[MinimalAction]

I understand this is ultimately due to AI boom.

A couple of naive questions:

1. What's the bottleneck in ramping up RAM production? Is it the availability of silicon itself? Or the factories are at capacity?

2. Is this supposed to ease up despite the AI boom? Definitely would ease up if busted.

[icyfox]

1. Factory limits basically. There's a limit to the amount of fabrication lines that can create ram. Combined with the market incentives right now to make high bandwidth memory (HBM) over server memory (DRAM)... HBM starts as DRAM dies, so it competes with normal DRAM for wafer starts / cleanroom fab capacity.

2. Eventually more plants will come on line. Most of the main manufacturers have announced expansions but these can take O(years) to come online.

[srdjanr]

Are more plants coming? I think I heard it won't be many of them, because it's risky.

If the bubble bursts and RAM demand drops, then they'll have big losses. And that's not an impossible scenario over the few X years that it takes to build a plant

[zettabomb]

The entire capacity of RAM production is basically booked out, for at least the next year. All fabs have sold their allocations already, and it takes years to build a new one. As a result, no it will likely not ease up if demand continues like this, again for at least a year.

[lenerdenator]

Also worth noting:

They sold their allocations to people who don't have a clear path to profitability, and were paid with massive amounts of money that don't exist in reality.

[root9876]

> and were paid with massive amounts of money that don't exist in reality

How's that possible?

[kuschku]

You know how fractional reserve banking can create money out of thin air? In theory, the stock market can be used in the same way. Which is what the AI providers have been doing.

[chadgpt3]

I agree to pay you within 90 days of receiving the product, which you'll send me in 6 months. You get to count this promise as income i've paid you right now. 9 months later I might or might not have the money to pay it for real.

The principle works for stable regular businesses where invoices do get paid, not for bubbles.

[lenerdenator]

I'd argue the opposite: It works even better for bubbles.

Stable, regular businesses generally don't take on massive amounts of debt as a matter of accomplishing their value proposition. Maybe they take on the debt as a part of an expansion into a new geographic market or on developing a new product, but they typically have cash reserves to help fund those things. Accordingly the debts are relatively small and there's at least some collateral to secure them.

Any kind of bubble, on the other hand, is, by definition, financially massive. People taking on debts take on huge debts with the hopes of huge rewards. The problem is, the debts are so huge that neither the debtor nor creditor can allow them to go unpaid.

When you reach the scale of debt that you see for AI - literally trillions of dollars - the creditor isn't just a bank or a few investors, they're entire segments of the economy. The debtors can use that as leverage to get what they want. The Great Recession is a prime example of this. You had the US government print out or borrow $700 billion-ish dollars to absolve financial institutions of the toxic mortgage-backed securities on their balance sheets. Now, could they have simply declared bankruptcy and had the assets on their balance sheets claimed by other parties? Maybe, but that would have taken too long for most people's comfort and, moreover, would have been embarrassing for the country's monied elite. There had to be a bailout to keep a bunch of things solvent in the near-term, and they got one.

You could see this here: AI companies have everyone on the hook for their bets, and if it doesn't work, you can't simply write off the value of shares of everyone who has backed the bet.

They're too big to fail. If you owe the bank $1 million, you have a problem. If you owe the economy $1 trillion, the economy has a problem.

[lenerdenator]

Credit.

[locknitpicker]

> 1. What's the bottleneck in ramping up RAM production? Is it the availability of silicon itself? Or the factories are at capacity?

For a RAM manufacturer, the incentive is to ramp up production AND prices. I doubt any of the names in the business is doing any work at all to lower their unit prices.

[Hamuko]

The current pricing isn't sustainable, and if you try to wring out the maximum amount of profit, you're gonna have competition spring up. The Chinese are probably salivating at this opportunity. Previously they would've had to sell memory at discount rates to get anyone to switch over to them, but now they'd have customers lined up for years if they were selling at the prices Samsung/Micron/SK Hynix were selling a year ago.

[locknitpicker]

> The current pricing isn't sustainable, and if you try to wring out the maximum amount of profit, you're gonna have competition spring up.

I don't understand what you mean by "sustainable". The whole industry tripled their process AND is maxing out their production lines. They are cashing out as expected.

It matters nothing if this trend can't be kept up for years, because by then their output still meets demand AND they will be sitting on a huge bag of cash.

I think you are confusing "I don't want to pay this much" with "this isn't sustainable". The sellers are cleaning up stock at a huge markup and buyers are still buying like crazy.

[Hamuko]

It's not sustainable because it'll basically ruin all of the industries connected to memory. OpenAI, Anthropic etc. all need a massive amount of compute to train and serve models, and increases in memory price will all drive up their already massive costs. And then on the other end, the people they need to use their AI services cannot either afford the devices they need to access the devices, or the device costs go up so much that they cannot afford to buy the services. If for example budget phones disappear completely because of memory prices, those consumers disappear from the customer pool and then the budget phone manufacturers disappear and stop buying memory, AI companies spend even more to build models for fewer people, etc.

Will there be a budget phone market with these prices? What about budget laptops? Can there exist a gaming console market when memory costs are so high? And what happens to the companies that sell goods and services to the users of those devices?

And obviously, if people see that you're extracting an exorbitant amount of profit, you're gonna have a lot more people eyeing your industry as ripe for competition, since some company would probably not mind eating some of your pie. I'd say that the current memory cartel has a very real risk from China right now. And if you get competition during the boom period, you're also gonna be competing with them during the bust period too.

[hurtigioll]

OpenAI/Anthropic are in the business of "replacing workers with AIs", not in the business of "selling AIs to workers"

that is a 20T market

they dont need you to afford a subscription, they need your boss to replace you with their service

[Hamuko]

I wonder about that. At the moment, it feels like OpenAI and Anthropic are in the “racing to IPO” business. I’m also not sure who is going to be buying all of this AI output when nobody’s employed.

[dopa42365]

Meet the new memory cartel, it's just like the old memory cartel but with CXMT being priced a tiny tad below.

Safe to say they're not in it out of sheer altruism.

[chadgpt3]

I heard a long time stock market analyst saying the semiconductor industry always has boom and bust cycles. This is a boom (for the semiconductor industry) and it will be followed by a bust as demand rapidly drops off, just like every other time.