Someone asked, "OpenAI has discovered a technology that can halve inference costs—could this be bearish for the semiconductor industry?"

Inference is actually better when it's cheaper; it reduces costs, not dependence on hardware. Of course, there is short-term bearish sentiment, but that depends on the details they disclose—otherwise it's hard to determine which upstream segment will be affected.

In reality, the only way to crash the memory storage market is oversupply: a sudden explosion in output, where everyone is no longer short of supply and can buy as much as they want.

Price increases are essentially to raise funds for expanding production—at its core, it's the downstream side (like Apple) repaying the debt they incurred from aggressively squeezing prices in the supply chain back in 2023.

As for efficiency gains, the impact of algorithms is temporary, and the market already expects them to keep improving. Besides the Jevons Paradox that everyone has heard so often it's become an earworm and is being repeated across the internet:

On one hand, the more you improve efficiency, the more I buy, which gives me a competitive advantage.

On the other hand, downstream demand is so tight right now that management won't stop placing orders just because of some news—who would save a little money now only to repeat the mistake of not being able to get supply the next time it's needed?

What these people are probably thinking now is: "If only I had placed ten times the storage orders last year."

Once bitten, twice shy. After experiencing such a huge shock, how could anyone possibly cut orders?

Therefore, I believe the demand for storage is long-term and stable.

Another possibility: a group of geniuses develops a substitute for NAND/DRAM.

But whatever the substitute, it will still be a semiconductor. And if it's a semiconductor, manufacturing it requires lithography machines and foundries.

Isn't the current bottleneck precisely the shortage of lithography machines and foundries? Can new capacity just fall out of the sky? The real reason for storage price increases isn't that DRAM is hard to manufacture or has high technical barriers—it's that there aren't enough factories.

So, we can fully interpret the current industry-wide signing of LTA contracts, and even institutions aggressively buying stocks and raising target prices, as helping MU and SNDK raise funds to build factories, providing new capacity for humanity's future.

This way, you won't always worry that storage is some kind of bubble and keep thinking it will crash.

Storage is truly different from silver and Bitcoin—at least, I've never seen anything like this in five years.

Tim Cook, on the other hand, says it's the first time in his 40 years.

Thank you, everyone.

😂😂😂