Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.
CXMT doing only 100K HBM wafer starts/month by 2028 might be unrealistic. Moreover, as I often say, there's not a world of difference between fabbing DRAM and "HBM"… anyway, they did a lot more homework this time. @zephyr_z9 https://ai-2040.com/supplements/compute-supplement#differences-with-nesovs-numbers https://x.com/teortaxesTex/status/2075419601563681025/photo/1
Interesting detail in AI-2040 supplementary: Despite the softened rhetoric on Chyna, it is grotesquely outclassed on compute, indeed even worse than in AI-2027. Domestic additions are on the order of 5.2M H100-equivalents by 2028. They realize that logic is no issue. HBM, tho… https://x.com/teortaxesTex/status/2075418983486927237/photo/1
CXMT doing only 100K HBM wafer starts/month by 2028 might be unrealistic. Moreover, as I often say, there's not a world of difference between fabbing DRAM and "HBM"… anyway, they did a lot more homework this time. @zephyr_z9 https://ai-2040.com/supplements/compute-supplement#differences-with-nesovs-numbers https://x.com/teortaxesTex/status/2075419601563681025/photo/1
Guardrails removed spam, off-topic, unclear, or duplicate replies.
Ask a question below.
Published answers will appear here.