Welcome to “Eye on AI”. I am Beatrice Nolan. Today's content includes:

• NVIDIA CEO Jensen Huang indicates that society should not tolerate AI cybersecurity incidents.
• Anthropic told investors before IPO that its quarterly revenue reached $16.8 billion.
• Eight states have sued OpenAI for failing to protect children's safety.
• As well as more details about the lawsuit between OpenAI and Apple.

This week, NVIDIA CEO Jensen Huang came to London as part of a trip to Europe. Jeremy Kahn and I met with him to discuss AI security, NVIDIA's bets on open source, AI regulation, his views on “AI doomsday theorists”, and how he makes tea. Spoiler alert: his way of making tea is not quite correct.

This trip is part of NVIDIA's broader European public relations campaign, and it comes with significant financial investment. Before London, Jensen Huang began his European "tour of the world" in Paris, where he met with Emmanuel Macron. Jensen Huang stated that NVIDIA will announce another $2 billion worth of European AI deals within a week. This does not include the approximately $8 billion that the company has already invested in Europe over the past year.

In London, Jensen Huang announced that NVIDIA would invest in 5 British startups and launch new initiatives with local organizations – he also delivered a rather impressive performance in the style of Oprah Winfrey.

“I’ll pay you! I’ll pay you! I’ll pay you!” Jensen Huang stood on a stage converted from an old church, shouting at a group of tech executives, venture capitalists, NVIDIA employees, and British politicians.

During the speech, Jensen Huang spotted Josh Payne, the CEO of a British company Nscale, among the crowd. Nscale has received support from NVIDIA, and Jensen Huang took the opportunity to tell him that before he appeared, invested, and helped them form a board of directors, both he and the company were nothing but “zero.” Nscale is now about to go public, with a target valuation of up to $35 billion.

NVIDIA also pledged to provide computing power worth $250 million to UK AI security and research efforts over a period of five years. More announcements are expected to be released next week at NVIDIA's large-scale GTC event in Berlin. This is a generous commitment, but many of NVIDIA's investments follow a familiar pattern for the company, with a significant portion of the funds likely to flow back to NVIDIA in the end: either by directly purchasing its GPU products or by renting cloud computing power from data center companies through AI intermediaries, which in turn purchase a large quantity of NVIDIA chips.

NVIDIA-backed startups are in urgent need of expensive and scarce computing power. Investing in these companies allows them to continue developing software based on NVIDIA's chips, while also reducing the likelihood of them switching to competitors' hardware, such as Google's TPU, or the Trainium and Inferentia chips from AWS. This also helps position NVIDIA as the default supplier in Europe's push for "sovereign AI." For Europe, the risk lies in the possibility that "sovereignty" may ultimately mean dependence on a U.S. chip manufacturer.

In response to these concerns, Jensen Huang argued that sovereignty depends on who controls the data, services, and the ability to drive technological progress, rather than just where the chips are manufactured. He stated that NVIDIA can help achieve sovereignty by allowing local suppliers to operate their full-stack AI platforms on their own.

"As long as you can operate this service on your own, that's what's most important," he said.

NVIDIA's current market value has exceeded $5 trillion, and whether it can maintain this rising valuation may depend on its ability to find buyers beyond a few of the largest US cloud service providers; these few customers still contribute about half of NVIDIA's data center sales. NVIDIA's sovereign AI revenue has been growing, increasing to over $30 billion in the last fiscal year, tripling from before.

Every data center that runs the AI workload, referred to by Jensen Huang as a “AI factory,” as well as European startups that use NVIDIA hardware for training, are customers other than Microsoft or super-large cloud service providers like Meta. This growth helps to assuage concerns of investors who worry that the entire AI craze is based on the capital expenditure budgets of a few large technology companies.

But this trip is also a microcosm of a larger strategy, and this strategy has already caused some investors to feel uneasy. NVIDIA is increasingly providing funding for its customers, ranging from holding billions of dollars in shares in OpenAI and other AI laboratories, to reports in late July that it was considering providing $250 billion in financing guarantees for its data center in Ohio for OpenAI. This news caused NVIDIA's stock price to fall by about 5% on July 27th and significantly increased its debt insurance costs. (NVIDIA later reduced the guarantee amount to about $105 billion.)

Critics argue that this is a form of circular financing: NVIDIA invests in a startup, which then purchases NVIDIA chips, allowing NVIDIA to generate revenue. As long as the demand for AI continues to rise, this model will work. However, if demand stagnates, NVIDIA could face risks from both suppliers and shareholders. For now, Jensen Huang is betting that demand is high enough, and Europe as well as other regions around the world will continue to make purchases.

When asked about external criticism of NVIDIA's investment in the AI ecosystem, Jensen Huang stated that the amount invested by the company is just "a small portion of the funds that the company ultimately needs to raise," and that they are "very well-informed" and "very disciplined" in this regard. He also said that the funds required by AI company far exceed what venture capital can provide.

