The high-bandwidth memory (HBM) market is continuing to show strong growth, with the latest earnings reveal coming from South Korean chipmaker SK Hynix, which posted a record US$5.6 billion profit for the December quarter.
Q4 figures also show a 12% rise in profit and 15% jump in revenue, with HBM making up 40% of DRAM sales.
The chipmaker is a leading supplier to the world’s largest company US$3.6 trillion tech giant NVIDIA (NASDAQ : NVDA) which is seeing a boom in GPU sales that underpin most AI systems and data centre requirements.
“SK Hynix emphasises that with prolonged strong demand for AI memory, the company achieved [an] all-time high result through [our] world-leading HBM technology and profitability-oriented operation,” the company said in its earnings release.
The $111bn tech manufacturer echoes industry forecasts of continued demand for HBM and high-density server DRAM - integral to high-performance computing.
“[Demand] will continue to increase as the global big tech companies’ investment in AI servers grows and AI inference technology gains importance," SK Hynix said.
“In the consumer market, in which inventory adjustment is expected, sales of PC and smartphones equipped with AI will expand and the market situation will pick up in the second half of the year.”
The sector is further boosted by tech giants Google, Microsoft, Meta and Amazon. These companies pour billions of dollars into AI servers, data centres and interface technologies.
A $130bn HBM market in eight years' time?
Bloomberg Intelligence says the high-bandwidth memory (HBM) chip market is set to rocket from a US$4 billion market in 2023 to US$130 billion by 2033, driven by the explosive growth of artificial intelligence models across business and lifestyle.
If correct, that’s 3,200% market growth in 10 years.
And memory performance is becoming a top priority for these customers.
“AI chips require memory with high speed, which could drive higher prices as the technology evolves and new generations are introduced,” Bloomberg tech analyst Jake Silverman said.
“As such, the market is set to expand at an average of 42% a year, making it more than 50% of the overall dynamic random access memory (DRAM) market in 2033 and comprising 10% of industry bit shipments — all led by the growing demand for AI infrastructure.”
“We’re witnessing an explosive growth period for the HBM market, as demand for HBM chips grows to keep up with an increasing number of large language models (LLMs) for AI.
If the number of parameters for LLMs continues to rise until 2033 at a similar rate to its current performance, HBM demand could even exceed 8.7 billion gigabytes.
$500bn injection
After being back in the White House for less than 48 hours U.S. President Donald Trump sent the chip sector into a frenzy by announcing Project Stargate - a half-a-trillion-dollar investment into AI infrastructure.
Equity funding for the Stargate project is provided in a JV with OpenAI, Oracle, MGX and Softbank. It will also feature Microsoft, NVIDIA, Arm and Oracle and Open AI again as key tech partners.
However, Elon Musk was sceptical about how companies could cough up that kind of cash, taking it to his social media platform X:
“They don’t actually have the money,” Musk wrote.
“SoftBank has well under $10 billion secured. I have that on good authority.”
Next-gen HBM
New generations of HBMs as technology develops may also lead to price increases that help grow the market.
“The next generation of HBM, dubbed HBM4, is expected to be introduced in 2H25/2026 and contribute to market revenue meaningfully by 2027,” Silverman says.
“This generation may require a more complicated manufacturing process that would reduce the number of dies per wafer and increase input costs, with a predicted pricing increase of 20% over the current HBM3E.
SK Hynix is expected to remain the leading global HBM supplier over the next decade. However, Bloomberg reckons its market share may decrease from 50% to 40% - as rivals like Micron (NASDAQ: MU) and Samsung catch up on technology and rapidly expand their capacity.
