The start of the Covid-19 pandemic opened the world’s eyes to the importance of supply chains, and the microchip shortage quickly became the most visible example of how bottlenecks can affect business. Three years later, the consequences of this squeeze continue to reverberate across the semiconductor industry, and suggest investors will have to think carefully about how to access cutting-edge technology opportunities.
For the past three decades, a globalised production model has created a series of semiconductor giants across the world. Their roles range from making the lithography machines used to imprint circuits onto silicon wafers (ASML (NL:ASML) in the Netherlands), using those machines (TSMC (TW:2330) in Taiwan), to designing the chips (AMD (US:AMD), Nvidia (US:NVDA) and Qualcomm (US:QCOM), all in the US).
This system, coupled with the ongoing boom in semiconductor usage, has been lucrative for investors: despite recent struggles, all these companies’ market caps have at least doubled in the past five years. However, as the pandemic highlighted, it has also created a supply chain littered with bottlenecks where knowledge is concentrated in specific countries or organisations.
Now, with east-west political tensions on the rise, governments around the world – most notably in the US but also those in continental Europe, Japan and the UK – are jostling to take more of a pie that is still anticipated to grow rapidly in the years ahead. Despite cyclical headwinds, the semiconductor sector is forecast by McKinsey to be worth $1tn (£0.8tn) by 2030 – up from $500bn in 2021.
The US has already shown its industrial policy can cripple overseas players. In 2019, having alleged the company was spying on US citizens, the Trump administration banned all high-tech US chips being sold to Huawei. The Chinese business was decimated as a result. No longer able to source design software, chip-etching tools or chips themselves from US suppliers, its revenue dropped 40 per cent between 2020 and 2021.
President Biden’s administration has clamped down further. In October last year, the US banned all advanced chips from being sold to Chinese companies, citing national security concerns over China's investments in supercomputers and artificial intelligence. Last month it was reported that the Netherlands and Japan had agreed to ban the export of chip manufacturing tools, following talks with US officials eager to ensure ASML, Nikon (JP:7731) and Tokyo Electron (JP:8035) followed its lead.
Incentives play an important role as well as sanctions and the US (as well as Europe, Japan and the UK – see below) is also trying to improve its domestic capabilities. The US has been the sector’s design leader since the integrated circuit was patented by Fairchild Semiconductor in 1961. However, its outsourcing of manufacturing has left it heavily reliant on Taiwan and to a lesser extent South Korea.
The US is uncomfortable with having 90 per cent of advanced chip manufacturing that close to China, particularly at a time when fears of a Chinese military intervention in Taiwan have been growing. Last summer, the US Congress passed the Chips and Science Act, which included $52bn of subsidies for companies building plants in the US. It is already having an effect: TSMC is building a $40bn facility in Arizona while Intel (US:INTC) intends to create plants in Arizona, Ohio, and New Mexico.
Near the bottom of the cycle
There is an added complication to this shift in priorities: in the short term, there are signs that the chip shortage has turned into a glut, one factor in this shift being that demand is slowing. Quarter-on-quarter revenue growth at TSMC slowed 1.5 per cent in its most recent results. “In 2023, our gross margin faces challenges from lower capacity utilisation due to semiconductor cyclicality, the ramp-up of entry, overseas fab [semiconductor fabrication plant] expansion and inflationary costs,” said chief financial officer Wendell Huang.
The secular demand for more technology creates structural growth trends for semiconductor businesses. However, in the short term, they are vulnerable to wider economic pressures. Currently, as far as the chip industry is concerned, the world is already in a recession.
In 2021, consumers were rushing to get their hands on consumer electronics while locked down in their homes. In response, consumer electronics companies stocked up their inventories in case of further disruptions. Then energy prices started to rise, and global economies slowed down. In the last quarter, Apple (US:APPL) saw year-on-year iPhone sales fall 8 per cent to $65.8bn. At Microsoft (US:MSFT), personal computing revenue fell 19 per cent.
The impact of weakening smartphone and personal computer demand can be seen across semiconductor manufacturers’ and producers’ results. Samsung (KR:005930), which makes the Dram memory chips for iPhones, saw operating profits fall 70 per cent in the fourth quarter. Jaejune Kim, vice-president of the company’s memory division, said that for PC and mobile “the decline in demand was steepening across the overall industry”.
