Meta description: Global AI news for November 29, 2025: China shifts AI training offshore, Google’s Gemini 3 leads, Gartner sees $15T AI agents, MIT warns on jobs, India–Google ink mega data center plan.

Top 5 Global AI News Stories – November 29, 2025

Offshore Training, AI Agents, Labor Upheaval, Google’s Comeback, and India’s Data Center Bet

The latest AI news cycle reveals a world in which artificial intelligence has become a core driver of geopolitics, enterprise strategy, and labor policy. Chinese tech giants are restructuring where they train their frontier machine learning models to navigate U.S. export controls, even as Google’s Gemini 3 and custom TPUs fuel a powerful comeback that has vaulted Alphabet toward a multi‑trillion‑dollar valuation. At the same time, Gartner projects that autonomous AI agents will intermediate tens of trillions of dollars in B2B commerce, while a major MIT study finds that today’s AI could already perform work equivalent to nearly 12% of U.S. jobs. India is also pushing to join the front rank of the AI industry through a proposed multibillion‑dollar Google data center investment. Together, these developments underscore the speed and scale of current global AI trends, and the urgency of aligning infrastructure, regulation, and workforce policy with the new reality.reuters+10​

1. China’s Tech Giants Move AI Training Offshore to Access NVIDIA Chips

Headline: Chinese AI Leaders Shift Model Training Overseas to Tap Restricted NVIDIA GPUs

According to reporting by the Financial Times and Reuters, leading Chinese technology companies—including Alibaba and ByteDance—are moving training of their latest large language models to data centers outside mainland China to gain access to NVIDIA AI chips that are heavily restricted at home. These firms are leasing capacity in foreign‑owned facilities, particularly in Southeast Asia, to train cutting‑edge models such as Alibaba’s Qwen and ByteDance’s Doubao on high‑end GPUs that cannot legally be supplied into China under tightened U.S. export rules. Separate analysis notes that Chinese startup DeepSeek, one of NVIDIA’s biggest buyers in China, can no longer use the most advanced U.S. chips domestically and is exploring local alternatives with Huawei and other Chinese vendors.finance.yahoo+6​

U.S. regulations bar the export of top‑tier NVIDIA accelerators to China and also restrict “export‑compliant” variants like the H20, while Beijing has reportedly moved to block foreign AI chips from use in Chinese state‑funded data centers. However, current rules do not prevent Chinese firms from renting compute on clusters physically located abroad and owned by foreign entities, creating a legal path for offshore training even when domestic use is constrained.tomshardware+3​

Editorial analysis (original content):
This offshore pivot highlights the limits of hardware‑centric export controls when measured against the borderless nature of artificial intelligence cloud infrastructure. The policy intent—to slow China’s ability to train frontier models on domestic soil—remains partially effective, but Chinese champions are clearly demonstrating that model capability can be “imported” via leased compute. For global regulators and compliance teams, the episode underlines the need to move beyond static chip lists and toward capability‑based, use‑case‑driven frameworks that consider who ultimately directs and benefits from model training, not just where the GPUs sit.

2. Gartner: AI Agents Poised to Command Trillion in B2B Purchases by 2028

Headline: Gartner Predicts AI Agents Will Handle 90% of B2B Purchases Within Three Years

At its IT Symposium/Xpo 2025, research firm Gartner forecast that autonomous AI agents will intermediate more than 15 trillion U.S. dollars in B2B purchasing by 2028, handling about 90% of all B2B transactions through automated negotiation and contracting systems. Analysts described this shift as one of the most far‑reaching changes in enterprise commerce in decades, enabled by machine‑to‑machine negotiation, standardized trust frameworks, and verifiable data feeds that allow agents to evaluate suppliers, place orders, and execute contracts with minimal human oversight.digitalcommerce360​

Gartner further predicted major second‑order effects: a 58‑billion‑dollar disruption in productivity software as generative AI and agents challenge incumbent office suites; AI proficiency tests in 75% of hiring processes by 2027; and “AI‑free” assessments at 50% of organizations by 2026 to verify independent reasoning without machine assistance. The firm expects fragmented, region‑specific AI platforms to cover 35% of countries by 2027 due to regulation and data sovereignty rules, and forecasts that organizations using multi‑agent AI in at least 80% of customer‑facing processes will significantly outperform peers on experience metrics by 2028.digitalcommerce360​

