AI dues are coming as soaring demand for memory chips set to boost computer prices

SINGAPORE - New technology comes at a price.

While chatbots such as ChatGPT, Gemini and Copilot are accessible for free, dues for the artificial intelligence (AI) boom will be paid through a different channel – more expensive electronic devices.

PCs are first in line, with prices of laptops and desktops set to rise by 15 per cent to 45 per cent through 2026. The higher end of the price range will be for custom-built, highly specialised desktops for demanding tasks such as high-end gaming, streaming and crypto mining.

Some of the world’s biggest PC makers, such as Lenovo, Dell and HP, are likely to announce new prices within weeks, according to Taiwan-based TrendForce, a tech industry consultant and market intelligence provider.

Surprisingly, the price surge is not driven by cutting-edge logic semiconductors that give computing power to devices.

Made by contract manufacturers such as Taiwan’s TSMC, South Korea’s Samsung Foundry and China’s SMIC, logic chips stole the initial limelight amid the unprecedented AI infrastructure build-out.

Instead, prices are rising due to the dramatic spike in the cost of memory chips – semiconductors that store and process data – that hyperscale data centres soon realised were key to delivering the promise of generative AI at scale.

This is a win for Singapore’s semiconductor industry, a critical node in the global chip value chain. The Government is also taking policy measures and investing heavily in research and manufacturing centred on AI innovation and other emerging technologies.

In January 2025, the US’ Micron Technology

broke ground for Singapore’s first high-bandwidth memory (HBM) advanced packaging plant

worth US$7 billion (S$9 billion) in investment, putting the Republic on the global AI supply chain map.

Including Micron’s HBM plant, the AI tailwind has brought more than $18 billion in research and development and manufacturing investments to the Republic’s semiconductor ecosystem over the past two years, according to Singapore’s Economic Development Board.

The latest generation of HBM chips provides faster data processing and unmatched power efficiency needed for AI workloads. Hence, data centres over the past year scrambled to book deliveries, sending memory chip prices, specifically for dynamic random-access memory (DRAM) chips, skyrocketing.

Prices for spot DRAM chips have almost tripled since 2024, with those for the latest-generation chips called DDR5 spiking by more than 300 per cent since September 2024, said TrendForce.

Some observers have said the upcoming surge in PC prices can lead to serious repercussions for chipmakers.

With the price increases coming at a time when PC sales by units are set to decline worldwide, there is concern that most PC makers will go for deep production cuts, which could then lead to lower overall demand for semiconductors.

While the top high-end mobile phone makers such as Apple and Samsung have so far demonstrated their pricing power by coming up with new and more expensive models every year, PC makers have seen their margins getting squeezed rather quickly.

But analysts such as Japanese investment bank Nomura’s Asia ex-Japan economist Sonal Varma have a different take.

She acknowledges that eventually, higher memory chip prices are likely to increase cost pressures and result in lower end-demand for both PCs and smartphones. Nomura estimates worldwide PC sales to drop by 2.8 per cent, while smartphone sales will shrink by 1.7 per cent in 2026.

But this time round, the chipmakers – especially those who make memory chips – are unlikely to suffer from the boom-and-bust cycle of the consumer electronics industry.

Memory chip price effects should turn even more favourable for chipmakers in 2026 amid a memory chip super-cycle, she told The Straits Times.

Nomura estimates that DRAM chip prices, which saw a 16.9 per cent year-on-year growth in the third quarter of 2025, will rise by 55.4 per cent in the second quarter of 2026.

Meanwhile, NAND chips – used for long-term storage of data, including audio and video on flash and solid-state drives – will surge from a drop of 20.5 per cent in the third quarter to an 82.3 per cent jump in the second quarter of 2026.

Powered by the promise of AI to boost productivity, automate processes and lower costs, the demand from industrial and business customers will continue to grow fast enough to compensate for any loss of demand otherwise.

Estimates of companies’ capital spending plans on AI range between US$5 trillion and US$8 trillion globally through 2030, said investment manager BlackRock.

While there are quite a few producers of logic chips, the global memory chip market is literally a de facto triopoly of South Korea’s

SK Hynix

and

Samsung

and the US’ Micron. They command a combined market share of more than 90 per cent in high-growth segments of DRAM and HBM chips.

This means that customers seeking to build devices with top-of-the-line memory chips have few options to seek a bargain elsewhere.

Mr Shawn Kim, Morgan Stanley’s head of Asia technology research, said prices quoted to customers by chipmakers for data centre servers for the fourth quarter of 2025 are up by nearly 70 per cent.

“While most contract transactions will likely be finalised later this month, acceptance appears inevitable,” he said in a report released in November. “Margin pressure is likely to persist in memory-consuming segments such as PC, mobile and broader consumer supply chains.”

NAND contract prices are also moving higher, up by 20 per cent to 30 per cent in the fourth quarter, driven by firmer enterprise solid-state drives quotations and limited alternatives for buyers, he noted.

“Given the scale of AI infrastructure investment and hyperscale customer dynamics, we expect peak pricing to surpass the previous cloud super-cycle high of early 2018,” said Mr Kim.

He added: “PCs, consumer electronics and non-premium smartphones are unlikely to absorb higher memory chip costs, leading to demand destruction and reduced set shipments in 2026 and 2027.”

Analysts believe memory chipmakers will retain their pricing power through 2026 as upcycles in the industry usually last for about six quarters.

Mr Kim said an accelerated computing build-out is under way across data centres, now expanding into agentic AI, AI video creation and soon, physical AI – representing a host of generative AI use cases.

While everyone is building AI infrastructure and racing to develop applications, tools and services on top of it, the ultimate question is whether the initial infrastructure cost will exceed the long-term value realised at the application stage, he said.

“But we remain early in this build-out phase. We think pullbacks are inevitable as we move from a bull market towards euphoria,” he said, referring to the tech frenzy, especially in the US stock market.

“Still, (tech company) earnings have provided a clear signal to remain constructive on the AI theme – particularly through memory chip stocks,” he added.