Hong Kong bourse demonstrated feeble rebound momentum this morning (5th) in lack of support from southbound funds. After opening down 8 pts, the HSI fluctuated throughout the session and closed at 24,040, up 68 pts or 0.28%, with full-day turnover at only HKD98.102 billion. The HSCEI closed at 8,051, up 21 pts or 0.26%. The HSTECH performed relatively better, closing at 4,183, up 25 pts or 0.62%.

Major techs jumbled. TENCENT (00700.HK) added 0.43%; BABA-W (09988.HK) gained nearly 1%; MEITUAN-W (03690.HK) hiked 0.64%; and XIAOMI-W (01810.HK) dropped 1.49% and JD-SW (09618.HK) dipped 0.49%.

Among blue chips, LENOVO GROUP (00992.HK) soared 4.66%, making it the best-performing blue chip. BEONE MEDICINES (06160.HK) mounted 3.02%. BUD APAC (01876.HK) lost 1.15% after the company expected to record a non-underlying withholding tax expense of USD52 million from internal restructuring in 3Q. MTR CORPORATION (00066.HK) plunged 3.41%, making it the worst-performing blue chip, after reaching a HKD2.25 billion settlement agreement with the government regarding the Hong Kong section of the high-speed rail and the Shatin to Central Link. HSBC HOLDINGS (00005.HK) rebounded 0.27% to HKD149.9, while AIA (01299.HK) sagged another 0.36%.

Goldman Sachs upgraded Z.AI (02513.HK) to Buy, with the stock closing up 6.15%. MINIMAX-W (00100.HK) lifted 0.58%, while XUNCE (03317.HK) dipped 0.33%.

Severe shortages of upstream materials triggered by surging demand for AI servers and high-speed networking have driven the PCB industry into a new upcycle. KB LAMINATES (01888.HK) leaped 11.94%; KINGBOARD HLDG (00148.HK) jumped up 8.36%; DELTON (01989.HK) rallied 6.86%; KINWONG (03228.HK) surged 9.21%; and VGT (02476.HK) elevated 5.69%.

NVIDIA (NVDA.US) set a record high last Friday, shoring up Hong Kong-listed chip stocks. SMIC (00981.HK) advanced 1.74%; HUA HONG GRACE (01347.HK) jumped up 4.27%; ILUVATAR COREX (09903.HK) swelled 3.94%; and MONTAGE TECH (06809.HK) escalated 5.25%.
~

AASTOCKS Financial News
Website: www.aastocks.com

AI startup Reflection, backed by NVIDIA Corporation (NVDA.US), is preparing to launch a powerful open-weight AI model to wrestle with Chinese AI companies such as DeepSeek as well as major US AI players, AXIOS, citing sources, reported.

The model is said to possess strong intelligence capabilities that could help enterprises build their own proprietary and low-cost AI systems.

Reflection will integrate with Nvidia GPUs to provide individuals and enterprises with a new and lower-cost alternative AI model.

Although the model is expected to lag behind the most advanced US AI models in the initial stage, it is expected to compete with China's leading open-weight models.

Reflection has purportedly been in discussions with parties in Washington and elsewhere over recent weeks to introduce details of its model ahead of the launch.

The company is also actively expanding computing capacity and recently signed major agreements with Nebius (NBIS.US) and SpaceX (SPCX.US) to lease Nvidia AI servers.
~

AASTOCKS Financial News
Website: www.aastocks.com

TSMC (TSM.US) is discussing cooperation plans with Elon Musk regarding the Terafab project, Culpium reported.

In addition to its existing operations in Arizona, TSMC is also planning investments in Texas, recent reports suggested.

The market believed the potential cooperation may be tied to the investment, with expectations that TSMC may seek to acquire and operate the Terafab factory.

