@Blueyatagarasu@teortaxesTex And, about money, even if they need to do exponential data centers, the capex needed to do hyperscaller datacenters in space would be so brutal that is not going to happen.
My guess is data centers would exist in space, but reduced edge datacenters for surveilance and military
@Blueyatagarasu@teortaxesTex Is a bet to make, but is a risky bet to put money on, so it has to be evaluated like that. That's why they are trying to do fast IPOs and access fast track to indexes, to socialize losses, and hedge their value with OUR money.
@Blueyatagarasu@teortaxesTex Investing in exponential hardware just because is only worth it if the real bottleneck is compute. And there is no evidence about that.
@Blueyatagarasu@teortaxesTex And what if the massive buildup of hardware brings nothing and the bottleneck is algorithmic? Or material based?
It's a possibility that would mean investing that much without reason, or even worse, a dead end.
The clock is the depreciation of hardware, and the investor money.
@Blueyatagarasu@teortaxesTex Not necessarily. If we achieve AGI, the economy will change completely. Everything could happen.
I just think there will be a plateau because of market laws. Money can't be churned continuously without return, and hardware depreciation will make that more severe in a few years
@Blueyatagarasu@teortaxesTex This is similar to the railroad mania of the 19th century. It will end up in a plateau. Algorithmic efficiency, and lack of ROI will cause that. Also, in that case the hardware depreciates quite fast. give it 4 to 6 years
@Blueyatagarasu@teortaxesTex mW, not kW* few days with almost no sleep sorry.
So my point is: it's just smoke for IPO. it will be used for some military usage, but not comercial for sure.
@Blueyatagarasu@teortaxesTex (btw my numbers were wrong, would be more in tje order of 15-20T/kW, but still really expensive, 0.8B just the lift)
Yeah, but computing will plateau in a few years probably, and there's no way a datacenter can be built in that time if the technology doesn't exist yet.
Intelligence should be open, accessible, and ready to build with, empowering every developer, everywhere.
GLM-5.2 is now available to all GLM Coding Plan users, including Lite, Pro, Max, and Team plans.
https://t.co/AedZACyzej
As our new flagship model, GLM-5.2 delivers powerful coding capabilities, usable 1M-context support, and continued strengths in long-horizon tasks.
API and Chatbot services will launch next week. The model will also be officially open-sourced next week under the MIT License.
The future of AI is open, and it belongs to the people.
@teortaxesTex@Blueyatagarasu But i don't see investors paying an extra without a clear benefit. Is only fit for some strategic data usage. For example, national security or surveillance servers. Not commercial scale. Investors want return on their money, so cheap wins.
@teortaxesTex@Blueyatagarasu Fuck, just noticed i said 200, when i meant 20, and did calculations with that. My mistake. Still not commercially viable, but more usable. The components are more expensive and just the freight is the same as the entire cost of a data center in earth. Good for some applications
@scaling01 Don't think so. I guess they are just treating AI as a military contractor now. That means, Anthropic and OpenAI now will be lockheed and boeing, and US taxpayers will pay for it.
Let's see what does Xi does in response.
My guess is, Open Souce AI is over now.
@teortaxesTex@Blueyatagarasu Ok, give me the source. Becuse i looked and that number is the estimation of just the power generation. like i said before, you would need a very very big refrigeration surface... and is not covered in that number. Sowy to hurt ur feelings
@teortaxesTex@Blueyatagarasu You are only counting power generation on that 6k, using ultra light power generatio relays. Not refrigeration, plumbing, robotics for assembly, shielding for night/day cycle termical... i've made the number given a guess of the cost based on ISS numbers
@teortaxesTex@Blueyatagarasu Is not comercially viable. Is a megaproject in such a scale that would increse inference cost to x20. I don't see the reason a investor would want that money burner.
@teortaxesTex@Blueyatagarasu The actual cost in the earth of such datacenter is around 1 billion. we are talking that is 4 times the price JUST transporting the materials, not talking robotics, cost (would be far more expensive to build), and very very complicated logistics