With AI, Chips Could Be the Next Trillion-Dollar Industry. Here’s Why.
AI promises to unleash the true potential of digitalisation, paving the way for the chip industry’s biggest wave of growth yet.
Since 2015, the market value of semiconductor companies has surged 4x, hitting USD4 trillion in 2021. Demand for chips has grown hand-in-hand with the boom in Internet services, video gaming and streaming, a trend that accelerated during the pandemic.
While personal computers and smartphones defined the last two waves of digitalisation, technologists believe the world is now entering a new “age of computing,” driven by innovations in artificial intelligence (AI). As machines become increasingly intelligent, chips will gain powerful new drivers of demand.
Training the Child Machine
In 1950, Alan Turing, a pioneer in the field of computer science, posed a challenge to scientists who were trying to build artificially intelligent machines.
“Instead of trying to produce a program to simulate the adult mind,” Turing wrote, “why not rather try to produce one which simulates the child’s?”
Today, this program is known as the Child Machine, and has been key to the development of artificial intelligence (AI).
Like a child’s brain, an AI program often starts with a blank slate. It then soaks up vast amounts of information from the world, learning from the data over time. In AI circles, this process is referred to as “training” or “deep learning.”
Data is necessary for AI to learn. In general, the more data, the better.
For much of history, humans generated most of all data, be it in the form of pictures, words, videos, voice messages or social media posts. But in 2018, machines generated more data than humans for the first time ever. This is considered a turning point in the development of AI, and computing in general.
By 2025, researchers at Applied Materials (NASDAQ:AMAT) believe machines like industrial appliances, vehicles, medical devices, sensors, and security cameras will generate 99% of the world’s data. Between 2018 and 2025, this will drive a 100x increase in the amount of data produced every year.
This is important because the explosion of data could help AI programs — such as the ones powering self-driving cars — learn faster than ever.
With enough training, AI programs can think — — taking what they have learned and putting it into practice in the real world. This is what AI gurus call “inference.” Inference is the technology Alphabet (NASDAQ:GOOGL) uses to spot scam emails, and what Netflix (NASDAQ:NFLX) uses to recommend movies to you.
According to ARK Investment, a fund manager, in 2020 AI powered nearly all major Internet services including search, social media, and video recommendations.
Indeed, AI is hard at work at the world’s biggest companies.
At ByteDance (the maker of TikTok), AI is used to keep users engaged and drive tens of billions of dollars in advertising spend. At Apple (NASDAQ:AAPL), AI powers Siri and a host of other functions, such as facial recognition. When former U.S. President Donald Trump caught COVID-19, he took an experimental drug made by Regeneron (NASDAQ:REGN), a biotech that uses AI to discover new treatments.
“AI is going to eat software”
The big story here is that AI is shaping up be an important part of many, if not all, of this century’s greatest technological achievements. This includes virtual (VR) and augmented reality (AR), driverless vehicles, biotech, and robotics. Some of the latest AI programs, such as OpenAI’s Generative Pre-Trained Transformer 3 or GPT-3, have started writing software code.
Alphabet and Nvidia (NASDAQ:NVDA) describe AI as the next era of computing.
“When people think about the Internet of Things (IoT), that’s an AI problem. When people think about self-driving cars, that’s an AI problem,” Jensen Huang, Nvidia’s chief executive, said in a 2017 interview with the Financial Times. “Just as software ate the world, AI is going to eat software.”
Ray Kurzweil, Alphabet’s Director of Engineering, has made even bolder predictions: By 2029, AI will reach human levels of intelligence; by 2045, machines will be smarter than humans. If Kurzweil is correct, AI could bring about a technological revolution bigger than all previous ones.
“AI is the most powerful technology force of our time. Learning from data, AI supercomputers can write software no human can…. Someday, trillions of computers running AI will create a new internet — the internet-of-things — thousands of times bigger than today’s internet-of-people.”
- Jensen Huang, CEO, Nvidia, Sept. 2020
AI is fueling a virtuous cycle of more data and more chips
A computer chip is like an organ. Some chips are like parts of the human brain, performing calculations at super fast speeds, or memorising information. Other chips are like ears and eyes, sensing the world around them, while some chips are used to “talk” — — communicating via airwaves. Combine several chips, and you get a ‘smart device.’
For machines to grow more intelligent, chips of all kinds will be needed to extract data from the world, store it, then learn and think. The cost of chips in driverless cars, for instance, is expected to be four times as much as those in today’s gasoline-powered cars.
Certain types of chips, for instance, are used to run and power the sensors that machines use to sense the world and generate data. Memory chips are needed to store and accumulate the vast amounts of data used to train AI programs. Working with in tandem with memory chips, logic chips — such as GPUs and CPUs — process information, helping AI machines “think.”
Networking chips (think 5G) help power communication, connecting data centers to consumers. They also link AI-enabled devices — — such as sensors — — with each other, creating what is known as the ‘Internet of Things (IoT).’ Finally, chips are needed for computers at data centers to make sense of all the data and create even “smarter” programs and machines.
All these types of chips already exist, but most have been designed for computers, servers, and smartphones. New types of chips, such as NPUs (neural processing units) and FPGAs (field programmable gate arrays) are being designed just to tackle AI applications, which are among the most demanding of computing tasks.
According to researchers at Georgetown University, these chips — also known as AI accelerators — are tens or even thousands times faster at training and inference, compared to CPUs. By 2025, AI-related chips could make up 20% of all semiconductor demand, up from 7% in 2017, according to estimates by McKinsey, a consultancy.
Another trend that has sped up the use of AI is the widespread availability of high-performance computing (HPC), clusters of super-powerful computers which can be leased from cloud service providers. Instead of having to buy, set up and run their own HPC systems, companies and scientists can now rent them from a cloud service provider like Amazon (NASDAQ:AMZN) or Alphabet. This gives them access to HPC systems using AI accelerators, which speeds up the process of training and inference.
“With the increasing need for computation, HPC will be the strongest driver of TSMC’s long-term growth,” TSMC (NYSE:TSM, TWSE: 2330) chief executive C.C. Wei said on Jan. 13, speaking on an earnings call. “Expect it to be the largest contributor in terms of our incremental revenue growth, with CPU, GPU and AI accelerators the main growth area for our HPC platform.”
Powering up the next trillion-dollar industry
Given the growing importance of AI, it should come as no surprise that many of the world’s biggest companies including Alphabet, Microsoft (NASDAQ: MSFT), ByteDance and Tesla (NASDAQ: TSLA) are all working on AI chips of their own. In recent years, tens of AI chip startups like Groq, Tenstorrent and Cerebras have emerged to challenge established names like AMD (NASDAQ:AMD) and Nvidia. Outside the tech industry, most people are not aware of these companies, but many of them have already secured billion-dollar valuations.
By 2030, the ‘AI Era’ will see chip industry revenue rising from USD500 billion a year to USD1 trillion (roughly SGD1.35 trillion) a year, according to forecasts by Applied Materials. Other major chip companies, including Lam Research (NASDAQ:LCRX), GlobalFoundries (NASDAQ:GFS) and TSMC, also project annual semiconductor revenues closing in on a trillion dollars by 2030.
To put things into perspective, it took more than 50 years for the chip industry to reach nearly USD500 billion in annual revenue.
In the age of AI, it will take only 10 years for it to grow by another USD500 billion.
- How TSMC cornered nearly the entire market for advanced chipmaking
