ZH version is available. Content is displayed in original English for accuracy.
Advertisement
Advertisement
⚡ Community Insights
Discussion Sentiment
50% Positive
Analyzed from 208 words in the discussion.
Trending Topics
#memory#ram#servers#gains#oversold#companies#chips#meta#https#zero

Discussion (2 Comments)Read Original on HackerNews
”Still, the analysts said, AI investors are taking risks so significant that much of the financial system now rests upon AI meeting expectations for productivity gains and profitability.”
The economics of computing will need to change in order to sustainably satisfy demand for AI services.
If there's going to be a crash, can the AI companies even liquidate the chips they bought? At mil-grade, they cannot.
Will there then be a glut of RAM instead of the current perfect storm of geopolitical stressors and alleged collusion?
Meta recently posted a solution for reusing decommissioned RAM with CXL and Vistara; "Meta reuses old RAM in new servers with custom bridge chip" https://news.ycombinator.com/item?id=48778956 .. https://aisystemcodesign.github.io/papers/isca26/vistara_cam... :
> to newer servers as expanded memory alongside servers’ local memory (e.g., DDR5 DIMMs). This approach offers a compelling combination of benefits: near zero-cost memory expansion through recycling, performance gains from higher memory capacity, and a reduced carbon footprint.
Circularly recyclable chips would increase the liquidation value of existing AI datacenters.
AI datacenters are angering people; cooling tower wash in the paltry remaining water pressure, loud hum, emissions from generators, steam for dumb cooling that's not zero water for example.
When the malarky hits the fan for some of these oversold AI companies, will there be constraints defining what is an acceptable AI datacenter?