Hyperscale: The Ambition and Excess of Big Tech's Data Empires

by Paris Marx

LONGLISTED FOR THE 2026 FINANCIAL TIMES & STANDARD CHARTERED BUSINESS BOOK OF THE YEAR AWARD

“A tremendous contribution to a debate that is of existential importance to us all.” —Naomi Klein, bestselling author of Doppelganger

“Hyperscale is a book Big Tech does not want you to read.” —Malcolm Harris, bestselling author of Palo Alto

An eye-opening account of how the data center boom is pushing our planet past its breaking point

Underlying our seamless access to AI, cryptocurrency, and unlimited cloud storage is a hulking, costly physical infrastructure. Massive data centers full of energy-hungry chips and processors power everything we do online. Scrambling to meet dubious market demand, the tech giants have leveraged our planet in pursuit of a market monopoly.

In Hyperscale, journalist Paris Marx uncovers the excesses of hyperscale data processing. He exposes the deceptive, exploitative ways tech companies are siphoning energy and resources from vulnerable communities in their pursuit of digital power, abandoning the pretense of social responsibility.

Amid energy shortages and tech-induced battles over drinking water, Marx makes a powerful case for technology that promotes the social good over the bottom lines of tech billionaires. He joins the incredible efforts of people and movements pushing back against the profit-driven interests of Big Tech, illuminating a path toward the sustainable future we all deserve.

The AI boom and its infrastructural demands have accelerated our many modern catastrophes. With razor-sharp reporting and incisive analysis, Marx shows us that the greatest threat is not the specter of AI overlords but a global environmental disaster already in the making.

Genres: Nonfiction, Technology, Politics, Artificial Intelligence, Business, Audiobook, Economics

318 pages, Kindle Edition

First published October 20, 2026

About the author

Paris Marx is a Canadian tech critic and host of the acclaimed Tech Won’t Save Us podcast. He speaks around the world about the politics of technology and writes the Disconnect newsletter. Paris is the author of “Road to Nowhere: What Silicon Valley Gets Wrong about the Future of Transportation” and “Hyperscale: The Ambition and Excess of Big Tech’s Data Empires”.

Reviews

[Sarah Jaffe · Rating 5 out of 5]

Reading for blurb so that will be available shortly but in the meantime here: it's hard to think of a more timely book than this one. You'll want it.

[Kathleen Littleton · Rating 4.25 out of 5]

This book is great with a couple of caveats. The author nails the electricity market and finance issues. His treatment of the water is limited by the fact that he focuses on drought areas with minimal consideration of areas with higher average rainfall. He mentions their issues with wastewater and stormwater runoff only in passing, although he identifies it correctly as a big problem. Likewise, he only mentions the issues with waste heat in passing despite its huge impact. The last chapter is not of the same quality of the rest of the book. He sets up a conflict between degrowthing and letting the tech billionaires control everything. This is a false choice. This basically accepts the billionaire's contention that they are smarter than everyone else and any other solution is "less'. There is no evidence for this. There is no reason to believe that regulation will lead to a "less" future. In fact, constraints often lead to a better outcome because it needs to be more carefully thought out.

[Kaleigh · Rating 5 out of 5]

Exhaustive and definitive—THE book on data centers (and kind of e-waste in general too). Wish I could make every politician that solicits data centers to their city/district/state read this!! Also every tech bro billionaire is an enemy to life itself and I spend every day amazed that they don't live in fear.

[Thomas · Rating 4 out of 5]

What is the backbone of the digital, and therefore physical, supply chain? Data centers. The storehouses of the servers that allow for the increasing interconnectedness of our world and the cloud storage and streaming so many of us rely on or even just make use of in our daily lives. However, with the birth of "AI", and before that cryptomining, not to mention the mass data collection for the sake of the modern hell of targeted advertising, the data centers have had to expand in size and total numbers. Now the ramifications in terms of land use, resource use, and the social destabilization that social media, AI, and digital technologies have become clear. Paris Marx does a fantastic job of describing the structure of these new hyperscale data centers, their impact on communities across the world, and the forces for and against their implementation. Most certainly recommended reading to get a full picture of what we, as humanity, are up against in the modern day.

Thank you to Riverhead Books and NetGalley for this ARC.

