/Semiconductor%20Chip.jpg)
If artificial intelligence (AI) is to become part of everyday life, one thing is clear - we're going to need a lot more computing power, and therefore a lot more of the semiconductors that power that computing.
Each model that has been developed by OpenAI has cost 25 to 100 times more than the previous one. The current version, GPT-4, cost $100 million to develop, while GPT-5 (currently in development) will cost an estimated $2.5 billion.
Perhaps more critically, there are estimates that GPT-5 could use about 1.5% of the world’s current computing power - and that figure could spike to 50% for GPT-6. To train a GPT-7 model, the world could need 15 times as many computers as are currently in existence!
While these numbers may be exaggerated, it is obvious the world will need to produce a huge quantity of the graphic processing units (GPUs) needed to train AI models.
At the moment, there is only one company in the world that can make GPUs at the scale needed for AI. And while Nvidia (NVDA) is the world's leading designer of these chips, it outsources all of its production to one manufacturer, Taiwan Semiconductor (TSM).
Let’s take a closer look at how AI is of growing importance to the company.
TSM and AI
It should not come as a shock that AI chips are a very fast-growing part of Taiwan Semi's business. Currently, AI makes up just a single-digit portion of its overall sales - but this revenue source is growing at a rate of 50% annually, and JPMorgan expects overall AI-related revenues to account for 25% of total revenue by 2027.
As it stands now, TSM's largest business segment is high-performance computing (HPC), at 43% of revenue last year. This segment includes the GPUs needed for training AI, along with data center chips. The next biggest segment was smartphones, at 38% of revenue, followed by the Internet of Things at 8%, and automotive at 6%.
Outside of the AI boom, it's worth pointing out that 2023 was a bad year for the semiconductor industry, with demand slowing almost across the board. Given its product mix, Taiwan Semiconductor’s net revenue fell by 8.7% in 2023.
However, in the fourth quarter, the upturn was obvious. Smartphone revenue rose 27% year on year, while HPC revenues jumped by 17%, and automotive was up 13%. For the current quarter, the company is forecasting that HPC demand will push revenue to a range between $18 billion and $18.8 billion, which would represent a 10% year-on-year increase. I strongly suspect TSM will easily blow past this estimate.
TSM Ramps Up Capex
Taiwan Semiconductor is an extremely well-run company. It understands that in order to meet the upcoming surge in demand for AI chips, it will need to make huge capital expenditures.
It ramped up spending shortly after OpenAI’s ChatGPT was released in late 2022. In 2023, TSM spent $30.4 billion on capex, up 72% from 2020. Management has promised to spend between $28 billion and $32 billion in 2024.
The fact that the company is always talking to its customers and their needs means it is really good at allocating capital. Taiwan Semi does not spend money on capital investments if there isn’t demand out there for the latest chips. That’s why its return on capital figure has been consistently above 20% for the past decade.
Semiconductor manufacturing requires a huge amount of spending because the chips need to constantly get smaller and more powerful to meet customer demands. TSM knows this, and has maintained its technological lead. For example, it is leading the way now with 2nm (nanometer) technology.
Its leadership position has given Taiwan Semiconductor a virtual monopoly, which allows the company to charge high prices, as reflected in the 43% operating margin.
Buy TSM Stock
Revenue at TSM was up 9.4% in the first two months of 2024, with sales of $12.6 billion. That’s despite signs that Apple (AAPL), which accounted for 25% of its business last year, is struggling to grow sales for the iPhone, particularly in China.
In simple terms, Taiwan Semi is the enabler for almost all AI processing at the data center as well as edge computing.
I believe organic growth of AI, Internet of Things, and high-performance computing applications will go on for decades. After all, AI and HPC play a central role in quickly processing data inputs to solve complex problems, like autonomous driving and language processing, which accentuates the need for more energy-efficient and faster chips.
And Taiwan Semiconductor should consistently earn higher gross margins than its competitors, thanks to its economies of scale and premium pricing, justified by its cutting-edge technologies. The company is also poised to benefit both from more semiconductor firms embracing the fabless business model, and from technology giants designing their own data center chips.
At a price just below $150 per share, TSM trades at a forward price/earnings ratio of only 23.8 times. Nvidia trades at a forward p/e about 14 points higher. Taiwan Semi's top-line compound annual growth rate (CAGR) should come in at 13.5% over the next five years.
The next valuation re-rating catalyst for the company’s stock will be exponential growth of generative AI used in devices like smartphones and computers.
Against this backdrop, TSM stock is a buy anywhere under $166.
