GateUser-2124263a

I took a high-speed train, and all the older folks around me were watching AI short dramas

So shorting semiconductors is the most crowded trade right now?

Interesting observation: after the 5090 surged and then fell, the 5080 and 5070 Ti started rising. It’s just like trading stocks and crypto

I don’t understand where homeview would be applied. Could it be the future interface for interaction between AI and humans?

The news that the 5090 will be discontinued seems pretty credible, so if you bought a 5090, you can quietly celebrate

strata is about to send memory usage soaring again

Approximate direction of sector rotation:
Optical modules - storage - MLCC - cloud - big tech

The proliferation of cloud-based agents is beneficial for traditional cloud providers, CPUs, and storage. Uh, storage was already in short supply—now it’s even scarcer.

Wasn’t the RTX Pro 6000 banned from sale in China by the US? Does that mean the 5500 is allowed to be sold in China? What standards has the US Department of Commerce set?

Micron’s earnings report on the 30th is likely to beat expectations.

If I set up a server rental business in the US offering clean residential IPs, I wonder if anyone would be interested? The initial idea is to use Mac minis + mobile 5G data.
The advantage is clean IPs—genuine residential IPs, with each server node having a unique, non-shared IP. The downside is that it’s expensive... It would suit people who are sensitive about using US residential IPs, such as those subscribing to Claude or opening cross-border e-commerce stores...

gpt-astra’s AGI score improved significantly because its computer-use capabilities have been greatly enhanced (on macOS). Theoretically, anything you can do on a Mac can be completed by astra.

I’ve always felt that dense architectures are intellectually superior to MoE architectures, but they consume too much computing power during training and inference—it’s an economic consideration.

Based on product trends, the RTX 6090 should have been equipped with 48GB of VRAM, but given the current shortage, I estimate it will have at most 32GB.

Following AI coding, AI for science is finally starting to gain momentum. First came AI for Mathmatics, followed by AI for medicine—when will it be AI for physics’ turn?

h3 and qwen3.8-27b saved the AI animation and short-drama industry.

qwen3.8-27b has already brought local deployment up to the level of GLM-5.2/OPUS-4.6—the dense architecture is still more powerful. The best configuration for local deployment is still a modified 48GB 4090 (two cards; one isn’t enough).