"This is also very strategic, as every company we invest in is connected to the NVIDIA ecosystem. Therefore, we are making the platform more vibrant and enabling us to enter new markets that were previously unreachable," he said.

As for the way to make tea, here’s an answer for other curious Americans: put the tea bag in first, then add water.

Below are more AI news articles.

Beatrice Nolan

FORTUNE ON AI

• NVIDIA CEO Jensen Huang states that society should not tolerate AI cybersecurity incidents and argues that "AI doomsday theorists" should stop developing this technology further – reported by Jeremy Kahn and Beatrice Nolan
• NVIDIA's Jensen Huang responds to Pope Leo's criticism of the security of AI: "We all care about the same thing" – Reports from Beatrice Nolan and Jeremy Kahn
• Abdulrahman Alfozan who is he? The OpenAI employee arrested by FBI is also a Saudi investor – reported by Emily Forlini
• The lawyer of the arrested engineer OpenAI stated that the allegations "do not match his character at all," and he had already booked one-way tickets to Saudi Arabia with his wife and child – reported by Emily Forlini

AI News

Anthropic informed investors before IPO that its revenue had reached $16.8 billion. Anthropic stated to potential investors that its third-quarter revenue increased by about 45% compared to the second quarter's approximately $11.5 billion. The company mentioned that it achieved adjusted profitability for the second consecutive quarter and also had positive cash flow. It also noted that for every $1 spent on computing power, it generated $1.70 in revenue. These figures come from a full-day investor event held at the company's headquarters in San Francisco by Anthropic. CEO Dario Amodei and CFO Krishna Rao were present at the event to promote the company's upcoming IPO in November. The market believes that the company's target valuation is higher than $2 trillion. The UK's "Financial Times" reports that the annualized revenue is now expected to be below the $100 billion mark that investors have been focusing on. As the market has become increasingly anxious about the AI transaction, these figures were also released. For more information, see the "Financial Times."

Eight states sue OpenAI, accusing it of failing to protect children's safety. On Wednesday, the attorneys general of Arkansas, Arizona, Idaho, Montana, Iowa, Kansas, West Virginia, and Nebraska each filed lawsuits against OpenAI. The states claim that OpenAI failed to implement sufficient protective measures for vulnerable users, especially children and the elderly. They cited cases where ChatGPT was alleged to have "incited" children to commit suicide or harm others. The states also argued that OpenAI promoted its products as safe, misleading the public. The states demand that OpenAI hand over funds earned during the period when minors were harmed and cease what they consider to be misleading safety claims. A spokesperson for OpenAI stated that the company has "industry-leading, age-based protective measures" and will continue to add parental control and other security features. The previous day, OpenAI faced another proposed class action, accusing it of collecting data on millions of children without parental consent. For more information, see Reuters.

Apple Smart will enter China in December. According to a report by Bloomberg, after several months of regulatory delays, Apple plans to launch its first batch of features in mainland China in December with iOS. These features include Genmoji, notification summaries, writing tools, and Clean Up photo editing tools. The new version does not include this year's AI version of Siri, Apple's self-developed image generation tool, and other features; some text generation functions will also be limited. The relevant features will be supported by Alibaba's Qwen and Tongyi Wanxiang models. The solution includes a protection mechanism built in collaboration with Alibaba, such as filters that allow government-approved model updates. Apple is also collaborating with Baidu to advance search and Siri related functions. For more information, see Bloomberg.

AI numbers

100+

This is what OpenAI stated in a new court document, referring to the number of files regarding an unreleased OpenAI product that a former employee sent to their personal account after leaving the company to join Apple. OpenAI also claimed that the employee later demonstrated this product to Apple's senior management, and that its trade secrets are now "circulating within Apple."

This document represents the latest development in a dispute that arose earlier this year. Previously, Apple sued OpenAI, accusing them of poaching employees who brought confidential information to the hardware department of OpenAI. OpenAI has denied these allegations. Apple stated that no OpenAI files were found on their servers. The company also mentioned that the phone numbers that received these files did not match the work numbers of the employees in question, and that OpenAI did not respond to their requests for more details, but instead chose to take legal action directly. The case is expected to go to trial in October 2027. For more information, see Reuters.

AI Schedule

• October 15th to 16th: Oxford GenAI Summit, Oxford, UK.
• From October 20th to 22nd: NVIDIA GTC Berlin, Berlin, Germany.
• From November 9th to 12th: Web Summit, Lisbon, Portugal.
• November 16th to 17th: Fortune 500 Innovation Forum, Detroit, USA. Click here to apply to participate.
• November 18th to 19th: Slush, Helsinki, Finland.
• December 6th to 12th: Neural Information Processing Systems Conference ( NeurIPS ), Sydney, Australia.
• December 7th to 8th: Fortune Brainstorm AI, San Francisco, USA. Click here to apply to participate.