PATTONPatton Harga

On August 27, Charu Chanana, Chief Investment Strategist at Saxo Markets, stated that NVIDIA's (NVDA.O) performance strongly validates the investment logic behind artificial intelligence infrastructure. She pointed out that as capacity continues to expand, sectors such as storage, networking, optical communications, power infrastructure, and some cloud service providers will benefit. However, the market increasingly needs to see returns from AI investments. Storage remains one of the clearest beneficiary areas, and NVIDIA's profit margin guidance confirms that supply remains tight. Following NVIDIA's earnings announcement, Asian storage chip stocks generally rose. However, Chanana added that the next phase of the storage chip market will increasingly depend on AI demand and whether the storage capacity of each accelerator can absorb the continuously increasing supply capacity, rather than relying solely on supply shortages to drive price increases.

On August 27, Charu Chanana, Chief Investment Strategist at Saxo Markets, stated that NVIDIA's (NVDA.O) performance strongly validates the investment logic in artificial intelligence infrastructure. She pointed out that as production capacity continues to expand, sectors such as storage, networking, optical communication, power infrastructure, and some cloud service providers will benefit. However, the market increasingly needs to see returns from AI investments. Storage remains one of the clearest beneficiary areas, and NVIDIA's profit margin guidance confirms that supply remains tight. Following NVIDIA's earnings announcement, Asian storage chip stocks generally rose. However, Chanana added that the next phase of the storage chip market will increasingly depend on AI demand and whether the storage capacity of each accelerator can absorb the growing supply capacity, rather than solely relying on supply shortages to drive price increases.

Zhiyu's flagship model GLM-5.3 will be open-sourced at 10 AM Beijing time on August 28. In fact, GLM-5.3 was already launched on Coding Plan on August 14, followed by the opening of its API. Tomorrow, the model weights will finally be available for download, which has been delayed by two weeks due to special reasons. Zhiyu stated that during the post-training process, the network security capabilities of GLM-5.3 grew at an 'unexpected' rate, not only becoming better at finding vulnerabilities but also learning to plan complete exploit chains across multiple steps. Therefore, the official team conducted a security assessment and reinforcement before releasing the weights. In official evaluations, GLM-5.3 scored 84.5% in the vulnerability discovery test Cyber Gym, slightly higher than Mythos 5's 83.8% and GPT-5.6Sol's 83.6%. In the more attack-oriented Exploit Bench, its score improved from 24.4% in GLM-5.2 to 54.4%, although it still lags behind Mythos5's 78.0%. Zhiyu also reported that the model has discovered 2,436 vulnerabilities across 269 open-source projects, of which 1,097 are classified as severe or high-risk. GLM-5.3 uses the same base model as GLM-5.2, with all improvements coming from post-training.