Memory chip stocks slide as AI efficiency breakthrough rattles investors

Memory manufacturers have been among the biggest beneficiaries of the AI boom

Memory chipmakers around the world shed close to $100bn in market value last week, after new research suggested AI systems may require far less hardware than previously expected.

On Tuesday, Google announced its TurboQuant algorithm, which it says can dramatically shrink AI models without sacrificing accuracy, allowing them to run on machines with significantly lower memory requirements.

The announcement sparked concern among investors, fuelling doubts about whether AI will continue to require such vast storage capacity.

In South Korea, shares of SK Hynix dropped 6%, while Samsung Electronics fell nearly 5%. Japan's Kioxia also declined by almost 6%.

The downturn followed earlier losses in the United States, where SanDisk and Micron Technology both weakened in trading.

Google “Turboquant”

According to Google, its "TurboQuant" technique could reduce the memory required to run LLMs by up to six times. It does so by optimising how models store and reuse previous computations, known as the "key value cache".

The breakthrough is part of a broader push within the industry to make AI systems more efficient, lowering costs and energy consumption.

Matthew Prince, chief executive of Cloudflare, described the development as "Google's DeepSeek", referencing efficiency gains by a Chinese AI firm that triggered a market sell-off last year.

"So much more room to optimize AI inference for speed, memory usage, power consumption, and multi-tenant utilization," he said in a post on X on Wednesday.

Investors reassess expectations

Memory chip companies have been among the biggest beneficiaries of the AI boom, as demand surged for the infrastructure needed to train and run models developed by firms such as OpenAI and Anthropic.

Investors had widely expected a prolonged shortage of such components, driving strong earnings growth well into next year.

But analysts now say improvements in efficiency could reshape those expectations.

Experts at Morgan Stanley noted that if AI models can operate with significantly less memory, the cost of running them would fall, potentially reducing the need for large-scale infrastructure.

"Thus, models that need cloud clusters can fit on local hardware, effectively lowering the barrier to deploying AI at scale," Morgan Stanley analysts wrote.

Ben Barringer, head of technology research at Quilter Cheviot, said declines in memory stocks were largely the result of investors taking profits.

"Memory stocks have had a very strong run and this is a highly cyclical sector, so investors were already looking for reasons to take profit," Barringer said.

But analysts caution against overreaction

Despite the market turbulence, some analysts caution against overreaction.

Morgan Stanley said the longer-term impact could be neutral, as lower costs may encourage wider adoption of AI technologies, ultimately increasing overall demand.

Others point out that techniques like quantisation, the process underlying Google's breakthrough, have been studied for years. What has changed is their readiness for large-scale, real-world deployment.

"The Google Turboquant innovation has added to the pressure, but this is evolutionary, not revolutionary. It does not alter the industry's long‑term demand picture. In a market primed to de‑risk, even an incremental development can be taken as a cue to lighten up," Barringer noted.

For businesses, the development could offer new opportunities. Companies that previously delayed AI projects due to high infrastructure costs may now find them more accessible.

At the same time, competition among major tech firms, including Microsoft, Amazon and Meta, is expected to accelerate efforts to improve efficiency.