers have generally been an extremely niche subject for research, and what little has been written about them is more than four years old from when they served a very different purpose."
Over the past four years, data centers have transformed dramatically in size and function, evolving from primarily storage facilities to centers focused on computing tasks.
Understanding their historical context is essential to grasping this shift, as prior to the rise of AI, data centers centered predominantly around data storage, with variations in security rather than size determining their quality.
High-end data centers were typically designed to safeguard sensitive information for governments, financial institutions, and other organizations, emphasizing security protocols that, despite their appearance, often had little to do with actual vulnerabilities.
"Since the leading cause of data center outages or leaks isn't an invading army, but human error, all this Cold War bunker-style security is pretty much just symbolic."
The industry often emphasizes robust security for advertising purposes, but the primary threats to data centers result from human mistakes rather than external attacks.
This disconnect between perceived and actual vulnerabilities is illustrated in popular culture, such as the depiction of the Steel Mountain data center in the television show Mr. Robot, which dramatizes security concerns to an unrealistic level.
"If data centers have always been bunkers, crypts that prefigure the resurrection, then the thing that they’ve kept buried has been our data, which thanks to AI, is now being reanimated."
The comparison of data centers to bunkers and tombs introduces a deeper philosophical discussion about their purpose and function in society.
Drawing on theorist Paul Virilio's ideas, bunkers are portrayed as shelters meant for survival amid crises, symbolizing a space where people conceal themselves while waiting for an eventual return to normalcy.
The critical revelation is that the data stored within these centers is akin to buried treasure, which, influenced by advancements in AI, is now rising to the surface, ready to resurface in various aspects of our lives, complicating our relationship with technology.
