Hyperscale: Paris Marx on the Ambition and Excess of Big Tech’s Data Empires

“I would take the crash over the success, because the crash gives us an opportunity to rethink.” — Paris Marx

“Scale” is such a loaded word. In Silicon Valley, it’s the thing-in-itself, the holy grail. So what happens if you scale up “scale”? You accelerate to Hyperscale, the title of a new book by the big tech critic Paris Marx, who has turned Silicon Valley’s holy grail against itself.

So what, exactly, does the H word mean? We say “big tech” which means everything and nothing. In his use of hyperscale, Marx means something very specific. His hyperscalers are multi-trillion-dollar conglomerates like Amazon, Microsoft, Google and Meta, who own the computation powering the AI revolution. These are the data center guys. Frontier labs like Anthropic and OpenAI aren’t.

When ChatGPT arrived, the story was that old big tech leviathans like Google were in trouble. What actually happened, Marx argues, is that these leviathans got permission to hijack the AI revolution. They had the models and data centers. What they didn’t have was cover, because the regulators were watching. Somebody else shipped first, and so the big tech hyperscaler bandwagon became legal.

Follow the money, Marx says. Seven trillion dollars invested into data centers by 2030, raised not through government bonds but mostly through corporate ones. Crypto’s collapse only hurt the people who bought crypto. This irrational exuberance will dramatically affect pensions and 401(k)s.

So I asked him the multi-trillion-dollar question. If this bet pays off, these companies become more powerful than ever. If not, we are all f&*ked. So, which is worse for a critic: hyperscale’s success, or its failure?

Marx is a man of the left. So he favors failure. Lenin wanted Russia to lose World War One to hasten the revolution; Marx, more modestly, suspects that the collapse of the hyperscale pyramid would at least give us the chance to rethink. And then what? Marx wants us to have “conversations” about the type of economy we want. Hyperconversations, more like. Very noisy ones.

FIVE TAKEAWAYS

1. At least the coal mine made jobs, and the line is his. Data centres arrive expecting a tax package, the construction ends, a skeleton staff remains and the abatement runs for years. He thinks the bargain is breaking, and that the opposition is hopeful precisely because it is not ideological.

2. The ICC lost its email. After US sanctions, the court’s chief prosecutor was locked out of his Microsoft account, and 1,800 workstations have since moved off Microsoft. Washington can reach into an institution it does not control, and the companies can reach into governments that outsourced their plumbing.

3. Why generative AI and not some other AI: because the people funding it are pursuing the replication of the human mind in a machine. That is a science-fictional priority, not an efficient one, and a narrower programme would need a fraction of the computation.

4. Nvidia is not a hyperscaler, it is the Levi’s of the gold rush — and a company with one revenue stream has nowhere to fall back to. Which is exactly why Michael Burry picked it as the cleanest short. He is similarly blunt about Oracle.

5. He breaks with Bernie Sanders on the record: welcomes the moratorium, is uncomfortable with the Bannon allyship, and argues Sanders has absorbed the executives’ account of the technology and so debates future harms instead of present ones.

THE BOOK

Hyperscale (Crown in the US, Verso in the UK, 2026) — https://www.penguinrandomhouse.com/books/797020/hyperscale-by-paris-marx/

Tech Won’t Save Us — https://techwontsave.us/about/

ABOUT KEEN ON AMERICA

Keen On America is a daily conversation about the ideas, books and arguments shaping the United States, hosted by Andrew Keen, author of The Internet Is Not the Answer and How to Fix the Future. New episodes every day.

CHAPTERS

00:00 — Intro
01:26 — What hyperscale actually means
02:14 — Is hyperscale a better word than big tech?
02:47 — The cloud companies and the frontier labs
04:23 — How SpaceX became an AI company
06:27 — Karen Hao, and the imperial reading
09:03 — Who is paying for all this
10:05 — The ICC loses its Microsoft email
11:00 — Crypto hurt buyers; this reaches pensions
13:10 — 1929, 2008, and too big to fail
16:27 — Why a bond crash is worse
19:20 — “I would take the crash over the success”
20:38 — AK: their goal is profit, not control
25:34 — Nvidia: hyperscaler, or Levi’s?
28:29 — Huang throws the labs under the bus
30:21 — How ChatGPT gave Google permission
35:39 — Bernie Sanders and Steve Bannon
39:47 — Digital sovereignty, and New York’s pause
41:04 — At least the coal mine made jobs
46:26 — What transportation should have taught us
50:29 — Closing