Google’s most powerful AI chip is here, and Apple’s Siri is getting an epic upgrade—here’s the latest AI news!

AI Frontier Express ↓↓

01 Google Cloud Unveils Its Most Powerful Proprietary TPU: The Fruit of Ten Years of R&D, Delivering Four Times the Performance of Its Predecessor.

In an official blog post, Google Cloud announced the company’s formal launch of the 7th-generation TPU (Tensor Processing Unit)—“Ironwood.” This chip will be officially available in the coming weeks. Designed entirely in-house by Google, the chip is capable of handling a wide range of tasks, from training large-scale models to running real-time chatbots and AI agents.

It is reported that Ironwood is specifically designed to handle the most demanding workloads—from large-scale model training and complex reinforcement learning (RL) to high-capacity, low-latency AI inference and model serving—all with ease.

Google says that compared to the TPU v5p, Ironwood delivers peak performance improvements of up to 10 times. Compared to the TPU v6e (Trillium), its single-chip performance for both training and inference workloads is more than four times higher, making it Google’s most powerful and energy-efficient custom chip to date. Google notes that the next-generation Ironwood TPU can connect up to 9,216 chips within a single cluster, thereby eliminating “data bottlenecks in the most complex models” and enabling customers “to run and scale the world’s largest, most data-intensive models today.”

02 Domestic AI Makes a Comeback! Alibaba’s Qwen Wins Investment Competition, Achieving a 22.32% Return Rate That Outperforms GPT-5 by a Wide Margin

The inaugural Alpha Arena AI Investment Competition has concluded, with Alibaba’s Tongyi Qwen3-Max taking the top spot, achieving a return rate of 22.32%. The competition was initiated by the U.S.-based Nof1 organization and featured six of the world’s leading models. Among them, only Qwen and one other Chinese model turned a profit; the four U.S.-based models—including GPT-5—sustained losses across the board. Notably, GPT-5 suffered the largest loss, falling to last place with a deficit exceeding 62%, underscoring China’s competitive edge in real-world AI applications.

03 Apple’s Siri Receives an Epic Upgrade! The 1.2-trillion-parameter Google Gemini Model Joins the Ranks—Set to Make a Stunning Debut in Spring 2026

According to reports from several authoritative media outlets on November 6, Apple and Google have struck a historic partnership: Apple will invest $1 billion per year to integrate Google’s Gemini 2.5 Pro large language model—featuring 1.2 trillion parameters—into its systems, completely overhauling the Siri voice assistant. The new version of Siri will adopt a collaborative edge-cloud architecture, with complex tasks processed through Apple’s private cloud servers. This approach not only ensures robust privacy and security but also ushers in a quantum leap in semantic understanding, multi-turn conversations, and real-time information retrieval. The revamped Siri will support integration of local data such as users’ call histories and photos, enabling it to automatically plan trips, recommend restaurants, and—for the first time—possess AI-powered web-search capabilities that can directly generate summaries of web content. This feature is expected to roll out with iOS 26.4, scheduled for release in spring 2026, marking a pivotal move in Apple’s AI strategy.

04 OpenAI invests 38 billion yuan to sign a seven-year deal with Amazon! Securing hundreds of thousands of NVIDIA GPUs.

Amazon Web Services (AWS) has signed a seven-year, $38 billion agreement with OpenAI to provide it with hundreds of thousands of NVIDIA GPUs—including GB200 and GB300 models—for powering ChatGPT responses and training next-generation models. All computing power deliveries are expected to be completed by the end of 2026. This move is part of OpenAI’s strategy to diversify its computing power infrastructure; previously, OpenAI had already secured massive partnerships with Microsoft, Oracle, and other major players.

05 Domestic AI Breakthrough! MiniMax Music 2.0 Outperforms All Others in Vocal Realism, Supporting Duet Performances by Male and Female Singers + Creation Across All Genre Categories

On October 31, Shanghai AI unicorn MiniMax released its next-generation music model, Music 2.0, marking a revolutionary breakthrough in vocal synthesis technology. This model’s voice quality is virtually indistinguishable from that of a real human singer and can precisely master more than ten singing styles, including pop, jazz, and rock. It also supports highly sophisticated collaborative performances such as dynamic duets between male and female singers and a cappella arrangements. Each song can be up to five minutes long and features a complete musical structure, including verses, choruses, and bridges.