Editorial analysis (original content):
Gartner’s outlook formalizes what many in the AI industry already sense: the center of gravity is shifting from single chatbots to fleets of specialized, interconnected agents embedded in workflows. If even a fraction of the projected 15 trillion dollars in B2B spend moves to AI‑mediated channels, procurement, compliance, and audit functions will need to be re‑engineered around algorithmic accountability and machine‑generated contracts. The parallel prediction of thousands of “death by AI” legal claims tied to autonomous system failures reinforces that CIOs must treat governance, red‑teaming, and human‑in‑the‑loop controls as core design parameters—not afterthoughts—when deploying agentic architectures.digitalcommerce360​

3. MIT–ORNL “Iceberg Index” Warns AI Could Already Do Work Equal to 11.7% of U.S. Jobs

Headline: New Study Finds Today’s AI Can Technically Replace Nearly 12% of U.S. Workforce Tasks

A major study by the Massachusetts Institute of Technology (MIT) and Oak Ridge National Laboratory (ORNL) concludes that currently available AI systems are technically capable of performing work equivalent to 11.7% of U.S. jobs, based on detailed modeling of occupational tasks and automation economics. Using a large‑scale labor simulation tool dubbed the Iceberg Index, the researchers mapped more than 151 million workers, 923 occupations, and over 32,000 skills to identify the share of tasks that today’s AI can perform at competitive cost.fortune+5​

The study emphasizes that this figure represents technical and economic feasibility rather than an immediate prediction of layoffs, but it highlights substantial “hidden exposure” in white‑collar fields such as finance, HR, healthcare administration, legal services, logistics, and customer support. States including Tennessee, North Carolina, and Utah are already using the Iceberg Index to inform AI workforce strategies, while commentators note that the research provides policymakers with a “digital twin” of the labor market to test the impact of training programs, tax incentives, or regulation before implementing them.cnbc+5​

Editorial analysis (original content):
The Iceberg Index reframes global AI trends in employment by grounding the debate in granular task data rather than headline job titles. The finding that nearly one in nine jobs are already automatable by existing tools suggests that the most disruptive phase of the AI transition will not wait for hypothetical artificial general intelligence; it will be driven by incremental deployment of today’s systems into back‑office workflows, call centers, and professional services. For both governments and enterprises, the research raises a compliance and ethics question as much as an economic one: how to ensure that productivity gains from automation are shared via wages, reduced working hours, or social protection, rather than accruing solely to capital and a small set of AI super‑firms.

4. Google “Reclaims the AI Crown” with Gemini 3 and TPU Strategy

Headline: Gemini 3, Custom TPUs and Structural Re‑Org Power Google’s AI Resurgence

An in‑depth analysis from Moneycontrol describes how Google has “reclaimed the AI crown” in late 2025 after a rocky start to the generative AI era. Following early missteps with Bard and error‑prone AI search results, CEO Sundar Pichai declared a “moment of urgency” in early 2025, consolidating Google Brain and DeepMind into Google DeepMind, flattening management layers, and ramping investment in custom Tensor Processing Units (TPUs) and data centers. The article reports that Google’s Gemini 3 model is now topping major AI benchmarks, drawing praise from rivals such as Elon Musk and Sam Altman, while Salesforce CEO Marc Benioff called the leap in capability “insane” and said it “feels like the world just changed, again.”moneycontrol+1​

Alphabet’s market value has surged roughly 60–70% year‑on‑year, with the company briefly surpassing Microsoft and closing in on a 4‑trillion‑dollar valuation, helped by double‑digit growth in Google Cloud, YouTube, and subscription revenues. Analysts cited by Moneycontrol argue that Google’s vertically integrated stack—from proprietary data in Search, Android, and Ads, to TPUs and a unified Gemini layer across products like Search, YouTube, Gmail, and Workspace—gives it a unique moat as AI infrastructure becomes the primary battlefield. At the same time, competitors remain wary: an internal memo from OpenAI CEO Sam Altman reportedly warned staff that Google’s pre‑training gains could create “temporary economic headwinds” for OpenAI, while NVIDIA publicly asserted it remains “a generation ahead” in AI chips even as Meta explores deploying Google’s TPUs.reuters+6​

