GO
| HSI1 | 24,917.60 | +111.97 | 192.63B |
| HSCEI1 | 8,284.58 | +38.25 | 51.08B |
| Back Zoom + Zoom - Block Traded | |
|
2026-09-11 15:23:50 OpenAI's GPT-6 Astra marks a new phase for frontier AI, demonstrating leading capabilities in computer usage, scientific research and professional work, CLSA said in a research report. Beyond improvements in benchmark testing, the shortening model iteration cycle is increasing the importance of product experience and brand premium. Meanwhile, rising training compute demand is reinforcing the scale advantages of leading frontier laboratories. In the broker's view, these trends will expand the AI total addressable market and accelerate progress toward more autonomous research systems and ultimately artificial superintelligence (ASI). Training compute intensity has once again become an investor focus. NVIDIA Corporation (NVDA.US) founder and CEO Jensen Huang disclosed that Astra was trained using more than 100,000 Grace Blackwell chips, while the next-generation training cluster accelerator could exceed 400,000 units, highlighting the capital-intensive nature of frontier AI. Based on the broker's estimates, calculated by revenue relative to inference and training compute costs, Anthropic's compute multiple will reach 1.8x in 2Q26 and the company has already achieved profitability, leading globally, while Chinese peers are at around 0.4x. US peers have already entered competition involving more than 10 trillion parameters, while Chinese laboratories remain at the 3 trillion parameter level. The key challenge in narrowing the capability gap lies in compute supply constraints, partially offset by greater emphasis on post-training, infrastructure efficiency and test-time compute. ~ AASTOCKS Financial News Website: www.aastocks.com | |