Why One Company's Chips Became a Target for Spies

The situation

This week Nvidia opened an internal investigation into a personal relationship between its CEO, Jensen Huang, and a businesswoman of Chinese origin whom US intelligence accuses of spying for Beijing. The company says it has no reason to believe secrets were compromised. The facts are still contested and much is unknown. But the structure of the story is not an accident. To understand why a foreign government would reportedly spend years cultivating access to one company's chief executive, you need to understand why that one company matters so much. The answer is a mental model about power: the chokepoint.

The model: a chokepoint

A chokepoint is a single, hard-to-replace node that a huge amount of activity must pass through. Control it, and you control everything downstream. The term comes from military geography — the Strait of Hormuz, the Panama Canal — narrow passages where a small force can block an enormous flow. Economists and security scholars (notably Henry Farrell and Abraham Newman, who coined "weaponized interdependence") extended it to the modern economy: in a connected world, whoever sits at the chokepoint doesn't just make money, they hold leverage over everyone who depends on them.

Nvidia is the chokepoint of artificial intelligence.

The mechanism: how one chip became a strait

Follow the chain.

AI models are trained by running vast numbers of simple calculations at once. A standard processor (a CPU, the chip in your phone or laptop) is built to do a few complex things in sequence. A GPU — a graphics processor, originally designed to render video games — does the opposite: many simple calculations in parallel. That is exactly what training an AI model requires. Nvidia built the GPU and, crucially, also built the software (called CUDA) that programs it. Rivals can copy the hardware; they have not matched the software and the ecosystem built around it.

The result: Nvidia controls an estimated 80–90% of the market for AI accelerators. OpenAI, Google, Anthropic, Meta, Amazon — every serious player — must buy from Nvidia. A single top chip runs around $60,000. The company is worth over $4 trillion.

Now layer on geography. Nvidia designs in California but manufactures in Taiwan, through TSMC. So the chokepoint runs through one company, in one small country, that Beijing claims as its own. And in 2022 Washington restricted sales of the most advanced chips to China — turning a commercial bottleneck into a national-security weapon.

That is the setup. When a country is locked out of the one resource it needs to compete, it has three options: build its own (slow), buy through the front door (blocked), or acquire access through the side — smuggling, influence, intelligence. The espionage allegations, whatever their truth, are the predictable shape of what happens around a chokepoint, not a bizarre exception to it.

The other view

Be careful separating structure from story. That chokepoints invite coercion and espionage is well-established. That this specific relationship was an intelligence operation is an accusation, not a proven fact — Nvidia disputes that secrets were exposed, and the key figures deny wrongdoing. It is also possible to over-read the geopolitics: some of what looks like statecraft may be ordinary smuggling for profit, and some of Huang's lobbying to sell into China is simply a company defending its largest potential market. Both can be true at once. The chokepoint model explains why the pressure exists; it does not confirm any single person's guilt.

Where else this pattern lives

Chokepoints are everywhere once you see the shape:

• Energy. For decades, control of oil transit routes shaped wars and alliances. Russia's leverage over European gas was the same model, applied through pipelines.
• Software. A startup that builds its whole product on one cloud provider's tools has handed that provider a chokepoint — and will discover it the day prices rise.
• Your own career. If you are the only person at your company who understands a critical system, you are a chokepoint. That brings leverage — and also makes you the node everyone watches, pressures, and eventually tries to route around.

The lesson repeats at every scale: whoever controls the narrow passage holds power far beyond their size, and that power attracts exactly the kind of attention — lobbying, coercion, intrigue — now swirling around Nvidia.

What to watch

Two signals will tell you which story is dominant. First, policy: does the US tighten chip export controls again, or keep carving out deals (as it did letting Nvidia sell a limited H20 chip to China in exchange for a revenue cut and rare-earth access)? Every carve-out shows the chokepoint being traded as a bargaining chip, not just guarded. Second, the technology: watch whether Chinese firms or Western rivals (AMD, Google's own chips, startups) start closing the gap. The day a credible alternative exists, the chokepoint widens — and the intrigue drains away with it.

Expectation: ~70% that US export controls on advanced AI chips remain in force or tighten over the next two years, because the strategic logic of holding a chokepoint almost always outlasts the commercial pressure to sell. The spying allegations may fade; the structure that produced them will not.

Distilled from El País (English)