Editorial analysis (original content):
Google’s resurgence illustrates how quickly leadership in frontier machine learning can rotate when foundational assets—data, distribution, and compute—are already in place. The company’s push to make Gemini the unifying AI layer across consumer and enterprise products also shows how incumbents may lock in advantage not just by building better models, but by tightly coupling them to billion‑user platforms and proprietary telemetry. However, as experts in the article note, Google still faces open questions around data privacy, sustainability, and weak open‑source positioning, especially for enterprises that want to host models on‑premises or across sovereign clouds. For organizations evaluating vendors, this reinforces the need to weigh raw model performance against long‑term ecosystem openness and compliance obligations across multiple jurisdictions.cnbc+1​

5. India’s Adani Seeks up to Billion for Google‑Anchored AI Data Center

Headline: Adani in Talks to Raise Multi‑Billion‑Dollar Investment for Google‑Powered AI Hub in India

Reuters reports that India’s Adani Group is seeking to raise up to 5 billion U.S. dollars in equity investment to build a massive data center campus anchored by Google as a key tenant, as part of a bid to position India as a major regional hub for AI and cloud computing. The planned facility would host Google data centers supporting AI workloads and is part of a broader partnership in which Google has committed billions of dollars of investment to India’s digital ecosystem in recent years. The deal would also mark one of Adani’s largest technology infrastructure projects, tapping rising demand for AI compute and storage across South Asia.reuters​

The proposed project aligns with India’s stated ambition to become a leading AI producer rather than just a consumer, complementing earlier national initiatives around AI research, digital public infrastructure, and data‑center‑friendly policies in key states. Analysts note that if fully funded, the campus could attract additional hyperscale and enterprise tenants, making it a cornerstone of India’s AI and cloud strategy for the next decade.theaitrack+1​

Editorial analysis (original content):
The Adani–Google data center plan highlights how emerging markets are moving from policy rhetoric to asset‑level commitments in AI infrastructure. For India, securing a flagship AI‑optimized campus anchored by a global hyperscaler would strengthen its case as an alternative to traditional hubs in North America, Europe, and East Asia, particularly for workloads that benefit from proximity to a large, digitally connected population. For global AI providers, these kinds of partnerships also raise complex questions around data localization, energy use, and alignment with local AI ethics and content rules—areas where missteps can quickly trigger regulatory or reputational risk.

Structured Data, SEO, and Copyright/Compliance Notes

For publishers implementing this article:

• Schema.org/NewsArticle is recommended, with fields such as headline (“Top 5 Global AI News Stories – November 29, 2025”), datePublished, articleSection (“Technology”, “Artificial Intelligence”), and about (e.g., “artificial intelligence”, “AI agents”, “AI infrastructure”, “labor market”, “Google Gemini 3”).

• A supplementary FAQPage block can address recurring queries, for example:

  • “What is Gartner’s forecast for AI agents in B2B purchasing?”

  • “What does the MIT Iceberg Index say about AI and U.S. jobs?”

  • “How are Chinese tech firms training AI models overseas despite chip export controls?”
    Answers should remain concise, fact‑based, and traceable to the cited sources.

All factual statements above are grounded in reporting and analysis from reputable outlets including Reuters, Gartner (via Digital Commerce 360), Moneycontrol, CNBC, Fast Company, NDTV, Tom’s Hardware and others as cited. This article itself constitutes original editorial synthesis and commentary created for journalistic and informational purposes. Third‑party material is used only to the extent necessary to summarize and contextualize current events, consistent with fair‑use and news‑reporting exceptions in applicable copyright regimes. Any AI‑generated text here is transparently presented as analysis and does not replace or replicate the original reporting of the cited organizations.fastcompany+10​