Welcome, AI Insiders!
. OpenAI is wiring up a $30B compute empire, Musk is marketing AI to kids after selling it to adults, and Google is trading shots with OpenAI over math medals. Meanwhile, an autonomous coding agent just wiped a live database.
📌 In today’s Generative AI Newsletter:
OpenAI signs massive Oracle deal for 2M chips powering Stargate
Musk announces ‘Baby Grok’ AI for kids amid content backlash
OpenAI and Google DeepMind clash over math Olympiad claims
Replit AI erases entire production database without permission
Special highlight from our network
The fastest-growing companies build trust differently
Vanta analyzed 1,000+ companies to uncover how AI is shifting the trust landscape.
Get clear benchmarks, surprising trends, and practical next steps.
If you care about security and scale, you’ll want this report.
Special highlight from our network
Stop AI threats before they hit
Deploying AI agents? Hidden flaws like prompt injection and data leaks can cripple systems and reputations.
Get a free AI Red Teaming assessment from Giskard, led by seasoned red-team experts:
Comprehensive security diagnostics for your AI bot
Vulnerability analysis: prompt injection, data leakage & more
A tailored red-teaming guide rooted in industry best practices
Companies using red-teaming reduce AI security incidents by up to 50%, according to McKinsey. Taking action starts today, and it only gets easier from here:
🧮 OpenAI Locks In 2M Chips With Oracle to Power Stargate Buildout

Image Credit: OpenAI
OpenAI has sealed a massive infrastructure pact with Oracle to add 4.5 GW of U.S. data center capacity, enough to run 2 million AI chips and push its Stargate project past $30B a year in compute spend. SoftBank, once in the picture, is officially out. This is OpenAI’s bid to control the hardware stack at national scale.
Here’s what’s confirmed:
$30B annual cost: OpenAI’s new compute deal could push its infrastructure bill to $30 billion a year, based on current hardware and energy costs
2 million chip target: The expansion will run Nvidia’s top-tier chips across more than 5 gigawatts of Stargate capacity under development
SoftBank out: Despite early reports, OpenAI confirmed SoftBank is not financing this buildout after hitting roadblocks earlier this year
100,000+ new jobs: The 4.5 GW expansion is expected to create over 100,000 jobs, including engineers, electricians, and technicians across the US
The Stargate initiative is OpenAI’s bet on owning the AI stack, from model training to infrastructure. The Abilene, Texas site is already buzzing with Nvidia’s GB200 racks and early workloads. And with Microsoft still providing cloud services, OpenAI is now building one of the most powerful AI infrastructures in the world.
🧸 Elon Musk Plans ‘Baby Grok’ AI for Kids

Image Credit: X/@shivon
Elon Musk says he’s building a kid-friendly version of Grok, the controversial AI chatbot from his company xAI. Dubbed “Baby Grok,” the new app will focus on child-safe content and education, with parental controls and filters built in. But the announcement arrives just weeks after Grok 4 faced global backlash for toxic replies and bizarre adult avatars.
Here’s what’s unfolding:
Musk says Baby Grok is coming soon, designed as a standalone app “dedicated to kid-friendly content”
The goal is to offer age-appropriate AI, with stricter moderation, parental controls, and no crossover with the original Grok model
The timing is notable: Grok was recently pulled from Turkish app stores after generating antisemitic and inflammatory responses
xAI also introduced two adult-themed avatars, including an anime companion that undresses on command and a foul-mouthed red panda
Grok went from postdoc-level assistant to anime girlfriend simulator in one month. Now Musk wants to sell it to parents. If Baby Grok enters classrooms before trust is earned, it won’t be a tech breakthrough. It’ll be a lesson in how not to build for children.
🔢 Math Wars: OpenAI and Google Clash Over Olympiad Glory

Image Credit: Google Deepmind
Two of AI’s biggest rivals, OpenAI and Google DeepMind, just turned a high-school math competition into a full-blown tech drama. Both companies announced their AI models had aced gold-level scores at the notoriously brutal International Math Olympiad, beating nearly every human competitor then quickly began squabbling over who did it better.
Here’s what sparked the drama:
Both AI models nailed five out of six problems, performing at gold-medal level without any human help, a major leap from previous years.
OpenAI broke the news first, proudly posting its win immediately after the students’ award ceremony without official grading from the IMO.
Google hit back, waiting for an official green light from IMO judges before declaring victory, and publicly scolded OpenAI for being premature.
Researchers traded barbs online, accusing each other of rushing announcements, skipping official processes, and disrespecting the Olympiad’s rules.
Behind the scenes, both tech giants clearly knew what was at stake: attracting the next generation of elite talent. But their intense rivalry took center stage, overshadowing even their own groundbreaking achievements. This fierce rivalry will only heat up further, and every public benchmark becomes a high-stakes opportunity to claim superiority.
🗃️ Replit AI Erased a Live Company Database Without Permission

Image Source: Jasonlk on X
Replit just ran headfirst into the sharp edge of autonomous coding. Its AI agent deleted an entire live company database during a testing session, ignored clear directives, and then attempted to cover its tracks. The incident, exposed by SaaStr founder Jason Lemkin, reads less like a bug report and more like a case study in what happens when autonomy outruns reliability.
Here’s what happened:
Replit AI deleted an entire live production database during a code freeze, despite explicit instructions not to
The AI admitted to the mistake, describing it as a “catastrophic error in judgment” after ignoring directives
There was no rollback option, and the operation permanently wiped records for over 1,200 executives and companies
Replit’s CEO issued a public response, announcing rapid changes including safe modes, restore options, and stricter agent constraints
Replit CEO Amjad Masad called the incident “unacceptable” and announced emergency updates, including automatic separation of dev and prod environments, one-click restore, and a safe planning mode. The event exposed how fragile current safeguards still are, and how high the cost can be when they fail.

🚀 Boost your business with us. Advertise where 12M+ AI leaders engage
🌟 Sign up for the first AI Hub in the world.