Morgan Stanley released a research report saying that Smart Spectrum's annualized recurring revenue had reached 1.8 billion US dollars as of mid-September, and raised the ARR guidelines from 2.4 billion US dollars to 3 billion US dollars at the end of the year, reaffirming the “increase in wealth” rating, with a target price of HK$1,800. The bank pointed out that the management framework shows that the financing of about 5 billion US dollars can support about 94,000 GPUs, of which about 40% is used for training and about 60% for inference; about 56,000 inference GPUs mean that the theoretical annual revenue limit is about 41 billion yuan. The maximum theoretical gross margin can reach about 80%. Alleviation of computing power bottlenecks is expected to open up the next phase of growth. Cloud cooperation can also improve capital efficiency and global distribution, and revenue sharing is expected to contribute from October. On the R&D side, the distribution of the training portion can support a cumulative annual training expenditure of about RMB 3 billion over four years, which is enough to support multiple GLM-5.3 level or larger training.

Morgan Stanley released a research report saying that Smart Spectrum's annualized recurring revenue had reached 1.8 billion US dollars as of mid-September, and raised the ARR guidelines from 2.4 billion US dollars to 3 billion US dollars at the end of the year, reaffirming the “increase in wealth” rating, with a target price of HK$1,800. The bank pointed out that the management framework shows that the financing of about 5 billion US dollars can support about 94,000 GPUs, of which about 40% is used for training and about 60% for inference; about 56,000 inference GPUs mean that the theoretical annual revenue limit is about 41 billion yuan. The maximum theoretical gross margin can reach about 80%. Alleviation of computing power bottlenecks is expected to open up the next phase of growth. Cloud cooperation can also improve capital efficiency and global distribution, and revenue sharing is expected to contribute from October. On the R&D side, the distribution of the training portion can support a cumulative annual training expenditure of about RMB 3 billion over four years, which is enough to support multiple GLM-5.3 level or larger training.