Massive increase in GPU demand in China after V3/R1 became available on Chinese cloud providers and Huawei integration

@Misterbignest If inference demand approaches infinity then nobody will pay 75% GM to NVDA ASICs are the best for inference Bullish on Broadcom

@phill__1 pre training wall or google is incompetent

@phill__1 Still doesn't change the fact that 1206 was trained on at least 50X more compute than R1/V3

Each day passes and whale mogging becomes harder Gemini 1206 is trained on at least 50X more compute compared to v3/r1

@icebreakers__ @Jukanlosreve Every cloud provider in China is serving r1/v3 They should help DS with compute in return for serving their models on their platform

RT @icebreakers__: @angelusm0rt1s @Jukanlosreve The government won't give them hardware directly, but it is possible that the government le…

@Presidentlin 1000K GPUs still can't beat the little whale

@icebreakers__ @Jukanlosreve What if government gives them a free cluster?? Will they buy 910Cs??

@YouJiacheng The traffic is that high??

@DavidLiptonWI @blob_watcher They can impose heavy fines

RT @zizhpan: Our official huggingface space demo for DeepSeek-VL2 Small is out! A 16B MoE model for various vision-language tasks. Thanks @…

LFG!!! Apple might put it in phones too

@teortaxesTex @1woods_ Yet couldn't kill it and huawei is still doing over $100B in revenue and profit margins have increased and has created a complete ecosystem of software and hardware Deepseek getting entity listed will ensure massive amounts of funding and heroic treatment

@teortaxesTex I think that guy failed to run it becuz he used ollama Use MindIE from Huawei if u want to run models locally on Ascends

@teortaxesTex Hope it's Feb 8 if Whale bros had more compute they would have annihilated everybody by now

Bessent was long on Yuan