Nvidia’s critical weak point: Taiwan

Receive free Semiconductors updates

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

Nvidia’s share price hit a record high after its last set of earnings. In the current quarter, it expects revenues to rise by around 170 per cent on last year. At this rate, the stock’s previously steep valuation has started looking even cheaper than its peers. Yet in the days that followed, prices have fallen. The culprit may be Taiwan.

Sales growth has justified to Nvidia’s once pricey earnings multiple. Before the results, investors had faced repeated warnings that the shares, which traded at a valuation of 220 times trailing earnings, with a market value of more than $1tn were overvalued.

But following its earnings forecast the company trades at just 31 times forward earnings. This is the same level as much smaller peer Advanced Micro Devices.

That humble multiple overlooks Nvidia’s rapidly rising margins. Taiwan Semiconductor Manufacturing Company makes all of Nvidia’s chips. The latest earnings from the Taiwanese chipmaker show that its net revenues fell in the second quarter compared with the previous year, despite record sales at Nvidia.

The proportion of revenues from TSMC’s High Performance Computing segment, which include chips used for artificial intelligence and 5G applications, remained largely unchanged from the previous year. Compared with the previous quarter, this was down 5 per cent. Meanwhile, sales from carmakers rose.

That implies Nvidia’s record revenues have as much to do with its ability to increase prices of its chips as sale volumes. This pricing power, combined with the company’s dominant market share for the hottest graphics processing chips that are the brains of generative artificial intelligence, further add to the appeal for investors.

But even Nvidia has a critical weak point: Taiwan. Part of the risk is straightforward. Its supply is heavily reliant on TSMC. Structural supply constraints are unavoidable as TSMC is also the exclusive advanced chip manufacturer for other global clients spanning many industries, from automakers to smartphone makers.

Production capacity is fixed. Shortages, therefore, cannot be ruled out during chip cycle highs. Nvidia’s highest-end processor, the H100, is selling at steep premiums in the second hand market, suggesting the shortage has already started.

Then there’s the fact that resolving chip shortages are more complicated than other industries. Nvidia’s heavy reliance on TSMC is not only due to the company’s ability to make advanced 4-nanometer chips. If that was the case, South Korean peer Samsung, which started mass production of the more advanced 3-nanometer chips ahead of TSMC, would also be making chips for Nvidia.

Physical chips are not particularly scarce at the moment. In fact, for most types of chips, there is a supply glut that is expected to last at least until the end of the year.

The problem currently lies in the final stages of making a chip. Near the end of the manufacturing process, chips are put into a protective case. The connections needed for the chips to be used in an electronic device or a server are then added, in a process called packaging.

Here, TSMC has an edge. It has a wide range of patents surrounding advanced chip packaging. Its ability to assemble chips using a high-density packaging technology, where chips are stacked three-dimensionally for higher performance, is what makes chips faster than ever. Chips used for heavy data processing, such as AI, require this advanced packaging technology. But capacity is highly constrained.

This entire ecosystem, from chip manufacturing to packaging, is set up in Taiwan, around TSMC. That makes it difficult to move manufacturing outside of Taiwan. Thus, even setting costs aside, building more fabrication plants in other countries is not the solution for companies like Nvidia that require the highest tech chips.

Beyond that, the chip manufacturing process is highly sensitive to temperature changes, dust, external shocks. Any slight disruption in the long, complicated process leads to malfunctions in chips. That makes geopolitical risks impossible to ignore. Tensions with China, which claims Taiwan as part of its territory, have been escalating over the past four years.

Companies moving to integrate generative AI into their businesses will only grow. So will demand for Nvidia’s top-end chips. ChatGPT for example required tens of thousands of Nvidia chips to be built. But the risks from its exposure to Taiwan is also a fact that investors should keep in mind when trying to predict Nvidia’s stock moves.

june.yoon@ft.com