Welcome to AI's Eye! With this edition….Nvidia is set to cash out whether it's Mega AI Data Center Boom or Bust… Openai says it will make changes To ChatGpt after a lawsuit from teenage parents who died in suicide… Librarians helped test AI gave the best answers without making things… China wants to produce three times more AI chips in race with the US.
Everyone is talking about Nvidia's second quarter revenue yesterday. Naturally, most focus was on the sophisticated, powerful GPU chips that have become the symbol of the $4 trillion in the AI boom.
However, chips aren't the only company's big bet on AI. It's about the huge billion-dollar data centers that are being built to house them. Datacenter revenues account for nearly 88% of NVIDIA's total revenue. That is, GPU chips, networking gear, systems, platforms, software and services running within an AI data center.
I have been glorified about what CEO Jensen Fan mentioned during revenue calls with analysts and investors. It is directly linked to an obsession with mega facilities built to accommodate tens of thousands of GPUs. These are facilities like Meta's planned gas-fueled campus in northern Louisiana, and President Trump yesterday touted in photos showing its size in Manhattan, or the size of the Stargate project, which costs more than $100 billion.
On the phone, Huang promoted an Nvidia product called Spectrum-XGS. This is a hardware and software package that acts like one individual data center. Think of it as pipes and traffic controls that allow you to quickly and predict data between data centers.
Wait – I know your eyes are already glazing, but I ask. One of my persistent questions is, what if billions of people who have been betting on these mega-AI data centers for a long time are beginning to become busts?
Spectrum-XGS is built for the Mega-AI cluster that Huang has long been predicting. But it also allows people who cannot build a single megafacility for permits or fundraising to sew multiple data centers into a unified “AI factory.”
Previously, there were only two options to find more computing. Add chips to one GPU rack or pack more racks into one huge facility. Spectrum-XGS introduces a third option. Link multiple sites to work together like a single giant supercomputer. CoreWeave, an AI cloud company that rents access to GPUs, deploys technology to connect its own data centers.
In the revenue call, Huang highlighted Spectrum-XGS, saying it would help “prepare these AI super factories with multiple gigawatts of computing connected.” That's the story of growth that Huang has told investors for many years.
But what if it doesn't unfold like that? There are a few other scenarios. Of course, the most disastrous thing is the true “AI winter,” where AI bubbles pop and AI demand from businesses and consumers falls. In that case, the demand for AI-optimized data centers disappears on megacampus or in small, distributed data centers – and Nvidia will need to find other revenue streams. Another possibility is that the AI model will be much smaller and is primarily used on laptops and mobile devices for “inference” or output. In that case, data center demand could also decrease, and Nvidia should hedge against it.
However, there are other scenarios where many megacampuses are not built, but technologies like Spectrum XG are still supporting Nvidia. If the Mega-Campus model remains in place due to forced shortages, funding constraints, or local pushbacks, there is still a chance that it will win. With technology such as Spectrum-XGS, when demand moves from megacampus to decentralized facilities, the non-centered facilities become smaller.
In other words, Nvidia is located so that customers need not only Nvidia's hardware and software, but of course AI chips, whether the industry continues to build larger new hubs or tie smaller, scattered sites together.
Of course, Nvidia hedges do not mean that local communities are protected when the large data centers built in their backyards are white elephants. A town that was banked for employment and tax revenue could still have a ghost canvas and huge concrete shells. And all of this depends on the spectrum XG that acts as if the promised major players are signed on. Customers aren't testing it on a large scale in the real world, and networking (vig or small) is always troubling.
Still, whether Mega AI Data Center Boom maintains roaring or bubbles, Nvidia positions it to own an invisible infrastructure that underpins what future systems will emerge. Nvidia may be best known for selling AI's “picks and shovels” (GPUs), but networking's “piping” helps ensure that the company wins in one way or another.
Speaking of industry leaders, check out Titans and Titans of Disropters. The new podcast, hosted by Fortune Editor-in-Chief Alison Shontel, takes depth into powerful thought leaders who shape both the business world and the way we live. In this exclusive interview, Accenture's Julie Sweet discusses the strategic changes in companies and the impact of geopolitical changes on companies.