AMD’s client segment revenue decreased 51 per cent year on year due to “reduced processor shipments resulting from a weak PC market and a significant inventory correction across the PC supply chain”. At Intel, client computing revenue was down 36 per cent. Across the industry, shipments of the x86 CPU used in PCs saw the largest decline in 30 years, according to Mercury Research.
Analysts are confident demand will return, but it is not obvious when, and supply chains complicate the issue. “When supply chain inventories return to healthier levels, the chip demand might start to grow from the second quarter of 2023 and more obviously from the third quarter, but this is also dependent on the uncertain global economy,” says TrendForce analyst Joanne Chiao.
Economic weakness coupled with rising interest rates caused a sell-off in the sector last year. This suggests the present moment is a better time to buy, but some think timing the market is harder than ever. “Investors used to buy just before earnings started getting better, but then everyone knew that, so started buying even earlier,” suggests Alastair Unwin, technology fund manager at Polar Capital (POLR).
Broker Stifel believes investors are already starting to look beyond the prospect of a recession. “The market is becoming more desensitised to negative estimate revisions, as focus begins to shift towards signs of recovery,” analyst Brian Chin said in relation to semiconductor manufacturing equipment supplier Applied Materials’ (US:AMAT) first-quarter results last week.
In any case, TSMC has weathered the downturn better than its rivals and its near-monopoly is not reflected in a consensus forward price/earnings (PE) ratio of just 16. There will be a geopolitical discount applied to the risk of Chinese intervention, but this should also be applied to its peers given the scale of global disruption that would cause, and the possible impact on other players’ Chinese operations (see chart).
The moat around TSMC remains sizeable. Its impressive gross margin of 53 per cent indicates pricing power, and management is confident in its ability to raise prices further because, in its words, “semiconductors have become more essential and more pervasive in people’s lives”. TSMC customers include leading chip designers Nvidia, Qualcomm and AMD.
Nvidia has a monopoly of its own in graphics processing units. Investors' Chronicle featured the company in our ideas section at the start of December, and its share price has since risen over 30 per cent, in part due to excitement over the role it plays in artificial intelligence infrastructure. Nvidia’s secular growth trends are strong, but its value is harder to justify at a PE ratio of over 50.
In terms of earnings, Samsung was hit harder than TSMC last year because of its greater exposure to smartphones, but it is confident that demand will return. In 2022, it invested $36.7bn in semiconductor capital expenditure, with $14.4bn coming in the fourth quarter. This is the same amount as in the previous year despite the big dip in profits. Its trailing 12-month PE ratio is just 7.6, so if demand can bounce back there looks to be good value in the Korean business.
Supercomputers and AI
Recent results would have been worse if the slowdown in consumer spending hadn't been offset by the continued growth in cloud computing, which is now crucial to the semiconductor sector’s prosperity. While cloud growth has also slowed, Google Cloud, Amazon Web Services and Microsoft Azure still saw growth rates in excess of 20 per cent in the fourth quarter of 2022.
At AMD, the fall in client segment revenue was offset by a 42 per cent rise in sales for its data centre segment, which helped drive group revenue up 16 per cent. Its rise is in contrast to Intel, whose failure in the data centre market means its market capitalisation has now dropped below AMD’s for the first time.
Intel invented the central processing unit (CPU) chip in 1971 but has faltered badly in the past two decades. First, it declined the opportunity to design chips for the iPhone. Steve Jobs turned to Cambridge-based Arm, then a small business, instead. Intel’s other big mistake was staying in the manufacturing business, where it has failed to match TSMC’s might.
AMD and Intel both specialise in CPUs. They are the control centre for computers; however, more specialised graphics processing units (GPUs) provide the computing power needed to train artificial intelligence models. The GPU was invented by Nvidia in 1999 and allows faster computing because it uses matrix calculations rather than linear calculations – known as parallel computing.
ChatGPT has brought AI to the world’s attention in the past few months, but this was only possible because of the investment going on behind the scenes. OpenAI’s ChatGPT was trained on a supercomputer built by Microsoft using Nvidia GPUs. Each individual GPU costs up to $250,000 and Microsoft’s computer contains around 10,000.
Nvidia also designs the cabling and software that connects GPUs with each other. In short, Nvidia covers the whole ecosystem needed to build the supercomputers that train AI models. The chief technology officer of a San Francisco-based cloud computing company, speaking on condition of anonymity, recently said Nvidia was “way ahead of all its rivals”.
Will subsidies work?
