The AI Revolution's Fragile Foundation

The AI revolution, we're told, is unstoppable. It will reshape every industry, redefine knowledge work, and represent the most consequential technological shift in human history. Except, apparently, it will not particularly bother the people whose livelihoods it is reshaping. This is the message that emerged from two high-profile statements in the past 24 hours — and the gap between them reveals something the industry would prefer not to name.

On 27 May 2026, Nvidia CEO Jensen Huang called Taiwan the "epicentre" of AI manufacturing, praising the island's dominance in semiconductor production. Hours later, OpenAI CEO Sam Altman insisted that AI was unlikely to cause a "jobs apocalypse," noting that white-collar job losses had been far less than he once feared. Both men were speaking publicly, both were broadly reassuring, and both were describing a system whose structural vulnerabilities they have every reason to minimise.

The question worth pressing is whether the people who tell us AI is inevitable actually believe it — and whether their own infrastructure suggests they do not.

The Concentration Problem No One Wants to Name

Taiwan produces approximately 90 percent of the world's most advanced semiconductors, almost entirely through a single contract manufacturer, TSMC. Nvidia's H100 and B200 GPU lines — the hardware that underpins the current wave of AI infrastructure build-out — are manufactured in Taiwan. So are the chips inside every AI accelerator from every American firm currently competing for data-centre contracts. Huang's framing of Taiwan as an indispensable cornerstone of AI is not incorrect. It is, however, incomplete in the way that serves his industry's interests.

Taiwan's geopolitical status is unresolved. Whether Taipei resolves that status through negotiation, pressure, or escalation is the single variable that could halt global AI development more quickly than any software capability could. The surrounding waters are contested. American policy has oscillated between strategic ambiguity and notional commitment in ways that do not inspire confidence in supply-chain continuity. Huang knows this. So does every other executive at every firm currently dependent on TSMC's foundries.

The industry has managed this reality by normalising it. Altman, meanwhile, offered a different reassurance — that AI's economic disruption had been "far less than feared." This framing addresses a different anxiety. It is designed to reduce public concern about automation, not about infrastructure. But the two anxieties are linked in ways the industry does not want to draw out. A supply chain concentrated on a contested island is fragile in ways that go beyond geopolitical theory. The semiconductor shortage of 2021 provided a preview: when a single point of production is disrupted, the effects cascade across every sector that depends on chips, including AI. The industry responded to that lesson by continuing to concentrate production further. Nothing structural has changed.

The Jobs Argument's Internal Contradiction

The "jobs apocalypse" framing Altman addressed deserves scrutiny independent of the supply-chain question, because his own industry's previous public messaging is in tension with his current position.

For several years, AI companies argued that their technology would be transformative — revolutionary, even — across every sector of the economy. That framing justified hundreds of billions in capital expenditure, drove stock valuations, and framed AI companies as the most important institutions of the age. The same industry then argued, when convenient, that AI's effects on employment were mild and that workers had little to fear.

This is a position that cannot hold on its own terms. You cannot simultaneously argue that a technology is the most consequential development in human history and also that its economic effects have been modest enough to avoid an "apocalypse." The framing Altman is now offering suggests either that the previous claims about AI's transformative potential were exaggerated, or that he is selectively emphasising the limited-distruption narrative because it serves interests that the transformative-potential narrative cannot.

The likely answer is the latter. When AI is being sold to investors, its potential to reshape industries is maximised. When AI is being sold to workers, its potential to displace them is minimised. Both framings have been deployed by the same executives. Neither, alone, is an honest account.

The Architecture of Manufactured Resilience

What the two statements share — Huang on Taiwan, Altman on jobs — is a structural function: they are designed to project inevitability and stability onto a system that possesses neither quality.

The tech industry has constructed a peculiar risk architecture. The people who profit most from AI development bear the least accountability for the concentration and fragility of that development's physical substrate. Executives receive performance bonuses calibrated to revenues that depend on a supply chain they do not own and cannot control, located on an island whose fate is determined by actors outside their influence. When analysts raise concerns about this concentration, the industry responds with reassurance: Taiwanese manufacturing is irreplaceable, but irreplaceability is presented as a feature rather than a vulnerability.

This is the structural frame. The companies building the AI revolution have optimised for efficiency and speed above all other variables, including resilience. They have been forced to confront the consequences of that optimisation — partly through the 2021 shortage, partly through ongoing geopolitical escalation — and their response has been to tell the public that everything is fine. That the island is an "epicentre." That the job losses are manageable. That the revolution is inevitable.

The public should assess those claims against the infrastructure they are built upon.

What Comes Next

The stakes here are not abstract. A disruption to Taiwanese semiconductor output — through diplomatic escalation, naval conflict, or any of a dozen lesser contingencies — would not merely slow the AI industry. It would halt it, at least in the near term and across all end-users simultaneously. Every auto manufacturer, medical-device firm, and defence contractor that depends on advanced chips is exposed to the same concentration. The industry has not diversified meaningfully; it has continued to bet on the same geography while publicly insisting the bet is sound.

The relevant question is not whether AI will disrupt jobs over the coming decade. It is whether the people currently assuring us of AI's inevitability and benign effects are offering genuine analysis or public-relations management. The evidence from their own infrastructure — from TSMC's foundry map and the geographic concentration of advanced manufacturing — suggests the latter. Taiwan is not an "epicentre" by choice; it is one by historical accident and industrial policy. The industry arrived at this concentration because it was efficient to do so, not because it was resilient. The reassurances from Huang and Altman serve their interests and their stock prices. Whether they serve anyone else is a question worth keeping open.