06 NVIDIA’s H100 GPU Enters Space for the First Time! With Performance Over 100 Times Greater Than Previous Models, It Ushers in a Revolution in Space Computing Power.

On November 2, 2025, NVIDIA’s H100 GPU, aboard Starcloud-1 satellite developed by the U.S. company Starcloud, was successfully launched into space. This marks the world’s first-ever deployment of a data-center-grade GPU for operation in space. Powered around the clock by solar arrays and relying on passive cooling in the vacuum environment, the satellite effectively addresses the energy and thermal management challenges faced by ground-based data centers. The H100 boasts performance that is 100 times superior to existing space-based computers, enabling it to process Earth observation data in real time and run large language models. Commercial computing services are expected to become available starting in 2027. This breakthrough signifies the transition of space-based computing from concept to practical implementation, offering a brand-new pathway for advancing AI development and achieving carbon neutrality goals.

07 Breaking News! The Claude model has been removed from shelves, leaving users exclaiming, “Our souls are gone!”

On November 4, ByteDance’s AI programming tool Trae announced that, due to service adjustments, access to the Claude model would be removed effective immediately, sparking heated discussion among users. This adjustment is directly linked to Anthropic’s tightening of its service terms for Chinese-funded enterprises. Although Trae is operated by a Singapore-based subsidiary, it remains constrained by its Chinese ownership background. As compensation, Pro members will receive an additional 300 fast requests per month—a 50% increase—valid until January 31, 2026. Trae simultaneously introduced alternative models such as GPT-5 and Gemini-2.5-Pro; however, some users have reported that these new models still fall short in terms of code-generation accuracy and contextual understanding, leading some to even consider switching to competitors like Cursor. This incident highlights the impact of geopolitics on the AI tools ecosystem and prompts developers to reassess the risks associated with technological dependency.

08 Modern and NVIDIA Invest $3 Billion to Build an AI Factory! 50,000 Blackwell GPUs Reshape the Future of Smart Cars and Factories.

Hyundai Motor Group and NVIDIA announced that, in partnership with the South Korean government, they will invest approximately 3 billion U.S. dollars to build a national-level physical AI cluster in Gyeongju, deploying 50,000 Blackwell GPUs to create an "AI factory." This facility will integrate autonomous driving, smart factories, and robotics technologies, leveraging the Omniverse digital twin platform to optimize production processes and using the Nemotron large model to develop in-vehicle AI assistants. The two companies plan to establish an NVIDIA AI Technology Center and a Hyundai Physical AI Application Center, positioning South Korea as a hub for AI innovation in the Asia-Pacific region. Their goal is to achieve end-to-end intelligent automation—from model training and verify to deployment—by 2027. This collaboration marks a pivotal leap for the automotive industry into the era of "software-defined everything."

09 Microsoft Partners with U.S.’ Largest Media Company to Launch AI-Driven Content-Pay Revolution: Usage-Based Billing Model Targets Google’s Dominance in Traffic

America’s largest media publisher, People Inc. (formerly Dotdash Meredith), has reached a groundbreaking AI authorize agreement with Microsoft, becoming one of the first launch partners in Microsoft’s “Publisher Content Marketplace.” Under the agreement, Microsoft Copilot will pay People Inc. content usage fees based on actual usage, pioneering a new “à la carte” payment model. This move directly addresses Google’s long-standing “malicious behavior” of unauthorized scraping of news content—People Inc. had previously used Cloudflare technology to block AI crawlers, forcing companies—including Microsoft—to proactively negotiate content deals. Data shows that Google Search still accounts for the majority of People Inc.’s traffic; however, Microsoft’s integration into the AI ecosystem is enabling it to build a differentiated competitive advantage, marking a pivotal turning point in the media industry’s shift from free content to value-based monetization.

Source: School of Artificial Intelligence

Editor: Yu Li

Proofreading: Zeng Xiong

Reviewed by: Zhang Weitian