Samsung Electronics warned its third-quarter profit would come in below market expectations and apologised for the disappointing performance with the tech giant lagging its rivals in supplying high-end chips to Nvidia in the booming AI market.

Samsung said its AI chip business with an unidentified major customer was hit by a delay, while Chinese chip rivals increased supplies of conventional chips, contributing to the decline in its semiconductor earnings.

The comments illustrate the challenges facing the company, which has been the world's biggest memory chipmaker for three decades but is battling growing competition in both conventional and advanced chips.

The world's largest memory chip, smartphone and TV maker estimated an operating profit of 9.1 trillion won ($6.78 billion) for the three months ended Sept. 30, versus a 10.3 trillion won LSEG SmartEstimate.

That would compare with 2.43 trillion won in the same period a year earlier and 10.44 trillion won in the preceding quarter.

"The earnings are a shock compared to what many analysts expected initially," said Lee Min-hee, an analyst at BNK Investment & Securities.

"I don't see its earnings improving in the current quarter," he said, saying it lags SK Hynix in increasing sales of high bandwidth memory (HBM) chips to Nvidia and its high exposure to the Chinese market hurts.

Samsung's late response to the AI chip market increases its reliance on traditional, lower-margin chips, making it more vulnerable to competition from China and slowing demand for smartphones and PCs, analysts say.

High-margin chips used in AI servers are driving a recovery in the chip market after a post-pandemic downturn last year. Still, Samsung has lagged SK Hynix in supplying high-bandwidth memory (HBM) chips to AI leader Nvidia.

"We have caused concerns about our technological competitiveness, with some talking about the crisis facing Samsung," Young Hyun Jun, Vice Chairman, Device Solutions Division, Samsung Electronics, said.

"These are testing times," he said, pledging to turn the challenge into an opportunity and focus on enhancing long-term technological competitiveness.

Samsung's share price, already down more than 20 percent so far this year, fell 1.6 percent after the earnings guidance.

HBM CHIPS DELAYED

Samsung said in a statement the start of sales of its high-end HBM3E chips to a major customer "was delayed relative to our expectations". It did not elaborate on the issue.

Earnings declined in the company's memory chip business as Chinese rivals increased supplies of "legacy" products and some mobile customers adjusted inventories, offsetting solid demand for HBM and other chips used in servers, Samsung added.

Samsung's contract chip manufacturing business, which designs and produces custom-made chips for other companies, likely continued to lose money in the third quarter as it is struggling to compete with leader TSMC, which counts Apple and Nvidia among its customers, analysts said.

Samsung's chief Jay Y. Lee told Reuters on Monday that he is not interested in spinning off the contract chip manufacturing business as well as its logic chip designing operation.

Samsung said one-off costs such as provisions for "incentives" and the unfavourable local currency also contributed to the chip earnings decline.

Earnings in its mobile division improved from the preceding quarter on solid sales of its flagship smartphones, while earnings at its display unit grew as its customers, which include Apple, launched new models.

Samsung will announce detailed earnings results later this month.

In May, Samsung abruptly replaced the chief of its semiconductor division, handing the reins to Jun in a bid to overcome a "chip crisis".

Samsung is also cutting as much as 30 percent of overseas staff at some divisions, Reuters reported in September, underscoring the challenges it faces.

Its US rival Micron last month forecast first-quarter earnings ahead of Wall Street estimates and reported its highest quarterly revenue in over a decade on the back of booming demand for its memory chips used in AI.