24-hour AI news: Microsoft's self-developed model debuted, Alibaba Cloud's AI growth rate is eye-catching

24-hour AI news at a glance: Microsoft's self-developed model unveiled, Alibaba Cloud AI growth rate leads; In the

past 24 hours, overseas, Microsoft has released two self-developed models, Anthropic has adjusted consumer data for training, Japan has launched Mount Fuji AI ash reduction simulation, Dell is under pressure on the AI server track, and South Korea plans to invest in AI research and development; Domestically, Baidu's intelligent cloud upgrade Baige 5.0 and Qianfan 4.0, Alibaba's financial report disclosed that the cloud business has grown strongly driven by AI, and the central control technology has released TPT 2, a trusted large model for the process industry, and formed an alliance, and the capital market and data element construction have strengthened simultaneously. The deduplication check of the same topic event on the designated channel in the past 72 hours has been completed, and only the topics with new progress have been included.

1. Microsoft released two self-developed models, MAI-Voice-1 and MAI-1-preview

1. Released on August 29: MAI-Voice-1 is a high-performance speech generation model that can "generate 1 minute of speech in 1 second" on a single-card GPU; MAI-1-preview is a large language model and has started public testing.
2. Application and testing: MAI-Voice-1 has been connected to Copilot Daily/Podcasts, and MAI-1-preview is open for evaluation on LMArena.
3. Significance: It marks the formation of Microsoft's self-developed model system, and while continuing to cooperate with OpenAI, it will improve its self-sufficiency and product diversity.

2. Anthropic announced that consumer chat records can be used for training (can opt out)

1. Time and scope: Effective from September 28 for consumer subscriptions such as Claude Free/Pro/Max and Claude Code, excluding commercial scenarios such as enterprise/government affairs and APIs.
2. Data policy: The chat and coding sessions of consent users can be stored for up to 5 years, and can be changed at any time in privacy settings, but the data used for training cannot be retroactively withdrawn.
3. Impact: Focus on the game of privacy and transparency, or promote the industry baseline of "default join/exit" and interface prompt design.

3. Japan released a simulation video of Mount Fuji AI ash falling for public disaster prevention education

1. Subject and purpose of release: The Tokyo Metropolitan Government and the Cabinet Office have successively launched generative AI videos, suggesting that volcanic ash may arrive in Tokyo within about 2 hours and impact transportation, power and material supply.
2. Status description: The official emphasizes that there is currently no sign of an "imminent eruption", which is a normalized disaster prevention drill.
3. Significance: AI content production and urban emergency communication are deeply integrated to enhance the public's intuitive cognition and willingness to reserve materials.

4. India's Reliance establishes a subsidiary of "Reliance Intelligence" to focus on AI

1. At the annual meeting on August 29, it was announced that it would establish a new wholly-owned subsidiary, positioning itself as the starting point for the transformation of "Deep-Tech+AI".
2. Industrial collaboration: Deepen cooperation with Google and Meta to strengthen ecology and distribution capabilities.
3. Significance: India's local giants are building a bridge between AI infrastructure and the consumer Internet, which is expected to drive the implementation of local scenarios.

5. Dell: High costs and competition suppress profits, still raising annual shipment expectations for AI servers

1. Market performance: The intraday stock price fell by about 10% on August 29.
2. Business guidance: The annual shipment forecast for AI servers was raised from $15 billion to $20 billion, while the adjusted gross profit margin in the second quarter fell to 18.7%.
3. Industry profile: Supply chain acceleration and price competition squeeze profits, and head customers (including xAI and CoreWeave) drive demand to continue to be high.

6. South Korea plans to significantly increase AI-related R&D expenditure in the 2026 budget

1. Draft budget: Total government expenditure is planned to be +8.1% year-on-year, of which scientific research expenditure will be a record +19.3%, focusing on AI.
2. Policy orientation: Hedging external tariffs and endogenous growth pressures with expansionary fiscal policy to promote AI-led structural transformation.
3. Regional significance: East Asia's AI investment continues to increase, and regional competition for computing power, talents and standards will become more intense.

7. Baidu Intelligent Cloud upgrades Baige 5.0 and Qianfan 4.0, and the training/inference system is fully accelerated

1. Platform upgrade: VPC/RDMA/X-link three networks and training and promotion integrated systems are synchronously strengthened, and Kunlun Xinchao nodes are launched.
2. Performance indicators: Through operator decoupling and adaptive parallelism optimizations, DeepSeek R1 inference throughput is increased by about 50%.
3. Industrial significance: Engineering delivery for enterprise-level AI, reducing the cost and delay of large model inference.

