Mark Zuckerberg wants to double Meta's AI computing infrastructure to 14 gigawatts by 2027 — a scale of centralized computational power that outstrips the energy consumption of entire nations — and he's locking in multi-year supply contracts to make sure nothing stops him. The stake for ordinary Americans isn't abstract: this is the infrastructure that will train the AI models deciding what you see, what you're allowed to post, and whether your voice counts on the platforms billions use to communicate.
According to an internal memo reported by Reuters, Meta plans to deploy seven gigawatts of computing infrastructure this year alone, then double that by 2027, with spending running as high as $145 billion — a meaningful slice of the more than $700 billion Big Tech is projected to pour into AI this year. The memo confirms Meta's in-house AI chip, code-named "Iris," enters production at TSMC in September after clearing bug validation in just six weeks with no major issues — an unusually clean result for a program that has, by Reuters' own description, "floundered since its launch more than half a decade ago." Broadcom is locked in as design partner through 2029. A new chip ships roughly every six months through 2027, against an industry norm of annual or slower cadences.
Follow the money. Meta has signed long-term contracts for memory from Samsung, flash storage from Sandisk, and fiber-optic equipment from Sumitomo Electric — multi-year lock-ins struck in the middle of a memory shortage severe enough that Morgan Stanley analysts say "chipflation" has become a macroeconomic concern. Apple is already raising consumer hardware prices. Ordinary Americans pay at the register while Big Tech hoards supply.
ZeroHedge, alone between the two outlets, framed the real tension: the market can't decide whether Meta is exercising capital discipline or abandoning it. When Bloomberg reported Meta was standing up a cloud business — "Meta Compute" — to sell surplus capacity and token-metered API access, the stock ripped nearly 9 percent higher. Days later, leaked town-hall remarks showed Zuckerberg conceding that agent development "hasn't accelerated in the way we expected," and the stock dropped. Now a memo describing a doubling of capacity looks, as ZeroHedge put it, "rather undisciplined when it comes to capex." Reuters buried this context entirely.
But the market's feelings aren't the real story. The real story is what 14 gigawatts of centralized AI compute means for free speech. Meta's AI powers Facebook and Instagram — platforms that have spent years building censorship machinery, from algorithmic suppression to coordinated fact-check networks to deplatforming. The AI models trained on this infrastructure will be shaped by the same censored data environment. When one company controls this much computational power and trains its models on data it has already curated, filtered, and suppressed, the output isn't neutral. It's an amplifier for whatever worldview the censors permit.
The memo itself is blunt about why in-house silicon matters: adopting the latest external GPUs at Meta's scale "has been a heavy lift, and it has cost us time." Independence from Nvidia and AMD sounds like good business. But independence from external chipmakers also means no external accountability. When the hardware, the training data, the model, and the platform are all owned by one company, there is no check anywhere in the chain.
Fourteen gigawatts isn't just a number on a capex spreadsheet. It's the architecture of control. The question isn't whether Meta can afford it — the question is whether a free society can afford to let one firm build it without answering for what that machine will be told to think, and what it will be told to silence.