[Demetri Papadimitropoulos · Rating 4.75 out of 5]

Every Euphemism Eventually Gets a Utility Bill
In “Hyperscale,” Paris Marx takes the cloud, compute, AI, and innovation at their word – then follows them back to cooling vents, transmission lines, polluted factories, locked-in customers, and the increasingly awkward question of who pays.
By Demetris Papadimitropoulos | October 14th, 2026

Drive through Ashburn, Virginia, and the future does not announce itself with chrome, glass, or anything so obligingly futuristic. In Paris Marx’s “Hyperscale,” it first looks like suburbia: broad roads, enormous houses, clipped lawns, a playground, a thick wall of trees. Then the sightline changes. White industrial walls “peek” above the foliage; farther off, high-voltage lines “tower” over them. Marx says the trees act as a “shield.” Those verbs do nearly a chapter’s worth of work. “Peek” makes the warehouse seem furtive; “tower” restores the disproportion; “shield” admits that concealment has been built into the view. The electrical apparatus is harder to hide than the white box itself, and the landscaping has taught both to pass as background. Digital life is not weightless. It has simply been placed behind the trees.

Even that first reveal is smaller than the book’s eventual field of vision. Marx begins with the physical machinery behind cloud computing and generative AI, then keeps following its dependencies until the warehouse has become a map of electricity, water, climate, semiconductor fabrication, land, finance, surveillance, military use, and political sovereignty. The nine chapters move from “Buildout” and “Cloud” through “Power,” “Water,” “Climate,” and “Chips,” then into “Control,” “Acceleration,” and “Human Scale.” The data center is less a destination than a junction box. Touch one wire and another system lights up.

Marx’s governing distinction arrives early: “the issue we’re facing today is one of scale.” That saves “Hyperscale” from the lazier argument that computation itself is the sin. A small facility can sit in a city without swallowing a municipal water forecast. The problem appears when facilities become enormous, cluster together, and serve a business model that profits by making ordinary digital life more computationally hungry. Marx then asks the more troublesome question: do we need “all this computation” in the first place? Most infrastructure debates start after demand has been accepted as fact. He moves the argument backward, to the people and institutions allowed to define demand as necessity.

Everywhere in the prose, abstraction is made to acquire a body. Marx likes to take a frictionless technology word and make it sweat. His cleanest trick is also his funniest: the only literal clouds in cloud computing are the plumes of water vapor escaping from cooling systems. The joke turns on one noun. “Cloud” promises altitude, softness, distance; vapor from a cooling vent gives you heat exchange, pipes, water, and a building bolted to the ground. A metaphor that helped erase infrastructure begins giving evidence against itself. The sentences are mostly medium-length, lucid, and impatient with jargon. Marx is not a stylist in the peacock sense. He does something more useful here: he turns words back into racks, turbines, cables, heat. The method is explanatory, but also adversarial. Every euphemism is asked to show its utility bill.

The movement from person to system is just as important. Brenda Rizzo’s Pennsylvania farm gives the electrical chapter its scale in human terms. A proposed transmission line would cross land that had been in her late husband’s family for decades, including the area where his ashes were buried; the initial compensation offer was one dollar. The scene does not prove every transmission project abusive, and Marx is better for not asking it to. It makes a phrase like “grid expansion” suddenly indecently concrete. Gigawatts have to cross somebody’s field.

Residents of South Memphis and former semiconductor workers in Taiwan extend that logic without merely duplicating it. In historically burdened Boxtown, people wake to a gas-like smell and connect their fear to xAI’s Colossus complex, whose electricity needs pushed the company toward methane turbines while permit questions mounted. In Taoyuan, the story of semiconductor ascendance passes through RCA’s contaminated plant and workers who suffered cancers, miscarriages, and poisoned groundwater. The digital and the industrial are not parallel metaphors here. They are one supply chain. A model may live in software, but the GPU beneath it has an industrial biography. Marx’s reporting is best when that biography is allowed to remain specific rather than instantly converted into a slogan.