Responding to the possibility of cooperation with TSMC, Musk said that it is only discussions at present, but something may come out of it. TSMC has yet to respond to the related reports.
~

AASTOCKS Financial News
Website: www.aastocks.com

SpaceX (SPCX.US) and NVIDIA (NVDA.US) jointly validated that AI has entered the era of Super Intelligence (SI), CLSA said in a research report, reiterating its Outperform rating on SpaceX with a target price of USD250.

The broker said discussions between Elon Musk and Jensen Huang at activities related to the US government were highly consistent with its initiation report on SpaceX and its 2024 views on the nuclear power industry.

Musk mentioned that China's power generation output is about three times that of the US, and stressed the need to ensure that the US has sufficient manufacturing capacity for logic and memory chips.

CLSA described SpaceX as the world's most technologically advanced enterprise group, possessing leading capabilities and market share across various sectors.
~

AASTOCKS Financial News
Website: www.aastocks.com


This article was automatically translated by AI, the Chinese version should be considered the authoritative version. AASTOCKS.com Limited does not guarantee its accuracy or completeness and accepts no liability for any damages or losses arising from the use of this translation.

The impact of AI open-source weight models on frontier model revenue has passed a local peak, as the capability gap has widened again and usage economics are increasingly favoring the model portfolios of frontier laboratories, Citi said in a research report. Although open-source models once markedly narrowed the gap with proprietary frontier models, much of the narrowing occurred when regulatory concerns constrained the release of advanced models, while open-source developers benefited from distillation and advanced overseas computing power. As mentioned by Poolside Co-CEO Jason Warner, in the long run, the value of open-source models lies more in the control brought by customization, deployment flexibility and intellectual property protection than in pure cost advantages.

Citi cited the latest weekly data that the highest Artificial Analysis Intelligence Index score for proprietary models rose from 53 to 58, while the highest score for open-source models increased from 44 to 46, widening the gap from 9 points to 12 points. Public benchmarks may still overestimate the degree of parity between the two, while private evaluations show that leading open-source systems lag further behind in long-horizon cybersecurity tasks and production reliability. Anthropic's Opus 5.5 improved the AA intelligence frontier by 9%, while reducing costs by 40% versus Opus 5 under typical workloads. Subsequently released Grok and Gemini models also improved intelligence while maintaining broadly stable token pricing, highlighting ongoing improvements in model efficiency.

The broker expected enterprises to adopt hybrid model portfolios as a de-risking mechanism, while AI laboratories are likely to increasingly embed routing, orchestration, governance and security capabilities into the model layer, making users specify individual models only in rare or highly specific scenarios. Differences in token efficiency and accuracy imply that workload-based routing is often more economical than uniformly adopting either open-source or closed deployments. NVIDIA Corporation (NVDA.US) acquired Hugging Face for USD12.9 billion and reached a USD6 billion licensing agreement with Poolside, enabling it to strengthen its open-source weight strategy spanning hardware, inference software, CUDA and Nemotron through model discovery, collaboration, deployment platforms and open-source weight development capabilities, making it one of the major beneficiaries of improvements and adoption in the open-source ecosystem.
~

AASTOCKS Financial News
Website: www.aastocks.com

OpenAI said it dismissed three researchers for allegedly improperly handling "sensitive information" and violating company policies. The incident also involved an external individual responsible for evaluating AI models.

At least two of the dismissed employees were reportedly engaged in AI Safety and Alignment research, which refers to ensuring that AI behavior and objectives remain aligned with human values, intentions and ethical standards.

The dismissals come amid intense public debate over AI safety issues and whether related technologies could pose an existential threat to humanity.

According to The Wall Street Journal, the three dismissed researchers were Jasmine Wang, Tomek Korbak and Mikita Balesni. They were accused of disclosing confidential company information to a third-party AI safety organization, while also frequently posting about AI safety topics on the X platform in recent months.
~

AASTOCKS Financial News
Website: www.aastocks.com

During IF26, investor interest in the global humanoid robotics industry remained strong, but discussions have clearly shifted from understanding the technology/industry to evaluating commercialization, value capture and investment opportunities, CLSA said in a research report. The broker maintained a highly constructive view on the long-term prospects of the humanoid robotics industry.