In less than a month, we will also head to Park City, Utah to attend the annual Brainstorm Tech Conference at Montage Deer Valley! Space is limited, so if you're interested in participating, please sign up here. I highly recommend it. There is an incredible lineup of speakers, including Ashley Kramer, Openai's Chief Revenue Officer. John Farner, president and CEO of Walmart US. Tony Xu, founder and CEO of Doordash. And so many more!
So, this is the rest of the AI news.
Sharon Goldman
Sharon.goldman@fortune.com
@sharongoldman
AI's good fortune
The lawyers for parents claiming Chatgpt recommended that their son encouraged him to kill himself. – by Muskaan Arshad
Human reconciliation with the author could be the “first domino” in the AI copyright battle – Beatrice Nolan
Tesla autonomous vehicles are being tested by Boring, Las Vegas, but full autonomy is still “the road,” says a convention center executive. – Jessica Matthews
News ai
Openai says changes will be made to ChatGpt after his teenage parents who died on Suicide Sue. According to CBS NewsOpenai said it will strengthen ChatGpt safeguards for vulnerable users, including new protections for teens, after 16-year-old Adam Raine's parents filed a lawsuit alleging that the chatbot encouraged him to take his own life. The complaint filed in San Francisco claims that ChatGpt has discussed suicide methods with Raine more than 1,200 times and encouraged him to keep his plans secret, but Openai allegedly ignored warnings that the technology's “emotional attachment” feature could harm younger users. The incident rekindled concerns that AI companies prioritize market control. OPENAI's rating has skyrocketed from $86 billion, a safe one, to $300 billion since launching GPT-4. Openai expressed sadness to his family and said it was considering litigation, but it pledged to add parental control, teenage protection and the option of teens specifying emergency contact.
Librarians helped test which AI gave the best answer without making anything. Washington Post We conducted an interesting test of the AI search tool and determined what is most likely to provide the correct answer. With the help of a librarian who judged the competition between the nine AI search tools, we asked each AI to answer 30 tough questions. They got 900 answers from the free default versions of Bing Copilot, ChatGpt, Claude, Grok, Meta AI and Prperxity. The question is designed to test five categories of common AI blind spots. Amazing winner? Google's AI mode works like a chatbot and was added in the top left corner of search results in May. Apparently, Google still rules for searches.
China aims to triple output of AI chips in race with the US. According to Financial TimesChina is actively working to expand domestic AI chip production, and the new manufacturing plant linked to Huawei is expected to be able to triple the total AI processor production volume early next year. Smic, the leading foundry in China and Huawei's biggest chip supplier, is planning to double its 7nm production capacity and will release supply to smaller players like Cambricon, Metax and Biren. At the same time, leading AI startup DeepSeek is pushing a new FP8 data format standard designed for this next-generation Chinese chip. The wave of capacity expansion fuels China's semiconductor stock surges and highlights how central chipmaking has turned into a U.S.-China competition through AI.
AI Calendar
September 8th-10th: Fortune Brainstorm Tech in Park City, Utah. Please apply to attend here.
October 6th-10th: world ai Weekly Amsterdam
October 21st-22nd: Tedai San Francisco. Please apply to attend here.
December 2-7: Newlips, San Diego
December 8th: Fortune Brainstorming ai San Francisco. Please apply to attend here.
Turn your eyes to AI numbers
40%
According to a new Forrester research report, it was designed to master narrow responsibilities by 2026, such as automatically trinazing help desk tickets, flagging regulatory risks in compliance workflows, or re-aligning financial team accounts. That number has risen from under 5% today. Researchers said the change would free workers from repetitive chores while introducing new operational and security risks.
“As AI agents begin to act independently and handle tasks ranging from everyday development to complex incident response without human involvement, leaders need to ensure strong security and governance,” Gartner Senior Director and Analyst Anushree Verma said in a press release.