If there is one recurring thought more interesting than the book’s headline argument, it is cost migration. A cooling design can lower direct water use while raising electricity demand. New renewable generation can feed a data center without displacing much fossil power if total load grows faster than clean supply. More efficient chips can coexist with rising total fabrication. The question becomes less “Did this company improve a metric?” than “Where did the burden go?” One small, almost comic example carries the logic beautifully: a data center that reused waste heat for a greenhouse looked clever until humidity helped turn the place into an insect buffet. The launch photograph was greener than the system. That distinction between improving a process and shrinking a total burden is one of Marx’s best defenses against corporate sustainability language, and it is also where the book is most intellectually reusable beyond data centers.

Part of the book’s personality comes from Marx’s ability to catch ideology at the moment it becomes hardware, zoning, or municipal signage. Ashburn contains Cloud Plaza, Performance Circle, and, most gloriously, Growth Mindset Lane – a management seminar paved into geography. Elsewhere there are locked doors, seven-foot server racks, metallic hum, warm exhaust, diesel generators painted Caterpillar yellow. These details keep the book from floating away into “Big Tech” as a generalized villain. Even Marx’s jokes work best when they stay attached to an object you could stub a toe on.

As a structure, “Hyperscale” keeps promising tidy compartments and then breaking them open. “Power,” “Water,” “Climate,” “Chips”: each title looks bounded until the chapter proves otherwise. Water means electricity. Chips mean water, chemicals, labor, and mining. Cloud ownership means legal jurisdiction. AI financing means new generation capacity. The effect is cumulative but also mildly corrective. Each chapter teaches the reader that the previous category was too small. This is why the book’s sprawl often feels earned before it begins to feel expensive. Marx is not simply adding topics. He is showing that the borders between them were administrative conveniences, useful to an industry that would rather have its environmental, financial, and political costs audited on separate spreadsheets.

Perhaps the smartest formal turn comes in “Control,” when Marx returns to South Dublin with the sentence, “But I didn’t tell you the whole story.” Earlier, the industrial park had meant grid strain, gas generation, security, and residents trying to learn what had moved into their neighborhood. On the return, the same gray buildings are reopened through controversies over Microsoft Azure, military surveillance, worker protest, and cloud ownership. Even two men riding lawn mowers past Marx’s group, despite grass too short to need mowing, take on a different comic unease. The detail is too slight to prove anything, which is exactly why it works: it registers as atmosphere before the chapter gives surveillance a harder institutional meaning. The first account was not false. It was partial. “Hyperscale” repeatedly asks what happens when the frame widens by one more system.

An intellectual risk follows from that widening. The path from cloud dependency to digital sovereignty is strong because the mechanism is visible: institutions build themselves on infrastructure that can be prohibitively expensive to leave and subject to laws made elsewhere. The path from hyperscale to every neighboring pathology of Silicon Valley is less secure. Surveillance, militarization, authoritarian politics, longtermism, social isolation, venture finance, and AI ideology all belong somewhere in Marx’s argument, but not with equal causal force. The later chapters sometimes treat adjacency as if it were another cable in the rack.

Dublin therefore cuts both ways. Its return proves that infrastructure can carry several meanings at once, but it also shows Marx’s appetite for making the data center explain almost everything around it. The problem is not thin reporting. It is prosecutorial overdetermination. Once a case has landed, Marx will occasionally call another witness, then another, then summarize the jury instructions himself. His anger is earned often enough that the habit rarely feels cheap; it does, however, reduce the amount of friction available inside his own argument.

In the middle of the book, the same pressure appears as repetition. Oregon, Virginia, Pennsylvania, Chile, Uruguay, Georgia, Ireland, Tennessee, Taiwan: the locations change, but a political sequence becomes recognizable. Investment is promised. Resource demand is minimized or obscured. Officials accommodate. Residents learn more. Resistance follows. A mitigation measure may simply move the burden elsewhere. Recurrence is necessary to prove a system rather than an anecdote, yet proof has diminishing returns. The pace improves whenever a new chapter changes the question instead of supplying another jurisdiction in which the old answer reappears.

