NVDA: The Best Business in the World at the Wrong Price

There is no serious argument against NVIDIA as a business. The numbers are extraordinary — $215.9 billion in revenue for fiscal year 2026, up 65% year over year, with data center alone generating $194 billion. Blackwell is sold out through mid-2026. The demand backlog is measured in trillions. By every operational metric, NVIDIA is executing at a level that almost no company in history has sustained at this scale.

The moat is real too, and it runs deeper than the hardware. CUDA — NVIDIA's parallel computing platform — has been built and refined over two decades. It runs on more than 100 million computers, is used by over 40,000 organizations, and has 4 million registered developers. You don't replicate that in a product cycle. When people talk about NVIDIA's competitive advantage, they often point to the chips. The chips are good. But CUDA is the actual moat — the switching costs are enormous, and the ecosystem compounds with every new developer who builds on it.

So the business case is easy. The stock case is harder.

The China problem is permanent, not cyclical

NVIDIA took a $4.5 billion charge in Q1 FY2026 for H20 inventory and purchase obligations after export controls eliminated demand for its China-specific chip. It then guided Q2 with an $8 billion quarterly revenue hole from those same controls. That's not a risk factor — that's a structural write-down of addressable market. China was a meaningful part of NVIDIA's data center business, and that revenue doesn't come back under current policy. The guidance absorbs it going forward, which means the growth model from here has a permanently lower ceiling than the one investors priced in 18 months ago. The Blackwell ramp in domestic markets compensates significantly, but the math changed and the stock hasn't fully repriced it.

The people who know the most are selling the most

On March 20, NVIDIA insiders executed 25 stock sale transactions in a single day, totaling $28.6 million. The largest individual sale was $17.3 million. Two executives sold at 4x and 7x their normal transaction size. This is not routine diversification — executives selling at multiples of their normal volume on the same day is a directional signal, not noise. Institutional investors are doing the same thing. Moore Capital, Spence Asset Management, and others have been trimming positions. The people with the best information about what's actually happening inside the business are reducing their exposure at current prices. That deserves weight.

The technical picture is maxed out

NVDA's 14-day RSI is sitting at 98.63. That is not a rounding error. A reading above 70 is considered overbought — 98.63 means the stock has been moving up with almost no corrective pressure for an extended stretch. Stocks at this technical extreme don't necessarily crash, but they do mean there is almost no room for the thesis to stretch further on price alone. Any disappointment — a guidance miss, a policy update on China, an AMD win announcement — hits a stock with no technical cushion. The downside is asymmetric to the upside at these levels.

Competition is no longer theoretical

For years, AMD's challenge to NVIDIA in data center AI was a story about potential. That changed. AMD's MI400 series just secured a $60 billion deployment deal with Meta. That is not a pilot program. That is a hyperscaler making a serious bet on an alternative architecture. NVIDIA still holds 90%+ of the market, CUDA still dominates, and one deal doesn't rewrite the landscape overnight — but the era of NVIDIA having no credible competition in this segment is ending. The moat is durable. It is not impenetrable.

The call

NVIDIA will almost certainly be a larger business in three years than it is today. The Blackwell supercycle is real, Rubin follows it, and AI infrastructure spending is not slowing. None of that is in question. The question is whether the current price reflects all of that, and the answer is yes — and then some. You are paying full price for a great business with no margin of safety, at a technical extreme, while the people closest to the business are reducing their positions at scale.

This is a hold-and-watch, not an entry. The setup for a clean addition requires one of two things: a meaningful pullback to a defined support level that resets the technical picture, or concrete robotaxi and Optimus revenue numbers that validate the next leg of the thesis and justify the premium. Neither exists today. Until one does, size stays flat.

What the data showed

I ran this analysis alongside DVRG, an AI-powered stock analysis platform that scores tickers across divergence signals — comparing sentiment, dark pool activity, institutional flow, insider behavior, and earnings revisions to detect when the market narrative is running ahead of or behind what sophisticated capital is actually doing.

The DVRG output on NVDA was consistent with everything above. Quality base scored at 100% — the business is exceptional, no debate. But the valuation model showed the stock trading 14.3% above intrinsic value with zero opportunity premium. The divergence score came in at 2.6 out of 10, placing it firmly in "No Timing Edge" territory. The divergence type flagged was Sentiment > Dark Pool — meaning the public narrative is bullish, but institutional block trades executed off-exchange are not confirming accumulation. The institutional flow reading was Weak. The system's final allocation output was 60% of maximum position size, further reduced by a 0.85x intensity modifier for the weak divergence phase.

In plain terms: great business, wrong time. The structured data and the qualitative analysis landed in the same place independently.