It's actually like 7 companies, but four of them are small enough that no one cares (Kingston, Kioxia, WD, and Nanya). But even if we assume 3, that has nothing to do with a "cartel." It's not even a natural monopoly or monopsony. They very much compete with each other, which is WHY there are only 3 big memory manufacturers.

Most people forget that just a short time ago, there actually used to be tens of memory companies. They don't exist anymore because most of them went out of business because memory is an absolutely terrible business. It's a largely undifferentiated product that's stupidly capital-intensive, has a very long return on capital window, and 9 years out of 10 the margins are functionally 0% after inflation. SK Hynix would make more money most of the time if they took all the capital they spent on building fabs and just dumped it into an index fund.

But occasionally, very very rarely, you have a year like this one (and next one) where demand skyrockets and you actually make profits worth caring about. The last time this happened, which was about 10-15 years ago, a bunch of smaller fabs decided the good times were going to last forever and dumped billions into new fabs... which were still being built by the time memory went back to commodity pricing and they all went bankrupt and got bought up for pennies on the dollar by the big three we know today.

That's why SK, Samsung, and Micron have all said they aren't dumping cash into accelerating capacity expansion: they only dominate the space because they understand how boom/bust cycles work and are conservative with capital.

Could they all spend $50 billion to build new plants/expand current ones? Sure. But those plants: 1. Wouldn't even break ground for at least 18 months. 2. Wouldn't be completed for at least 36 months. 3. Wouldn't start meaningful production for at least 42 months. 4. Wouldn't start producing a positive ROI for a minimum of 60 months, and really closer to 120 months (10 years).

If Samsung started planning a new fab today, it wouldn't meaningfully increase the amount of available memory on the market for FOUR YEARS. And in four years, all of the big new datacenters would be mostly finished and suddenly demand for memory would drop off a cliff and Samsung would be sitting there with its dick in its hand looking like an idiot and out several billion dollars. Building new fabs now is the equivalent of buying GameStop stock when it was $400/share.

And here's the proof that this isn't just price-fixing or "cartels" or late-stage capitalism or whatever other nonsense people come up with: there are exactly two places where it makes sense to build memory fabs right now. China and Vietnam. Both are functionally state-controlled economies were the government can (and does) intervene to expedite annoying things like planning, environmental impact, zoning, land-use approval, etc. Those are the only two countries where you can conceivably build a memory manufacturing facility in less than three years. And they aren't going all in on it, even though China's (and Vietnam's) entire economy is still largely based on exploiting lucrative niches cheaply (a.k.a. "bootlegs" and "knockoffs"). Because even they realize that this is a temporary demand spike that'll settle out long before they can make enough profit to be worthwhile.