Marx is also less exact when his rhetoric reaches for maximum voltage. The Introduction rises from documented concern into a neat crescendo: chatbots are “breaking people,” image generators are “breaking reality,” and data centers may “break the world.” The parallelism is memorable, but its syntactic perfection makes unlike harms march in suspiciously perfect formation. The International Energy Agency’s 2026 outlook provides a useful check on the temperature: it projects global data-center electricity consumption at roughly 950 terawatt-hours by 2030, around 3 percent of worldwide electricity demand, even as AI-focused facilities grow far faster. The International Energy Agency also stresses uncertainty, which matters in a sector whose efficiency, capital spending, and adoption curves are moving quickly. None of this shrinks Marx’s case. It locates it. A load can be a modest slice of global consumption and still overwhelm a particular grid, municipality, aquifer, or rate base.

Inside that distinction, “Hyperscale” becomes more precise than its fiercest lines suggest. Electrical systems do not experience a worldwide percentage; they experience a giant request for connection at a particular node. A household does not experience an average water footprint; it experiences a well, a rate increase, a drought restriction, a planning decision. Marx keeps restoring that granularity. His real subject is not simply resource consumption but the political sequence by which a private forecast becomes a public obligation. If a company announces that AI will require gigawatts of new capacity, why is the first question so often how quickly the grid can provide it rather than what the proposed use is for?

The hard part is that some demand is real, and some benefits of scale are real too. Marx’s own cloud history makes that inconvenience impossible to dodge. Amazon Web Services did not conquer business computing solely because executives were hypnotized by Jeff Bezos. Elastic capacity solved problems. Companies could survive traffic spikes without owning enough hardware for their maximum possible load; startups could launch without building server farms. One of Marx’s interviewees, looking back on a cloud migration, says, “We let the behemoth in through the back door.” The line bites because the door was opened for a reason. Convenience became dependence through a rational choice, not a bargain that looked evil at the time.

Resistance to Marx’s thesis should begin there. Large systems can create interoperability, redundancy, availability, scientific capacity, and efficiencies people actually value. A huge public system could serve social needs; a tiny private one could still be exploitative. “Human scale” is therefore a moral and political orientation, not a measurement. Who decides which purposes warrant enormous computation? Medicine? Climate modeling? Language preservation? Entertainment? Defense? Search? Marx persuasively argues that the answer should not default to whichever company can raise the most capital. He is less specific about institutions capable of making those judgments without becoming slow, captured, or technologically dependent themselves. Nor can public ownership magically solve the difficulty: “Control” has already reminded us what states can do with surveillance and military systems. A democratic veto can be captured just as a market can be monopolized. The book knows enough to make its own prescription troublesome, and that trouble should remain open rather than be tidied into a policy slogan.

Once “Human Scale” begins, the argument gets an unexpectedly mundane stress test. An Amazon Web Services outage knocks smart beds, Roblox, Ring doorbells, businesses, and hospital communications offline because of a DNS failure. Nearly two decades after Bezos imagined cloud computing as utility-like infrastructure, the dream has arrived in an awkward form: AWS is essential enough to resemble a utility, privately governed enough not to behave like one, and embedded enough that leaving can be ruinously expensive. Marx’s choice of examples is sly. A smart mattress and a hospital system do not belong to the same moral category, but the outage places them on the same technical dependency graph. After the grand rhetoric of “Acceleration,” this is a delicious correction. Civilizational computing trips over a blank DNS entry.

Placed immediately before that outage, the end of “Acceleration” becomes even stranger. Ilya Sutskever is quoted imagining the earth’s surface covered with data centers and power stations, a “tsunami of computing” arriving “almost like a natural phenomenon.” The metaphor removes agency as efficiently as any press release. “Tsunami” supplies inevitability; “natural phenomenon” supplies innocence. Tsunamis do not hold earnings calls, hire lobbying firms, seek tax abatements, sign power-purchase agreements, or choose jurisdictions with accommodating utility policy. They happen to you. By borrowing the grammar of natural force, technological ambition makes political decisions sound meteorological. The next chapter title, “Human Scale,” corrects the grammar before it corrects the politics. One page imagines planetary computation as weather; the next shows a private utility failing because two automated processes collide.