Investors are increasingly focused on identifying potential winners, assessing supply chain beneficiaries, and monitoring key catalysts such as the roadmap for Tesla, Inc. (TSLA.US) Optimus Gen3 and the industry's move toward mass production. Discussions are also placing greater emphasis on the commercialization timelines, production capacity ramp-up, and real-world deployment cases of leading US and Chinese developers, rather than debating whether the technology itself is feasible.

The broker said market expectations for catalysts are broadly in line with its views, including the listing of YUSHU TECHNOLOGY (688836.SH), the Tesla's Optimus Gen3 roadmap, and future funding rounds for leading private developers. More importantly, the transition from pilot deployments to mass production over the next 3-18 months may be the most important catalyst for boosting market confidence in commercialization.
~

AASTOCKS Financial News
Website: www.aastocks.com

Chinese automakers are ambitiously expanding exports amid meager domestic sales and competition from more than 100 domestic rivals, CLSA said in a research report. At the CITIC investor forum, CLSA met with investors and mainly discussed how Japanese automakers can respond to Chinese competition outside China, the acceptance of Chinese imported vehicles across different countries, and how automakers operate under low-growth industry conditions and geopolitical pressure.

The broker noted that markets most vulnerable to penetration by Chinese automakers are countries without domestic automobile industries, such as Australia and Singapore, where Chinese automakers already have relatively high market share. This is followed by markets with domestic automobile industries, such as the US, the European Union and India, which have established high tariff barriers, while the EU Industrial Acceleration Act also poses risks.

CLSA's top picks in the global auto sector are Nissan Motor (7201.JP), Suzuki (7269.JP) and General Motors Company (GM.US). Nissan was rated High Conviction Outperform with a TP of JPY600, supported by cost reductions, attractive hybrid model launches and higher utilization at US plants. Suzuki was rated Outperform with a TP of JPY2,300, as the broker believed high growth in its India business could drive earnings above expectations.

General Motors Company (GM.US) was rated Outperform with a TP of USD105, with nearshoring production and new growth businesses expected to support re-rating. The broker rated Tesla, Inc. (TSLA.US) Underperform with a TP of USD320, while Toyota (7203.JP) was rated Outperform with a TP of JPY3,700.
~

AASTOCKS Financial News
Website: www.aastocks.com

Advanced Micro Devices, Inc. (AMD.US) CEO Lisa Su is expected to visit South Korea next month to meet with executives of South Korean semiconductor companies and government officials, South Korean media reported.

This will mark Lisa Su's second visit to S Korea in about seven months following her trip in March this year. She is considering attending a conference in Seoul on October 8, where participants are expected to include CEOs of major South Korean AI semiconductor companies.

She may also meet with South Korean Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon during the event.

The report added that Su is also arranging meetings with executives of major South Korean companies, including Samsung Electronics.
~

AASTOCKS Financial News
Website: www.aastocks.com

Meta Platforms, Inc. (META.US) announced the establishment of Meta Enterprise Platform, a new business unit focused on bringing the company's AI capabilities to enterprise clients and developers, BofA Securities said in a report. CEO Mark Zuckerberg described this as the next major pillar of Meta's business. The broker maintained its Buy rating and USD810 TP on Meta.

The broker noted that the initial products include Muse agent, Meta Business Agent, Muse API and Muse Code. The company also hired MongoDB CEO Chirantan CJ Desai as Chief Enterprise Platform Officer, and the broker expects him to focus on driving product sales.

The new platform is not entirely unexpected, said the broker. Meta's launch of enterprise AI solutions could generate more sustainable monetization revenue that is less affected by macro conditions. The enterprise AI solutions market is expected to exceed USD1 trillion by 2028, while Meta's consumer data and communication applications may help it sell AI products to consumer-oriented enterprises.
~

AASTOCKS Financial News
Website: www.aastocks.com