24-Hour AI News: Blackwell Enters Mass Production in the US, Agent Launches into Enterprise, and Chinese Applications and Investments Synchronize

In the past 24 hours, computing power and enterprise software in North America and Asia-Pacific have reached a new turning point, and Europe has continued to speak out on the implementation details of the AI bill; China's applications and investment and financing have progressed simultaneously, and medical AI, retail AI and education large-scale models have been promoted simultaneously.

1. Nvidia Blackwell begins mass production in Arizona, marking another step forward in localizing its supply chain

1. Updated on October 18th, NVIDIA announced that the Blackwell architecture GPU has entered mass production at TSMC's Arizona factory, with a node of 4 nanometers, which is an important milestone for its manufacturing in the United States.
2. Simultaneous layout of supporting links: Carry out packaging and testing cooperation with Amkor and SPIL in Arizona; Build two "supercomputer manufacturing plants" in Texas, and mass production is expected to begin in 12-15 months.
3. Performance and product matrix: Blackwell Ultra targets 15 PFLOPS-level AI computing power, supporting cost/performance optimization for enterprise-level reasoning and training.

Salesforce strengthens "Agentic Enterprise" and Agentforce 360 enters engineering phase

1. An analysis on October 18 pointed out that Salesforce will focus on upgrading from "demo to production" and build an operations layer with agent observability, governance, and traceability.
2. Propose the "Service as Software (SaSo)" paradigm: Based on the data cloud, unify semantics, policies and tool calls to reduce marginal costs and improve repeatable delivery capabilities.
3. Competitive landscape: Competing with ultra-large model platforms and other SaaS vendors for the "system intelligence layer" relies on end-to-end verifiable results and scalable stability.

3. Germany's Digital Industry Association Speaks Out on the Implementation of the AI Act: Concerns Over Regulatory Fragmentation and Insufficient Resources

1. On October 18, BVDW pointed out that the draft implementation plan of the German national AI law has problems such as multiple supervision and unclear authority boundaries.
2. Personnel and capacity building: By 2027, the planned expansion will only reach 129 positions, which the association believes is difficult to match the complex market supervision tasks.
3. Timeline background: The EU AI Act will take effect in stages, and the general-purpose AI (GPAI) obligations will be implemented from August 2025. Enterprises must complete the compliance path design as soon as possible.

Baidu Health launches "AI Butler": real-time doctor verification, positioning 24/7 health assistant

1. Released on October 19, the training and online process introduced large-scale labeling and secondary real-time verification by real doctors, claiming to improve the security and usability of medical AI.
2. Product coverage: dozens of functions including popular science Q&A, file management, medication consultation, AI pre-diagnosis, report interpretation and skin testing, covering the entire link from pre-diagnosis to post-diagnosis.
3. Usage: Aggregate on the mobile portal for convenient access to enhance the accessibility of public health management.

5. iFlytek plans to raise 4 billion yuan to increase its computing power base and education model; performance is under pressure in the short term.

1. On October 19, the company disclosed a plan for a private placement, aiming to raise no more than RMB 4 billion, which will be invested in the construction of a computing platform, the development of a large-scale model for Spark Education, and the replenishment of working capital.
2. The number of issuers shall not exceed 35, and the related parties of the controlling shareholder intend to subscribe for RMB 250 million to RMB 350 million to strengthen the certainty of strategic investment.
3. Financial progress on the same day: Net profit in the first three quarters was RMB 99.36 million, a year-on-year decrease of 76%; Q3 net profit was RMB 25.79 million, and R&D/development investment increased significantly year-on-year, reflecting the phased characteristics of "investment for momentum".

6. Double 11 Enters the Year of AI Acceleration: Shopping Guides, Recommendations, and Logistics Improve Efficiency at Multiple Levels

1. The platform claims that its computing power has increased 40 times year-on-year, and the perception of user behavior sequences has been extended from half a year to 10 years, leading to an increase in purchasing efficiency of approximately 25%.
2. New products in the shopping guide product matrix: Applications such as "AI Universal Search" for complex semantics are launched, covering the closed loop from demand understanding to decision-making recommendations.
3. Logistics: JD.com has invested in the "Super Brain + Wolf Clan" equipment cluster on a large scale for the first time, claiming that the first-line fulfillment efficiency has increased by nearly 20%, strengthening the supply-fulfillment coordination.

7. Jensen Huang confirmed his attendance at the APEC CEO Summit in South Korea from October 28 to 31, focusing on AI and robotics.

1. On October 19, Nvidia CEO will meet with executives from leading Korean companies to discuss industrial cooperation in areas such as AI, robotics, digital twins and autonomous driving.
2. The Asia-Pacific high-level economic and trade dialogue window will provide a multilateral scenario for AI supply chain, storage/high-bandwidth memory cooperation and ecological docking.
3. The clear time and place will help us observe the next step of coordination between the Korean storage and AI accelerator industry chains.

Frequently Asked Questions (Q&A)

Q: What direct impact does Blackwell's "mass production in the US" have on the supply and demand landscape?

A: In the short term, it will help Nvidia to diversify its production capacity risks and shorten part of the North American delivery chain; but the current node is N4, and the most advanced processes are still concentrated overseas. The overall supply improvement needs to be evaluated in conjunction with the ramp-up period (12-15 months) of the packaging and testing and complete machine assembly production lines.

Q: How will Germany’s BVDW’s concerns about the implementation of the AI bill affect your business in Europe?

A: Enterprises will face the challenges of interfacing with multiple entities and uncertain authority boundaries, leading to increased compliance and filing costs. It is recommended to proactively review GPAI transparency and data governance requirements, establishing standardized assets for model cards, data provenance, and risk assessments to accommodate the obligations that will take effect in phases from 2025 to 2027.

Q: What is the core security point of Baidu Health’s “AI Butler”?

A: We introduce "doctor annotation + real-time secondary verification" in the front-end and back-end links of the model, placing generated recommendations under doctor supervision and mitigating the risk of misdiagnosis in AI-based consultations alone. This is a tool for health management and triage, not a replacement for clinical diagnosis.

Q: Is it contradictory that iFlytek is raising funds while its profits are declining?

A: No contradiction. Large models and computing power are capital expenditure- and R&D-intensive sectors, and financially, "investment first, profits lag." The key lies in whether improvements in computing power and algorithm efficiency within the fundraising projects can create a closed loop in terms of ARPU and penetration within the education and industry models.