What he means is this. Many people want to run AI right on the computer, and think that with an open source model, there is no need for a data center anymore. But looking at the perspective of allocation for the whole world, according to Altman, having the limited amount of DRAM in personal computers, which is only used about 5% of the time, is very wasteful. While the supply is tight like now, he wants RAM to be in the cloud to run at full capacity.

Altman is not wrong when it comes to performance

I think we need to be fair first. Looking purely at the numbers, Altman's idea makes sense. The computer at home spends most of its time just waiting, surfing the web or opening a few tabs, and the RAM in the server is almost always busy. The same amount of memory chips, placed in a data center, can serve more people.

The context of the statement is that he is criticizing the idea of ​​running AI right on the machine. What he wants to emphasize is that running an open source model at home does not make the need for data centers disappear.

SK hynix AI server DDR5 RAM bars on display at MWC 2026

The problem is who is holding most of that RAM

The sentence would have sounded much lighter if it hadn't come at this exact moment. According to The Verge, Samsung and SK hynix have signed long-term contracts with cloud operators and projects like OpenAI's own Stargate, locking in the majority of the world's RAM output for the next few years. The billions of dollars these parties are willing to spend to buy chips is the reason why laptops, phones and other technology items you buy are getting more expensive.

In other words, RAM in personal computers is less and more expensive partly because it is reserved for data centers. Now the head of one of the biggest buyers is saying RAM lying on your desk is a waste, so I understand why many people have a hard time hearing it.

Where have RAM prices gone

TrendForce in a report at the end of September said that DRAM manufacturers are still prioritizing advanced production lines for high-performance server products, so the whole market continues to lack goods. As for regular DRAM, the fourth quarter contract price is forecast to increase by 10 to 15% compared to the previous quarter.

According to Commercial Times cited by TrendForce, overall DRAM contract prices in the second quarter increased by about 80%, in the third quarter by about 40%, and in the fourth quarter are expected to increase by about 20%. The increase slowed down, but it was still an increase on top of the increase, not a decrease.

Advertisement

DRAM contract price increase by quarter, fourth quarter is forecast

When it comes to retail buyers, the numbers are even more unpleasant. TweakTown said the cheapest 32GB DDR5-6000 kit on the market has increased to 449 USD, while a Corsair 32GB DDR5-4800 kit that previously sold for about 80 USD is now 399 USD, nearly five times more.

Previous and current price of a Corsair 32GB DDR5-4800 kit

The price is not just in the RAM

This doesn't stop at PC builders. TrendForce said fourth-quarter laptop inventory reflected higher component costs and was slowing sales. PC manufacturers are still buying RAM heavily because they believe that the supply will still be tight in 2027, and the amount of DRAM for PCs next year may decrease as more capacity is transferred to servers.

The same goes for the phone. RAM for phones is being strictly allocated because chip manufacturers are shifting less to servers, high costs are dragging down smartphone output. Intel, in its third quarter report call, forecast that PC sales in 2026 will decrease by about 12%, and will not see a recovery until memory prices calm down.

Advertisement

Memory showcase for SK hynix's AI data center, where capacity is being prioritized

Why do I still want RAM in the device

I understand the argument about efficiency, but personal computers are not just about resource utilization. The RAM in the device is what helps you work without the internet, the data is in your hands, playing games or editing videos without having to wait for anyone to allocate it. Running AI on a machine is the same, it has its own reasons for privacy and not being dependent on a service.

If you push everything to the cloud, of course you can run more RAM, but it also means an additional monthly payment, and what you used to own is now something you rent. For me, 5% of usage time doesn't mean the remaining 95% is useless. That is the surplus so that the machine is available when needed.

What should you think now

In the immediate future, there are no signs of easy breathing. SK hynix once warned that the memory shortage could last until 2030, and PC companies are buying stocks because there is still a shortage in 2027. So I think like this:

If the device is usable then keep it, there's no need to upgrade the RAM when the price is at its peak. If you must buy a new computer, choose one with enough RAM for a few years, because upgrading later will likely cost more. If you have extra RAM in your old computer, don't rush to sell it cheap.

What do you think about Altman's idea, should RAM be on your desk or in the cloud?

Source: The Verge, TweakTown, TrendForce, Intel