8. Alibaba's financial report: Cloud business revenue +26% year-on-year to 33.4 billion yuan driven by AI

1. Time and caliber: Cloud business revenue for the quarter ended June 30 was 33.4 billion yuan, +26% year-on-year.
2. Investment intensity: The cumulative AI-related investment in the past four quarters has exceeded 100 billion yuan.
3. Structural changes: The increase means e-commerce and AI infrastructure, short-term profits are under pressure but the momentum of cloud business is strong.

9. Central Control Technology released TPT 2, a trusted large model for process industry, and established the "Industrial AI Data Alliance"

1. Release time: The global new product launch conference was officially announced on August 28.
2. Model positioning: Trusted AI models and industrial agent platforms for time series data and process industry, emphasizing closed-loop controllability and auditability.
3. Industrial significance: The large-scale modeling of in-depth scenarios in the manufacturing industry has accelerated, and industry standards and data have collaborated towards platformization.

10. Semi-annual report of Unigroup Co., Ltd.: Launched UniPoD S80000 super node, significantly improved training and promotion efficiency

1. Financial overview: Revenue in the first half of the year was 47.42 billion yuan, and net profit after deducting non-attributable to the parent company was 1.12 billion yuan, both about +25% year-on-year.
2. Product highlights: The 64-card single-cabinet supernode is networked with +25% training efficiency and +63% inference efficiency compared with 8×8-card servers.
3. Implementation progress: It has been deployed in multiple projects and can evolve to 1024 cards on demand to serve trillion-level parameter-level models.

11. Industrial Fulian: AI servers drive high growth, with a market value exceeding 1 trillion yuan

1. Performance impulse: single-quarter revenue in the second quarter exceeded 200 billion yuan for the first time; In the first half of the year, AI server revenue increased by more than 60% year-on-year.
2. Chain spillover: optical modules and complete machine supporting enterprises are synchronized with high prosperity, and the "network side" of the industrial chain has become the core of new additions.
3. Market profile: The market value of A-share AI hardware chain leaders jumped, reflecting the structural preference for computing power investment.

12. Baidu AI Coding and "Second Match" no-code platform accelerate the implementation

1. Product trends: AI coding and low/no-code tools are expanded for non-professional developers to shorten the application delivery cycle.
2. Ecological expansion: Combine with enterprise knowledge base and process orchestration to promote the large-scale use of "AI workflow".
3. Application significance: Lower the threshold for AI application and promote the implementation of lightweight agents in multiple industries.

13. Progress of data elements: national high-quality dataAccording to the collection, the cumulative transaction volume has reached 4 billion yuan

1. Authoritative report: The National Data Administration disclosed that up to now, the cumulative transaction volume of high-quality data sets in the country is about 4 billion yuan.
2. Supply capacity: more than 35,000 high-quality data sets have been built in various places, with an overall volume of more than 300PB; Chinese training data accounts for 60%-80% of most models.
3. Mechanism innovation: The Beijing Digital Exchange accounts for nearly 80% of high-quality dataset transactions, and many places have piloted "data corpus price shares".

Frequently Asked Questions (Q&A)

Q: What are the key highlights of Microsoft's self-developed model release?

A: First, the extreme performance of the voice model MAI-Voice-1 on a single card GPU ("1 second generates 1 minute of speech" level); The second is that MAI-1-preview will start public evaluation and will enter Copilot in stages, showing that Microsoft is completing its self-developed models and inference stack while maintaining cooperation with OpenAI.

Q: Why did Alibaba Cloud's business outperform overall revenue this quarter?

A: AI pulled significantly, with cloud revenue +26% year-on-year to 33.4 billion yuan; At the same time, the company disclosed that the cumulative investment in AI-related products in the past four quarters exceeded 100 billion yuan, and capital expenditure and product upgrades jointly drove cloud-side growth, but it put pressure on profit margins in the short term.

Q: Does the AI ash reduction simulation of Mount Fuji in Japan mean that the risk is increased?

A: No. The official is clearly a disaster prevention science popularization content, and there is currently no "imminent eruption" warning; However, the video gives a visual scene of "volcanic ash arriving in Tokyo in about 2 hours", which helps the public to stock up on supplies and formulate travel/emergency plans.

Q: What is the latest focus of "intelligent computing + data elements" in China?

A: On the one end, the computing power and platform (such as Baige 5.0/Qianfan 4.0, UniPoD super node) improve training and delivery efficiency; On the other hand, the large-scale supply and trading mechanism of high-quality data sets (cumulative transaction volume of about 4 billion yuan and total volume of more than 300PB) forms a closed loop of "data-model-application" for the industry.