With decades of knowhow behind them, rebalancing the global semiconductor market will not be easy, even with the massive subsidies now being handed out. As technology advances, the knowledge gap risks growing wider rather than narrower. Transistors can now be switched on and off with the repositioning of a single atom, enabling more to be crammed onto each chip. But ever-better machinery is required to facilitate this progress. AMSL’s lithography printing machines, for example, cost up $200mn each.
And having the right machinery is not enough without skilled workers. “Making wafers is like cooking. You can have the best equipment in the world, but you need to be a good chef otherwise the meal will be awful,” says Tim Pullen, chief financial officer of UK wafer production company IQE (IQE).
McKinsey thinks the US is badly short of the necessary labour despite its expansion plans. “To complete the $400bn-worth of construction projects in advanced industries, including semiconductors, the US would need about 200,000 to 300,000 more skilled labourers such as electricians, mechanical workers, welders and pipe fitters,” it stated in a recent report.
TSMC has said it will transfer hundreds of new US workers to Taiwan to learn from its factory there.
Europe, equally concerned about its reliance on Asia, plans a €43bn (£38bn) Chips Act of its own. Intel has committed to building a 'mega-site' in Germany and the bloc is keen for TSMC to follow suit. But again there are concerns over a skilled worker shortage, and there are those who say the continent should focus on existing expertise such as chipmaking equipment like that provided by ASML, and equipment inputs such as the mirrors and lenses provided by Germany’s Carl Zeiss SMT.
In the UK, there is a similar fear that companies will never be able to catch up with their US and Asian competitors. But here, too, there are specialisms. The UK's most successful semiconductor company is chip designer Arm. Bought by Softbank (JP:9984) in 2016 for $32bn, it is set to be listed by the Japanese company this year. Successive UK prime ministers have been lobbying for a domestic listing to form part of the flotation process.
A US float, rather than a dual US-UK listing, remains most likely. The few listed semiconductor companies in the UK are much smaller than their US peers and have complained about a lack of support from the government compared with their US and EU counterparts. The industry continues to await publication of the UK government's semiconductor strategy, first scheduled for autumn last year.
It's highly unlikely the UK will ever catch up when it comes to silicon technology, but it was reported by Bloomberg last month that the UK strategy will involve using taxpayer money to ramp up the manufacturing of compound semiconductors.
IQE is the leading manufacturer of compound wafers outside of China. Its wafers are shipped to manufacturers and sliced into smaller pieces before the transistors are printed onto them. Compound wafers are more expensive than traditional silicon ones, but can function at a higher frequency. This makes them useful for next-generation technology, such as 5G, electric charging and 3D sensing. The UK is also home to semiconductor manufacturers focused on other chips, such as CML Microsystems (CML) – which secured a place in Simon Thompson's 2023 Bargain Shares portfolio.
IQE is based in Newport, but recently warned it would move more manufacturing to its US facilities if the UK didn’t provide further support. The government, however, is keen to promote the Newport 'cluster', and in November blocked the sale of Newport Wafer Fab to China’s Nexperia on national security grounds.
After five years of stagnant growth, IQE hopes demand for its compound wafers will be driven by the secular growth trends of 5G and electric vehicle charging. This year’s FactSet forecast revenue of $170mn is expected to be 4.4 per cent below 2020. Management has warned on demand for its wireless wafers as the handset market slows down, and we currently rate the company as a hold. However, it expects growth to speed up in 2024 as the economy bounces back.
A recent Investors’ Chronicle visit to IQE’s Newport facility underlined the potential for expansion: its floorspace features just 10 of its machines and much more empty space. Management hopes to add 90 extra machines, but first needs to establish demand for the product. New chief executive Americo Lemos, who previously worked at US chipmaker GlobalFoundries (US:GFS) and Qualcomm, spends most of his time in California negotiating with potential customers.
Some of these negotiations have been successful. IQE has signed six commercial deals since May. It doesn’t share client names, but house broker Numis believes a multiyear deal with Sony is amongst them. In a note published in January entitled ‘Buy more on any weakness’, Numis analyst John Karidis said the broker remained confident the sector will benefit greatly from “major developing tailwinds”.
That is also the story of the industry currently. Despite near-term fears of recession and medium-term concerns around de-globalisation, demand for semiconductors will continue to grow. Monopolies such as TSMC, Nvidia and ASML promise strong investment returns – but this strength increasingly means they are also political weapons that governments can and will turn on enemies.
