Singapore is poised to emerge as a critical supplier and processing node for semiconductors that can handle the vast computing and memory needs of running generative artificial intelligence (gen AI) applications on devices as compact as smartphones.
As the global market for gen AI products – encompassing both hardware and software – enters a widely expected boom over the next few years, chipmakers here in Singapore and elsewhere are looking to increase their production capacity and broaden their talent pool.
Many chipmakers are in the race, including Qualcomm, which has a test centre of excellence in Singapore that carries out design verification and failure analysis for the smartphone chipmaker’s new products and technologies.
But the company that gives Singapore the largest footprint in the gen AI chip market is Micron Technology – one of the world’s top producers of memory chips that process and store data.
Because of Micron, Singapore commands a 10 per cent share in the high-end global memory market, according to HSBC.
And demand for these memory chips is set to increase as exponentially as that of logic chips, which perform computing functions.
Mr Sumit Sadana, executive vice president and chief business officer at Micron, believes that 2024 will be the first year when personal computers armed with AI devices and applications will start getting introduced. Some latest smartphones with AI capabilities are already in the market.
Production of these devices will then ramp up quite heavily in 2025 and 2026, he said. But this period will eventually be seen as just the beginning of a much longer investment and innovation cycle.
“It is really in its early infancy of what will be a 10- and 20-year wave of investment that will drive AI into every corner of the economy and every part of the lives of consumers,” he told The Straits Times in an interview.
Mr Sadana, who joined Micron in 2017 and has more than 30 years of technology industry experience, believes that gen AI will change everything – how we work, communicate, entertain and interact, and how businesses boost productivity.
“This growth over the years will need to be supported by additional investments in manufacturing and investments in headcount,” he said.
Statista Market Insights projects the gen AI market to reach US$36 billion (S$49 billion) in 2024, up from US$20.5 billion in 2023 and US$5.5 billion in 2020. There will then follow a period of explosive growth, estimated at a compound annual rate of 46.5 per cent, and the market will become US$356 billion in size by 2030.
Over the next 10 years, Bloomberg Intelligence estimates that the market will grow to US$1.3 trillion.
The evolution of the gen AI market began with companies training the so-called large language models, or LLMs, on supercomputers.
LLMs are machine-learning programs that can recognise and generate text, as well as other forms of content such as images and videos.
LLMs are trained on large amounts of data, including nearly everything that has been written on the internet over a large span of time. The LLM segment of the gen AI market is so far almost entirely dominated by graphics processing units (GPUs) made by Nvidia.
For example, Nvidia GPUs form the backbone of the AI platform of the National Supercomputing Centre Singapore that manages the Republic’s first national petascale high-performance computing facility.
Its supercomputer uses six units of Nvidia DGX-1, which is a deep-learning system based on eight of Nvidia’s Tesla V100 GPUs – the world’s first GPU to break the 100 teraflops barrier of deep-learning performance.
A teraflop is a unit of computing speed equal to one trillion calculations per second, while peta-scale refers to petabytes, which is equal to 1,000 terabytes of data. Then in 2023, Nvidia launched the H100 GPU chips aimed at data centres – such as those operated by Amazon’s AWS, Alphabet’s Google Cloud, and Microsoft’s Azure – which plan to keep upscaling their gen AI offerings on the cloud.
Nvidia says the H100 is four times faster than the chip’s predecessor, the A100, at training LLMs, and is 30 times faster when replying to user prompts. In March 2024, Singtel announced it will be launching its GPU-as-a-Service (GPUaaS) in Singapore and Southeast Asia in the third quarter of 2024, powered by Nvidia’s H100 clusters in existing upgraded data centres in Singapore.
The GPUaaS – aimed at companies looking to boost productivity and accelerate growth and innovation without investing in expensive AI infrastructure – will then be expanded to run in three new AI data centres in Singapore, Thailand, and Indonesia over the next three years.
The next stage of the gen AI market’s evolution will be on the edge – which refers to devices or networks that are close to the user, such as PCs, TVs, mobile phones, and other handheld consumer products.
And the consumer electronics market has quite a diversified base of semiconductor suppliers. While Nvidia is the biggest supplier of GPUs used in gaming desktops and laptops, the space is contested by the second-largest GPU maker, AMD, which has a significant 16 presence in Singapore.
AMD and Intel, who also use Singapore-based semiconductor assembly, test and packaging companies to process their chips, are market leaders in the central processing units (CPUs) and chipsets used in PCs.
Both companies are working on their own versions of AI chips and processing devices. Meanwhile, Qualcomm is the top player in the smartphone CPU and chipset market and has already launched a mobile platform that supports a wide range of AI models including LLMs such as Baichuan-7B, Llama 2, and Gemini Nano.
Mr Ehsanul Islam, vice president of engineering and regional head of Qualcomm Southeast Asia, said the company’s latest smartphone chip can run gen AI models ondevice with or without having any cloud connectivity
Hence, as the gen AI market grows, different chipmakers will find their own niche and invest accordingly. “Qualcomm will be looking to increase the capabilities of its test centre here in Singapore,” he told ST.
Mr Sadana said: “Almost the entire production of Micron’s Nand flash memory chip, about 98 per cent, is in Singapore, and it is our main focus for driving the future road map of innovation.”
HSBC’s economist for ASEAN Liu Yun said: “Singapore is the only ASEAN economy with a sizeable share of memory chips, a high-end segment in electronics that has long been dominated by South Korea and mainland China.” But it is not the only area that gives Singapore an edge
“Despite seeing some pull-down in recent years, Singapore has an impressive 20 per cent global share in amplifier chips and 10 per cent for all other types of chips, including processors and controller chips,” she said.
