Enthusiastic Nvidia investors may need a reality check

Receive free Nvidia updates

We’ll send you a myFT Daily Digest email rounding up the latest Nvidia news every morning.

Nvidia’s share price has more than tripled this year. Expectations of unprecedented demand for artificial intelligence chips have pushed the US chip designer to a market capitalisation of more than $1tn. 

So why have Asian chipmakers — the companies that make all of Nvidia’s chips — reported their lowest earnings in years?

The risks facing the global advanced chip industry should provide an important reality check for the investors who believe Nvidia’s record-beating rally can continue.

Nvidia’s advanced chips are essential for powering the new generation of AI-related tech, from ChatGPT to autonomous driving. These chips, or graphics processing units, are thousands of times faster than general purpose chips when applied in AI models and account for most of its sales.

Companies are boosting computing power by increasing the number of Nvidia chips they use in data centres. This trend is just starting, meaning there is ample room for growth. That explains why Nvidia is forecasting revenues of $11bn for the current quarter, up nearly two-thirds from a year earlier.

Making Nvidia’s high-end chips requires the most advanced chip manufacturing technology available. Together, Taiwan Semiconductor Manufacturing Company and Samsung make 100 per cent of the world’s supply of advanced chips, which use a manufacturing process of 5 nanometres or less.

The fortunes of TSMC, Samsung and Nvidia should therefore be inextricably intertwined. All of Nvidia’s AI chips are made by TSMC. Samsung makes chips that are crucial for operating servers needed for generative AI.

That suggests investors who buy Nvidia should be investing in TSMC and Samsung with equal enthusiasm. Relatively cheap valuations should add further appeal.

Yet their share price gains look paltry next to Nvidia’s. Warren Buffett has been one of the most high profile investors to dump shares in TSMC this year, exiting completely in the first quarter. And record AI chip demand has done little to offset earnings weakness at the world’s two largest advanced chipmakers. Samsung’s second-quarter earnings are down 96 per cent, the lowest in 14 years. TSMC’s net profit fell 23 per cent, its first quarterly decline in four years.

Part of the reason for this is that AI is still a small sector — analysts estimate that less than 5 per cent of global demand for chips comes from AI-related uses. At TSMC, nearly 90 per cent of sales come from consumer electronics, smartphones, PCs and cars.

Time lags make a difference too. Chip orders are placed well in advance, typically once a year. When chips were in short supply over the past two years, companies stockpiled them. Now, as they work through that inventory, orders have shrunk.

But a bigger reason that shareholders of Asian chipmakers do not share the same fervour as Nvidia investors is that they are all too familiar with the limitations of chip production. 

Unprecedented demand is much less exciting when output is capped. Chip production is capital intensive. Building a chip fabrication facility costs more than $20bn. Parts must be continually upgraded due to the rapid pace of chip technology change, and weak earnings mean there is less money for capacity expansion.

For now, Nvidia and TSMC’s gross margins are comparable at about 60 per cent. In the future, Nvidia will have the edge. Its software and intellectual property adds value and avoids the steep costs of chipmakers.

But on the flipside, this means that Nvidia’s growth is also constrained by capacity. AI chips will remain a small fraction of chipmakers’ sales for years. Nvidia must share chipmakers’ resources with key clients such as Apple. A new iPhone launch, for example, means orders for hundreds of millions of high-end chips.

The current chip oversupply has glossed over the risks ahead. Moreover, the gaming and crypto sectors, where Nvidia chips are just as critical, compete for their supply. Disruption risks from a chip engineer shortage and rising Taiwan-China military tensions should also be factored in.

As the complexity of machine-learning models used for tasks such as generative AI grows, increasing computing power requirements can only be met by more chips. Longer term, Nvidia is positioned to be the biggest beneficiary of the AI wave. Yet for now, investors must price in both the risk of supply constraints and temper expectations about how quickly the market can grow.

june.yoon@ft.com