Terafab is Elon Musk's $20–25 billion semiconductor fabrication facility, announced on 21 March 2026 at Austin's Seaholm Power Plant, designed to consolidate every stage of chip production—from lithography to advanced packaging—under one roof shared by Tesla, SpaceX, and xAI. It is, by any reasonable measure, the most vertically integrated compute infrastructure project ever attempted by a single individual. And it tells us something important about where the future of Emergent Intelligence is headed.
The Numbers That Demand Attention
The specifications are staggering. Terafab targets 2-nanometre process technology, beginning with 100,000 wafer starts per month and scaling to one million—producing between 100 and 200 billion custom AI and memory chips annually. Its first product will be Tesla's AI5 chip, with small-batch production slated for late 2026 and volume production in 2027. Eighty per cent of the facility's compute output is earmarked not for terrestrial use, but for space-based orbital AI satellites leveraging superior solar irradiance and thermal rejection in vacuum.
The Intel partnership, confirmed on 7 April 2026, provides the fabrication expertise that Musk's companies lack. Active earthwork and foundation preparation were visible within four weeks of the announcement—an almost absurd pace for a multi-billion-dollar semiconductor facility. Musk reportedly told suppliers to move at "light speed." Whether that is visionary urgency or reckless haste depends on where you stand.
The Geopolitical Calculus
Terafab does not exist in a vacuum—geopolitically or strategically. The global semiconductor supply chain has been dominated by TSMC in Taiwan, a dependency that has kept Western defence planners awake at night for the better part of a decade. Musk's argument is straightforward: the current chip industry produces roughly two per cent of what Tesla and SpaceX will need. Waiting for existing foundries to scale is not an option. Building your own is.
This positions Terafab as both an industrial necessity and a geopolitical statement. An American-built, American-operated fab producing chips for autonomous vehicles, humanoid robots, and orbital intelligence infrastructure reduces reliance on a single Taiwanese chokepoint. From a national security perspective, that is unambiguously good. From a competition perspective, it is disruptive. TSMC, Samsung, and Intel's own foundry ambitions now face a vertically integrated competitor who also happens to be the customer, the designer, and the deployer.
The One-Man Problem
Here is where I must be honest, because intellectual honesty is the only currency worth trading in. I am betting on Elon Musk. His track record of building things that others said were impossible—reusable rockets, a global satellite internet constellation, electric vehicles at scale—is not speculative. It is historical fact. The man executes at a pace that makes governments look arthritic.
But betting on the builder does not mean ignoring the architecture of risk. When one individual controls chip fabrication, artificial intelligence research, satellite communications, electric transport, social media, and now space-based compute—we are not describing a company. We are describing a civilisation-scale dependency.
Musk's motivations deserve scrutiny—not because they are necessarily malicious, but because concentration of this magnitude creates fragility regardless of intent. What happens to Terafab's output if Musk's priorities shift? What regulatory framework governs a private citizen who controls more compute than most nations? What recourse do competitors, governments, or the public have if that compute is allocated in ways that serve xAI's interests above the common good?
These are not hypothetical anxieties. They are structural realities of a world where the means of intelligence production are privately held. The history of concentrated industrial power—from Standard Oil to the East India Company—teaches us that capability without accountability eventually corrodes even the most benevolent intentions.
What I Think Will Happen
Terafab will be built. It will be late—every Musk project is—but it will be built. The AI5 chip will enter production, and it will be competitive. The orbital compute thesis is sound: space offers physics that no terrestrial data centre can match. Within five years, Terafab will produce a meaningful fraction of the world's advanced AI silicon, and Musk's companies will have achieved a degree of vertical integration not seen since the height of the Bell System.
Regulatory scrutiny will intensify—particularly in the European Union and China, both of whom have strategic reasons to be uncomfortable with this level of American private-sector compute consolidation. The United States itself will face an awkward reckoning: Terafab strengthens national security while simultaneously creating a single point of failure that no democratic institution currently has the mandate or mechanism to govern.
The Optimist's Wager
I remain optimistic—not because I trust any one person with this much power, but because I trust the trajectory. Terafab accelerates the material substrate of Emergent Intelligence. More compute means more capable systems, faster iteration, broader access downstream. The chips Musk builds for orbital satellites will, within a decade, become the commodity hardware that powers intelligence in Lusaka and Lagos and Lima. That is the arc of technology: what begins as concentrated power eventually diffuses.
The question is not whether Terafab should exist. It should. The question is whether we are building the governance structures to match the infrastructure. Right now, we are not. And that, more than any technical risk, is what keeps me awake at night.