Of all the book’s countermodels, Te Hiku Media in Aotearoa New Zealand gives that correction a physical form. The Māori-owned nonprofit has archives, broadcasting work, a language community to serve, and technical problems arising from that purpose. Its small server room, easily mistaken for a storage closet, supports speech-recognition tools built around Māori language needs and community trust. When Te Hiku asked for language recordings, the community supplied hundreds of hours under a Kaitiakitanga License that treats the data through guardianship rather than indiscriminate ownership. The organization calls itself a caretaker, not an owner. That verb matters as much as the hardware. Te Hiku does not build maximal capability and then search for markets large enough to justify it. It begins with a community, asks what that community needs, and chooses technology accordingly. Purpose precedes scale.

Under the appeal of that scene sits a useful complication. Te Hiku still needs servers and cloud services, and Marx concedes that Amazon’s globe-spanning network offers conveniences smaller systems do not yet match. His best alternative is therefore not local good, global bad. It is a system in which scale keeps answering to purpose, ownership, accountability, and a knowable constituency. Marx’s visits to Catalyst Cloud and T4 sharpen the point: both are commercial operators and both want growth, yet their national orientation and more candid accounting alter the relationship between provider and user. “Human scale” matters most when it names accountability, not acreage.

Less comfortably for the prescription, sovereignty does not guarantee restraint. The European Commission’s 2026 European Technological Sovereignty Package seeks less dependence on non-European Union providers while also encouraging major expansion of European cloud and AI capacity. A different flag over the server hall does not lower the power draw. Nor does jurisdiction by itself answer Marx’s question about necessity: Europe can own more of the stack and still decide that the answer to American hyperscale is European hyperscale. That policy tension is a better companion to Marx than another alarming statistic because it tests the distinction he wants to make. Control can move without demand shrinking. If sovereignty becomes only another race for capacity, one vulnerability recedes while the growth logic survives intact.

On my scale, “Hyperscale” lands at 88/100, which translates to 4.75/5 on Goodreads. The score reflects a book whose conceptual architecture and reporting outweigh its repetition, rhetorical maximalism, and reluctance to let opposing cases exert sustained pressure. It is especially good at making one kind of sentence – “AI requires more compute” – expand until its hidden subjects appear: utilities, regulators, residents, miners, fabs, gas turbines, investors. The figures will date; some named projects will be overtaken before the hardcover has gathered much dust. What should last is the suspicion Marx directs at the moment when a commercial forecast quietly becomes a public necessity. Once that transformation is visible, “innovation” stops functioning as a permission slip.

Smallness, finally, is not what makes Te Hiku’s server room memorable. It is that the room has an answer to the question the giant facilities so often evade: what is this for? “Hyperscale” cannot tell us exactly how much computation is enough, and I trust it more for leaving that measurement unsettled. The useful reversal is subtler. The billion-dollar facility usually arrives with its need already presumed; the community is asked to accommodate it. Marx asks that presumption to run in the other direction. Make the purpose answer first. Then let the scale defend itself. The next time a white wall appears beyond the trees and the transmission lines follow, the question is no longer merely how to connect it. The question is why it must be there at all.

[Matthew Bruemmer · Rating 4.25 out of 5]

Thanks to Riverhead for the advanced copy.

As someone whose work is tied to data center growth and the hyperscale of technology, I was impressed by how simply and effectively Marx communicates on the topic. He is clearly an expert in the field and knows how to relate to the average reader. This is a very clear and concise book about the effects of data centers and leaving the control into the hands of technology oligarchs. At times, I do think he can be overly black and white about specific companies and individuals without truly reviewing his bias. However, much like our political society, sometimes the facts are this black and white and I think he overall does a good job staying in the lane of facts vs bias throughout the book. I would have enjoyed more of his commentary on what can actually be done in our systems and laws from a practical approach in our capitalistic society.

[Eliza's Reading · Rating 5 out of 5]

Hyperscale by Paris Marx would be a fascinating read if not for it being so terrifying. The ramifications of what will be left of this planet and humanity because of the free rein that governments have given to big-tech companies is criminal. Living in Virginia, the place with the most data centers worldwide, this book is a must read. Paris Marx goes into great detail over the demands that AI places on our environment and communities with the incredible amount of water and electricity that is required to run each data center. And it’s not just the data centers but also how we are obtaining the minerals and materials to make the chips to power this technology. It’s always the most vulnerable among us that suffers the most.

