Why Nvidia has a concentration problem

In 1990, economists won a Nobel Prize for identifying the risk of having an overly correlated portfolio. However, in 2024 the stock market put that risk aside as it became historically concentrated amongst a handful of companies, and one technological bet in particular.

It is the importance of covariance between assets that underpins the capital asset pricing model (CAPM), invented by Jack Treynor and William Sharpe, which won a Nobel Prize in 1990. Before CAPM, the rule of thumb was to have a large portfolio, but it wasn’t until Treynor that investors started caring about how much an individual stock adds to the variance of a portfolio. As economist Alex Tabarrok explains, CAPM was the point when “we get, really for the very first time, a scientific understanding of what risk is in the financial markets”.

This logic doesn’t just apply to asset portfolios. Covariance risk is also a concern for individual companies’ revenue streams. Analysts are often keen to point out if a small business has just a handful of customers. Likewise, management teams are eager to boast when customer concentration reduces. In its last results, Aim-listed software company Alfa Financial said it had “significantly reduced” its customer concentration, pointing to the fact that its top five customers represented 34 per cent of its revenue down from 61 per cent in 2019.

However, another Aim company, Zoo Digital, hasn’t been so lucky. Zoo provides translation and dubbing services for streaming companies. During the pandemic, it boomed and in the two years to March 2023, with Netflix spending billions on content, Zoo Digital’s revenue more than doubled to £75mn. The unforeseen problem was concentration risk. In the summer of 2023, the actors and writer’s guilds in Hollywood went on strike. This was only a small problem for Netflix because of its massive back catalogue and millions of customers, but for Zoo Digital it was crippling. In the year to March 2024, Zoo Digital’s annual revenue halved to £35m. And its share price is now 80 per cent lower than its 2023 peak.

The benefit of huge companies is this concentration risk usually doesn’t exist. Meta and Alphabet have billions of users and millions of advertisers, while Amazon has hundreds of millions of users and thousands of merchants. All these businesses are also dispersed geographically.

One of the exceptions amongst megacaps is Nvidia. The AI semiconductor company is the second largest company in the world with a market cap of $3.3tn. Yet, 50 per cent of its revenue comes from four customers: fellow megacaps Microsoft, Amazon, Meta and Alphabet.

So far, the market hasn’t worried as, since ChatGPT was released just over two years ago, Nvidia’s revenue has increased 380 per cent, while operating profit is up more than 1,000 per cent. However, the risk for Nvidia has always been its four largest customers stop spending as much or look for alternative suppliers for AI chips.

It looks like the latter has come first. Two weeks ago, semiconductor designer Broadcom said it was in negotiations with new hyperscale clients, thought to be OpenAI and Apple, around the development of custom AI chips. Broadcom’s Application-Specific Integrated Circuits (ASCIS) are designed to perform specific tasks. In partnership with Broadcom, Google has already developed the Tensor Processing Unit (TPU) which it deploys alongside Nvidia’s graphics processing units (GPUs) in its data centres. The benefit of the TPU is it is made for specific tasks, making it more cost-effective than the GPU. Now, OpenAI and Apple are looking to do something similar.

Off the back of Broadcom’s bullish earnings call, its share price increased 25 per cent. In the same week, Nvidia’s fell 10 per cent. Usually, the market doesn’t see the world as zero-sum. More AI spending should be good for all AI stocks, but Nvidia has benefitted from a near-monopoly, and the market now feels Broadcom’s gain means Nvidia’s loss.

Demand for AI chips will continue to outstrip supply in the short term. In an interview with the Financial Times, Broadcom’s chief executive Hock Tan said companies wouldn't stop spending until “they run out of money or when shareholders put a stop to this”. There is going to be plenty of cash to go around for both Broadcom and Nvidia. 

What the market has seemingly woken up to, is that with such a small number of customers, Nvidia isn’t guaranteed a monopoly. When all your customers are dispersed they cannot act collectively to negotiate down the price. When your already very few customers are also multi-trillion dollar companies, they don’t need to act collectively to cause damage.

Reducing correlation isn’t just a rule for investing; it’s a safeguard against overexposure to any single risk. Nvidia has put all its bets on AI, and for now, it is paying off. But increasing covariance lurks, just as it can in any aspect of life. Get married, but not to a gold digger. Be friends with your colleagues, but not at the expense of those from other parts of life. And please form a band, just absolutely not with a sibling. It may be great for a while, but it increases the risk everything comes crashing down at once.