24-Hour AI News: Oracle Secures Major Meta Cloud Order, OpenAI Adds Backup Computing Power, Domestic AI Museums and Projects Launch

In the past 24 hours, it has been reported that Oracle and Meta are negotiating in-depth negotiations on about $20 billion in AI cloud cooperation, OpenAI plans to lease about $100 billion in backup computing power in the next five years, and Meta has submitted an application to enter wholesale electricity trading. The China Huawei Full Connection Conference disclosed the AI infrastructure and roadmap, the East Expo set up a 10,000-square-meter AI pavilion for the first time, and Anhui announced the first batch of AI scenario innovation projects with more than 370 million yuan, achieving balanced coverage at home and abroad.

1. Oracle and Meta negotiate a large AI cloud order of about $20 billion

1. On September 20, a number of authoritative media reported that Oracle and Meta were in in-depth negotiations on a multi-year AI cloud cooperation, with an amount of about $20 billion.
2. The focus of cooperation will be used to train and deploy generative AI models to supplement the training capacity of Meta's existing cloud service provider system.
3. If achieved, it will mark the further platformization of ultra-large-scale AI computing power procurement and form a strong pull on the supply of GPU and network interconnection.

2. OpenAI plans to lease about $100 billion in "backup computing power" in five years

1. On September 20, according to authoritative media, citing internal communication information, OpenAI plans to add about $100 billion in backup server leases in the next five years.
2. The company's cumulative server leasing expenses by 2030 are estimated to reach 350 billion US dollars, which is equivalent to an average annual value of about 85 billion US dollars.
3. The strategy aims to enhance supply elasticity and business continuity, and can be transformed into external marketable production capacity to support products and research growth.

3. Meta applies to enter wholesale electricity trading to deal with energy consumption in AI data centers

1. On September 20, the media disclosed that Meta had submitted an application to the U.S. federal energy regulator to participate in wholesale electricity trading.
2. The company's goal is to achieve self-sustaining or flexible scheduling during peak electricity consumption, and to resell electricity when there is surplus, so as to improve the proportion of clean energy and cost efficiency.
3. The integration of "computing power-green electricity-trading" in AI data centers is becoming a new paradigm, or guiding large technology companies to lay out their own energy assets.

4. Huawei Full Connectivity Conference Focuses on New Paradigms of AI Infrastructure and Interconnection

1. From September 18th to 20th, Huawei emphasized SuperPoD/Supercluster and high-bandwidth interconnection and proposed a new architecture for long-term computing power needs.
2. According to foreign media, the multi-year roadmap of Ascend Accelerator and the Atlas supercomputing solution have become the focus of attention, pointing to the continuous promotion of "independent and controllable" domestic AI computing power.
3. The open source and ecological trends of UB-Mesh and other interconnected technologies have triggered the industry to re-evaluate the cost and scalability of large-scale AI clusters.

5. The 22nd East Expo set up a 10,000-square-meter AI pavilion for the first time, and regional cooperation heated up

1. On September 20, the East Expo released a number of data: a 10,000-square-meter AI pavilion was created for the first time, and nearly 200 high-tech enterprises were displayed.
2. Multilingual exhibition agents, medical and service robots, etc. were intensively unveiled in the exhibition hall, and cross-border scenarios were significantly implemented.
3. China-ASEAN industry and trade with AI as the starting point will accelerate synergy, forming an exploration window for regional standards and mutual recognition of data.

6. Anhui released the first batch of AI scenario innovation projects, with an amount of more than 370 million yuan

1. On September 20, Anhui announced the first batch of artificial intelligence scenario innovation projects in 2025, with a total amount of more than 370 million yuan for 18 projects, driving a total investment of about 580 million yuan.
2. The project focuses on "artificial intelligence + manufacturing" and other directions, emphasizing scenario supply and computing power cost optimization.
3. Localities accelerate the implementation of models and industrial agglomeration in the way of "list system + track".

7. CICE Expo: iFLYTEK creates an AI learning machine and interactive experience area

1. On September 20, at the China International Consumer Electronics Expo held in Qingdao, iFLYTEK built an AI learning machine, Go robot and creative interactive experience area.
2. The on-site emphasis is placed on the stratification and content ecology of personalized precision learning ability for 3-18 years old.
3. The combination of education terminals and cloud services continues to increase the penetration rate of families and campuses.

8. Beijing Zhiyuan "Scholar Funding Program" was launched, facing original innovation

1. On September 20, Zhiyuan Research Institute launched a new annual funding program for global talents, covering AI and interdisciplinary.
2. Goal Through the support of the whole chain, promote the output of disruptive technologies and original innovative achievements.
3. The industry-university-research joint mechanism is regarded as the key starting point for domestic AI to shift from "large-scale training" to "breakthrough methods".

9. British startup Mantic AI ranks in the top 10 in international prediction competitions

1. On September 20, foreign media reported that Mantic AI ranked eighth in the Metaculus Cup event, with an overall performance outpacing a large number of human players.
2. The event covers dozens of real-world problems, emphasizing calibration ability and uncertainty reasoning.
3. The value of "inference and calibration AI" in investment research, policy evaluation and risk control has been further verified.

Frequently Asked Questions (Q&A)

Q: What is the largest transaction in this issue of Overseas AI News?

A: Oracle and Meta are negotiating a multi-year AI cloud computing cooperation worth about $20 billion.

Q: What is the direct purpose of OpenAI's addition of "backup computing power"?

A: Enhance computing power redundancy and business continuity, mitigate supply fluctuations, and monetize surplus capacity to support product growth and research iteration.

Q: What are the quantitative progress of "hard landing" in China?

A: The East Expo set up a 10,000-square-meter AI pavilion for the first time, and the first batch of AI scenario projects in Anhui exceeded 370 million yuan and drove a total investment of about 580 million yuan, all of which are clear landing data.

Q: What are the technical highlights of Huawei's Full Connectivity Conference?

A: The core of the system engineering around SuperPoD/Supercluster and high-bandwidth interconnection, as well as Ascend's multi-year roadmap, is to solve the bottleneck of large-scale AI training and inference with interconnection and computing power efficiency.