Marx writing on this subject is both easy to understand and accessible for those of us not really computer or AI savvy. It’s obvious Marx knows and has researched this topic endlessly. Seeing how we are hearing more and more people and communities effected by these centers that are located near them, this book is very timely and necessary. Now we need to find out what we can do to hold these companies accountable and what legislation we can put in place that don’t contain loopholes in order to protect our environment and humanity at large. Doing nothing is not an option.

I beg everyone to read this book so you are prepared when you discover an unknown company is buying up land near you.

Thank you, Paris Marx and Riverhead Books for providing this e-ARC via NetGalley for review.

[Robert Dohrendorf]

Rarely has a book landed at a better moment.

Hyperscale follows the physical footprint of the AI boom: enormous data centers, the energy and water they consume, and communities that learn about them only after the decisions have been made, often hidden behind non-disclosure agreements. Marx combines history and reporting to make an otherwise abstract topic very tangible.

The section on dark data was particularly eye-opening. Vast amounts of data are stored that will never be used, yet keeping them alive consumes real energy, around the clock. Think of the data generated by autonomous vehicles: how much of it will ever serve a meaningful purpose? It is a striking example of a system that no longer seems to ask what all this is actually for.

While reading, I couldn't shake an image from Nick Bostrom: the paperclip maximizer. Marx is not writing about superintelligence, but the underlying logic rhymes. Here, the metric is scale, the paperclips are data centers, and the “everything else” is power grids, water tables and local democracy.

What stayed with me personally: the companies I once admired as innovators now carry a reputation my younger self could never have imagined. I suspect the social licence they have enjoyed for so long is going to be increasingly contested.

Highly recommended for anyone who wants to understand the infrastructure behind the AI hype — and is willing to question it.

[P · Rating 1 out of 5]

There is some good content in this book. For example, if you're really, really ignorant about how the internet and technology work, there is some valuable material here. But even this small compliment comes with a devastating caveat -- much of it is wrong or poory researched. The author talks about some of the truly evil and depraved people behind AI but then turns around and quotes others to support their arguments. They talk about water usage and rely upon statistics that have been completely debunked and wholly ignore modern techniques. Many of the themes have been well worn over, and this book adds nothing new to the discussion, but it's a solid summary of the doomer angle. It's written messily (sort of like this review), lacks insight, and it's only driving force seems to be "screw all this!" It really could have used more erudition from those who actually know what they are talking about -- her acknowledgements read like a who's not who of people who I've never heard of in AI. There are plenty of reputable AI pioneers and experts one could quote, and plenty of valid criticisms of data centers, but this screed does a disservice to the discussion with its lack of seriousness. I was quite thoroughly disappointed. Do better, Paris Marx. 2/10

[Fay Abernethy · Rating 5 out of 5]

Have you ever wondered: why, oh why, is generative AI being pushed at us so relentlessly, despite its questionable benefits for normal people? Why are hyperscale datacentres being built with, frankly, indecent haste, to the detriment of local communities and our environment?

Paris Marx, host of the brilliant Tech Won’t Save Us podcast, has conducted an in-depth enquiry into just this subject. The results are scary. The push for AI has little to do with business logic (although making obscene amounts of cash while the party lasts is definitely a factor) and more to do with a particular ideology that is detrimental to all life on Earth, with the exception of a few individual humans at the top of the pile. Guess who they might be?

This book is a rallying call for everyone who thinks there are better ways to structure our technology.

[Brian Mikołajczyk · Rating 3.75 out of 5]

A survey of the impact of data centers with a focus on recent scaling and the how AI companies are all growing at breakneck speeds for little ROI.
Each chapter covers a specific issue: Water, Power, Climate [change], Jobs, etc.

[Wilhelm Arcturus · Rating 2.5 out of 5]

Interesting and relevant information, but a tedious read.

[Oli · Rating 1.5 out of 5]

Maybe I know too much about data centres already (possible) but I found this book quite surface level in its exploration. The writing style ok to good though quite a scattergun approach to examples.

[Tim · Rating 3.5 out of 5]

Very accessible a read. You've probably heard all the stories that are in this book if you've been paying attention to tech journalism. Lend to a family member that is uninitiated.

[Sinéad Leavy · Rating 3.75 out of 5]

Irelands features quiet a bit in this short book.