The world’s biggest memory chip maker said the worst is over for the global semiconductor market but announced plans to extend production cuts because a demand recovery is primarily constrained to high-end chips used in artificial intelligence. The move underscored the unprecedented semiconductor downturn that led the South Korean firm to incur a record 8.9 trillion won ($7 billion) operating loss from its bread-and-butter chip business in the first six months of this year.

On a conference call, Samsung’s executive vice president Jaejune Kim told analysts that the company would continue to cut production, making adjustments for specific products, while doubling capacity for high-density memory like IBM, which is used in AI. He said this resulted from robust demand for AI-driven technologies, including 5G, the Internet of Things, graphic processing applications, and booming investments by major hyperscalers.

He added that memory demand is expected to recover gradually, with customers destocking inventories while purchasing new products. But he added that “the recovery speed will depend on the macro economy and customer inventory levels.”

Investors brushed off a 96% plunge in quarterly operating profits as Samsung said it expects demand for memory chips to recover steadily in the second half and beyond. It also unveiled a plan to introduce 2-nanometer production for mobile phone chips by 2023, a significant step toward becoming more efficient in its foundry business. That was seen as a way to help it better compete with Apple’s (AAPL.O) premium iPhones.