Nvidia CEO Jensen Huang resigns as Google launches Gemini 4 Hydrogen

The end of the Huang era

Nvidia announced that Jensen Huang has resigned from his roles as president, chief executive, and board member. This departure marks the end of a 33-year period in which Huang transformed a graphics card business into the primary engine of the global AI economy. The news broke during morning programming on CNBC and sent shockwaves through the tech sector.

The sudden resignation follows reports of personal controversies involving Huang and several individuals, including a direct report. While the board has expressed full confidence in the remaining leadership team, the company faces an immediate transition period. Colette Kress will step in as the interim chief executive while a permanent replacement is identified.

The market reaction to the news was swift and volatile. Nvidia shares, which had recently closed at a record high of $239.24, fell to an overnight low of $156.20 following the announcement. Investors are now looking toward the company's third-quarter results, which are scheduled for release on November 17.

Google's biggest search update

In a parallel development, Google launched Gemini 4 Hydrogen as the default model for its AI Mode in Search. This rollout is available to all users globally and represents what Google describes as the largest single change to its search engine's model. The update is designed to handle more complex user intents and provide more nuanced results across various languages.

The Hydrogen model introduces several specific improvements for the average searcher. It is built to better understand misspelled or complex questions and to fan out queries across multiple subtopics to surface a broader set of links. Additionally, it provides richer visual answers, including tables and custom tools, to help users find information more efficiently.

Industry analysts expect these changes to primarily affect AI Overviews and AI Mode rather than the core search rankings. The launch occurred during the November 2026 core update, positioning Hydrogen as the primary tool for handling difficult questions. This move places Google's newest model at the forefront of its competition with other AI-driven search experiences.

Benchmarks and performance

While Google highlights the improvements of Hydrogen, independent testing provides a more complex picture of the current AI landscape. A recent Multimodal Knowledge Agent benchmark released by Google DeepMind and Artificial Analysis compared several leading models. The results showed that Anthropic's Claude Fable 5.2 outperformed Gemini 4 Carbon in quality, despite being significantly more expensive to run.

Google's own internal testing places the Carbon model second in quality among the models they tested. However, the gap between Carbon and the top-performing Fable was described as statistically insignificant in the 25-page report. These benchmarks highlight the ongoing trade-offs between model accuracy and the cost per task for developers and companies.

The performance of these models is becoming a critical factor as AI moves into more practical applications. For instance, Google is extending Gemini's ability to operate food, ride, and reservation apps to several countries. As these models become more integrated into daily tasks, the balance between high-quality output and operational costs will remain a primary concern for the industry.

Impact on web traffic

The rise of AI-generated answers is already beginning to reshape how people navigate the internet. Data from SparkToro shows that visits to the top one million websites have fallen by 12.2% since early 2022. While the online population has grown by about 24% in that same period, the total number of visits per internet-connected human is declining from its peak.

News publishers are feeling these consequences more acutely than many other types of websites. Only 10 of the 50 largest English-language news sites grew in visits year over year, with The New York Times being the only top-ten riser. Search traffic now accounts for only 12% of North American publisher pageviews, a significant drop from 21% a year earlier.

This shift is leading to a contraction in the journalism trade, with some organizations already cutting staff and pausing long-running conferences. As AI models provide direct answers, publishers are searching for new ways to monetize their content. Some are moving toward collective licensing models to ensure they receive payments when their content is used to train or inform AI responses.

Regulatory and policy shifts

As AI models become more prevalent, regulators are moving to tighten the rules around data privacy and advertising. The European Data Protection Board recently adopted guidelines stating that cross-site ad tracking will rarely pass the legitimate interest test. This means that personalized advertising will be treated as direct marketing, making it easier for users to object to tracking.

In the United States, the Red Iberoamericana de Proteccion de Datos ruled that client data is off-limits for AI training by default. They clarified that simply mentioning the use of algorithms does not satisfy a company's duty to inform users. These protections are becoming standard as companies face increasing scrutiny over how they handle personal information to feed their models.

Additionally, Google has updated its policies to ban ads that use AI to impersonate public figures, even if the advertiser discloses the use of AI. The company also clarified that targeting users based on prohibited categories, such as counterfeit goods, remains strictly forbidden. These policy updates reflect a broader effort to establish guardrails as AI-generated content becomes a larger part of the digital advertising ecosystem.