Nvidia’s Real Risk Sits in the Boardroom, Not in the Courtroom

Nvidia's Real Risk Sits in the Boardroom, Not in the Courtroom

The market is once again staring at the wrong screen. When Nvidia drops eleven percent in a session and the chief executive's name trends for reasons that have nothing to do with silicon, the commentary writes itself: the bubble, the litigation, the private life. All of it is noise. The two things that actually matter are duller and far more dangerous — a governance structure creaking under the weight of its own success, and a software moat that China is quietly draining one generation of chips at a time.

Start with the boardroom. Nvidia has become the centre of gravity of the entire artificial-intelligence economy — in October 2025 it became the first company in history to reach a $5 trillion market value — and it is starting to look like a financial ecosystem unto itself: circular finance, a myriad of investments in small companies, guarantees extended here and there, and those same small companies coming back as customers. Add employees in open revolt and a palace coup inside the board of directors, and you have a set of headlines that, for any normal company, would be a curiosity. For the company whose chips effectively price the future of the world economy, they are a systemic event.

The point is not that Nvidia is a fraud. The point is that when a firm's revenue is partly financed by its own balance sheet — in November 2025 Nvidia announced it would put up to $10 billion into Anthropic, a customer — the market loses its independent yardstick. This is the oldest problem in finance dressed in new clothes: an asset must generate future cash flows to be worth something; without them there is only a price, never a value. A share price held up by circular flows of capital tells you about the flows, not about the value. And when internal governance wobbles at the same moment — a restive workforce, a board at war with itself — the correction, whenever it comes, arrives without a shock absorber.

Then there is the moat. Everyone treats CUDA as permanent, like a language nobody can stop speaking. It is not. The most interesting development is that AI models running on CPUs and GPUs are already producing partial alternatives to CUDA. Partial today. But in software, partial has a habit of becoming sufficient. And the Chinese are not waiting for anyone's permission. While Washington debated export controls, Beijing kept building 7, 5 and 3 nanometre-equivalent capability. The sequence was almost comic: America announces it will not sell advanced chips; two years later China answers that it will ban the imports itself, because it now makes them at home. A ban designed to slow a competitor became the forcing function for its independence.

This is what I have argued for years about Chinese planning. It behaves like a very large company with a board that writes plans with clear objectives and deadlines, and then meets them. You can dislike the system; you cannot ignore the delivery. Every node they reach narrows Nvidia's addressable market from the top — China still accounted for $17.1 billion of Nvidia's revenue in fiscal 2025, and its data-centre chips now barely sell there — and every alternative compiler narrows it from below.

Meanwhile, the press devotes itself to the chief executive's love life. A strange country, this one, that can seize a head of state, kill another nation's religious leader, foment wars for decades, decide European foreign policy and sustain genocides — and then discover that two lovers put the greatest technological revolution of the century at risk. And the revelation that one of them was an employee? At the top of a company that size, everyone is an employee, a supplier or a client. If that is the scandal, the scandal is arithmetic.

None of this is an argument about the next quarter. It is an argument about where to look. If you want to know whether Nvidia deserves its place at the centre of the world, skip the court filings and skip the gossip columns. Read the board minutes and read the compiler stack. Watch whether the guarantees and the small-company investments turn into genuine external demand, and watch how fast the "partial" alternatives to CUDA become the default in the places where most of the world's AI will actually be built.

And if you want to read those signals yourself, there is a lesson I keep repeating: study engineering, mathematics, physics. Not finance, not business, not whatever is fashionable this year. Thirty-five years after my own degree in electronics engineering, it is still what lets me tell the difference between a share price and a technology.

The litigation cycle will pass. The gossip will pass. The governance and the moat will not.