Yes, I'm sure they can afford to buy tens of billions in servers just to store until obsolescence while they're trying hard to even become profitable lol
OpenAI bought up memory wafers, not ready-made systems from nvidia. That's something they may actually be hoarding to create a bottleneck.
Also true.
But if you choose to run software that uses a lot of RAM, ask yourself why you haven't created an alternative that doesn't use a lot of RAM. If the answer is "I don't have the time to", then that's probably also why the developer hasn't made it use less RAM.
There's also those of us whose builds date back to 2020. Holding out another 5-10 years is a bit of an ask.
It takes double digit billions to start a manufacturing plant and that's when you already have people who know what to do.
Most countries can't really afford this in their budgets and I'm saying countries because it'd be a stupid endeavour for most private enterprises to even attempt. CXMT (DRAM) and YMTC (NAND) absolutely are sponsored by China, which is the only real way to get one of those companies going these days.
Google, Apple, etc could start their own memory companies if they wanted to. But it's a hell of an expense to justify to your investors.
Why wouldn't China just charge the increased prices, same as the DRAM cartel? Primarily they want to make money.
Unless they offer below-cost RAM for a few years to drive the competition out of business before jacking up their prices of course.
Run less software then. You can do that today.
Locally, not necessarily true. Globally, almost definitely.
It was still far from a "big" car. A contemporary Mercedes S-Class was nearly half a meter longer than the 4 door Range and that's a sedan lol.
Range Rovers are now as big as luxury sedans were in the 1980s.
Claude's still there, seems Alibaba's attack wasn't really all it's cracked up to be.
Now the US gov's attack seems to be working since Claude Fable 5 is still not there.
What's the power scale of yours?
I suspect the big ones use evaporative cooling because they're trying to build in the gigawatt scale and IIRC there was talk about single racks reaching a megawatt soon, currently they're ~150 kW.
The power density of those new nVIdia GPU compute servers is nuts and using evaporative cooling means less energy use than closed loop.
What I'm saying is, some of those planned datacenters wouldn't be feasible with closed-loop water cooling. And yes, I agree with you that this should be legislated. If they can't cool their servers without evaporating a bunch of drinking water, they can... have fewer servers.
Nobody is buying to destroy. They're all starved for compute right now. Hence the dynamic pricing to stop people using at peak times.
And when they put all of that into their new RAM business, their investors are going to have a bunch of questions.
Even with tens of billions in cash it's not that simple. Like I said, you first start off by convincing a bunch of people their non-competes and give up their massive annual bonuses, so that's going to be an investment that the existing companies don't have to make. Then you'll have caught up with RAM manufacturing in about 10 years or so once you've ironed out all the details.
China, the biggest economic powerhouse in the world and the epicenter of modern electronics manufacturing, is still not making HBM3 or 4 or whatever nvidia uses now for Rubin at a commercially viable scale at CXMT, they're only barely making enough DDR5 to get contracts.
That's after starting the company in 2016 and getting DDR4 manufacturing going in 2020 and getting people from Samsung to steal trade secrets (and come work for them).
Just because you can build a fab for 20 billion doesn't mean it's only 20 billion to get started, or a quick 2-3 year endeavor to do so. Publicly traded companies can't really do this because it's a bunch of expenses over multiple years without a single cent of revenue for many years. This needs to be properly private cash or government backing.