24-hour AI news: SoftBank x OpenAI joint venture in Japan, Gemini takes over Maps, domestic projects launched through a competitive bidding process.

In the past 24 hours (November 4-5), on the Chinese side, there were developments such as the release of annual AI innovation task guidelines by ministries, updates to industry large-scale models, and deepening of "existing power" applications; on the overseas side, SoftBank and OpenAI established a joint venture in Japan, Google deeply embedded Gemini into Maps, and the global market saw fluctuations surrounding the "AI bubble" and controversies over benchmarking methodologies.

1. SoftBank and OpenAI establish "SB OAI Japan" joint venture to provide enterprise-level AI services.

1. On November 5, it was announced that "SB OAI Japan" has been established in Japan, which is positioned to provide localized enterprise-level AI solutions and operation services for large Japanese companies.
2. The first batch of customers were launched from within the SoftBank group, and the roadmap covers the entire chain of delivery from enterprise data access to customized intelligent agents.
3. It is being promoted in parallel with the plans for data centers and ecosystem partners in Japan, and is regarded as an accelerator for the large-scale application of AI by Japanese companies.

II. Google integrates Gemini into Google Maps, enabling more "conversational" navigation and assistant capabilities.

1. Starting November 5th, it will be gradually launched, supporting route planning, location Q&A, and multi-task commands using natural language.
2. In parallel with Gemini's integration in scenarios such as YouTube/Search, it forms a unified AI interaction portal across applications.
3. Industry insiders interpret this as the further development of the "AI native operation layer" on mobile devices, strengthening its competitiveness against third-party assistants.

III. Global tech stocks were pressured by concerns about an "AI valuation bubble," leading to increased volatility in the chip and computing power supply chains.

1. On November 5, stock indices in many regions retreated, with technology and AI sectors leading the decline, as the market focused on the uncertainty of the return cycle of computing power capital expenditure.
2. Some leading stocks are under short-term pressure, and analysts warn of risks related to "concentrated holdings + earnings realization pace".
3. Short-term fluctuations do not change the long-term investment theme of AI infrastructure, but funds are more inclined to invest in targets with clear commercialization progress and cash flow.

IV. The research team pointed out that hundreds of AI safety and performance benchmarks have methodological flaws.

1. A joint study released on November 4-5 reviewed 440+ benchmarks and pointed out issues such as coverage, leaked questions/overfitting, and evaluation consistency.
2. It is recommended to shift from static question-and-answer to task-based, multimodal, and adversarial evaluation, and to establish an open and reproducible experimental pipeline.
3. Issue warnings about the "score-based meritocracy" in regulatory, bidding, and academic competition scenarios, and call for the industry to jointly establish new benchmarks.

V. The Ministry of Industry and Information Technology released the 2025 Guidelines for the "Challenge-Based Approach" to Promote the Artificial Intelligence Industry and Empower New Industrialization.

1. On November 5th, the annual task list was released to various regions, covering key technologies for large-scale models, industrial software and intelligent manufacturing.
2. Emphasize "enterprises posing the questions and heroes taking on the challenges" to promote collaborative research and development between industry, academia, and research institutions and the implementation of related applications.
3. Define clear assessment indicators for the deep integration of informatization and industrialization, and promote the large-scale application of AI in key industries.

VI. "Storage Power China Tour" Beijing Station: Advanced storage computing power supports the optimization of efficiency and cost for large model inference.

1. On November 5th, the China Academy of Information and Communications Technology (CAICT) hosted an event focusing on the critical role of "storage power" in the reasoning process in the AI era.
2. Under the constraints of inference cost, latency and quality, a new architecture path with high bandwidth, low latency and hierarchical storage is proposed.
3. It integrates with application scenarios such as finance, customer service, and medical imaging, emphasizing the system collaboration between data and models.

7. The "Intelligent Industrial Model 3.0" was released in Dalian, deeply integrating the industry-wide model into the production system.

1. Version 3.0 was released at the Petrochemical and Chemical Industry Digitalization Conference on November 4-5, focusing on process optimization, safety, environmental and quality control, and predictive maintenance of equipment.
2. Overcome pain points such as industry data silos and fragmented standards, and emphasize the combination of cross-site data governance and knowledge graphs.
3. This marks a new stage in the large-scale model development of the process industry, moving from pilot verification to system-level deployment.

8. Experts at the Hongqiao Forum proposed three principles for "AI for Good": negative list, openness and transparency, and a combination of prevention and crackdown.

1. On November 5, it was proposed to establish an enforceable negative list under the framework of centralized governance by administration and law.
2. The platform is required to meet certain transparency standards to facilitate social supervision and scientific evaluation.
3. It proposes to correct the imbalance between "excessive law enforcement" and "ignoring illusions," and advocates rational governance.

Frequently Asked Questions (Q&A)

Q: What will SoftBank and OpenAI’s joint venture in Japan achieve in the short term?

A: We initially served SoftBank and large Japanese companies with enterprise-level intelligent agents and data platform projects. Starting in 2026, we will cooperate with the expansion of local data centers to promote large-scale delivery, focusing on localization compliance and industry scenario adaptation.

Q: Google has put Gemini into Maps. What is the fundamental difference between this and previous voice navigation?

A: The shift from "command-based" to "conversational" allows for the expression of multiple constraints (road conditions, waypoints, parking, etc.) in natural language at once, and enables cross-application information retrieval, forming a more unified AI interaction layer.

Q: Will the so-called "flaws in AI benchmarks" affect corporate procurement and oversight?

A: Yes. Research suggests shifting from a single score to multidimensional assessment and reproducible experimental pipelines. Bidding and regulation may increase task-oriented, adversarial, and long-term consistency test items, weakening the "score-only" approach.

Q: What does the "open competition for key projects" approach mean for the industrial chain in China?

A: By directing resources toward bottlenecks and quantifiable scenarios, companies can commit to technologies and indicators according to the list and obtain policy and financial support, thus shortening the cycle from technological breakthroughs to industry implementation.

Q: Why is "storage capacity" frequently mentioned in the era of large-scale models?

A: The bottleneck in the inference stage is shifting from "pure computing power" to "storage bandwidth/latency + data orchestration". Efficient tiered and near-memory architectures can significantly reduce latency and cost under the same hardware conditions.

Q: Will market concerns about an "AI bubble" change the long-term trend?

A: Short-term fluctuations mainly stem from the pace of earnings realization and the uncertainty of capital expenditure returns, but in the long run, applications/infrastructure with verifiable ROI and clear cash flow are still viewed favorably.