China Merchants Securities International: Zhi Spectrum (02513) completed strategic financing of approximately US$5 billion to maintain an “increase in holdings” rating and target price of HK$1,600

The Zhitong Finance App learned that China Merchants Securities International released a research report saying that Zhi Spectrum (02513) recently announced that it has completed strategic financing of about 5 billion US dollars, mainly to invest in next-generation GLM basic model research and development, fully self-training systems, and computing power infrastructure. The bank maintained its “Overweight” rating, with a target price of HK$1,600.

The bank quoted the company's management exchange meeting on September 16 as saying that after this expansion, computing power has begun to take shape, and short-term computing power supply is no longer the main constraint on revenue growth; the company has signed revenue sharing agreements with leading domestic and foreign cloud service providers, and related revenue has been confirmed since October; the year-end ARR guidance was raised from 2.4 billion US dollars to 3 billion US dollars. Currently, the full business ARR (monthly basis) has reached 1.8 billion US dollars, a further increase from 1.6 billion yuan two weeks ago; the cumulative order amount within one month of the launch of the first co-work scenario has already exceeded 1 billion yuan.

The bank pointed out that this round of financing anchors computing power infrastructure and a completely self-training system. Combined with net cash of about RMB 28.5 billion at the end of the first half of this year, computing power reserves have increased significantly. The additional computing power portion can theoretically support more than 6 billion US dollars in inference revenue. The actual situation depends on utilization rates and discount rates. The computing power multiplier increased 14 times year-on-year in the first half of this year; since domestic chips were included in the main inference computing power at the beginning of the year, the inference cost per token has dropped 80% compared to the beginning of the year.