📊 MARKET PULSE
OpenAI’s Revenue Run Rate Doubles — But Profit Could Be Years Away
OpenAI recently hit an annualized revenue run rate of $10 billion as of June 2025—nearly double the $5.5 billion figure reported in December 2024. This surge, driven by booming demand for ChatGPT across enterprises and individual subscribers, positions the company well to reach its $12.7 billion full-year target.
The number excludes licensing revenue from Microsoft and large one-time deals, reflecting purely recurring income—from consumer ChatGPT subscriptions, roughly 3 million business/education users, and its API. → Continue reading here.
🔮 Altman Predicts Insight-Generating AI by 2026
OpenAI CEO Sam Altman says AI will likely generate “novel insights” by 2026, signaling a shift toward models that can help scientists form new hypotheses. His essay suggests OpenAI is leaning into discovery-focused AI. But experts argue today’s systems still fall short of true originality, lacking the spark to ask transformative questions.
🤖 Meta’s V-JEPA 2 Teaches AI to Predict Actions
Meta has unveiled V-JEPA 2, a new world model designed to help AI agents reason about and predict physical outcomes from video. Trained on real-world scenarios, it boosts understanding—letting robots anticipate object behavior and human actions. This is key to building intelligent, action-aware machines. Three new benchmarks accompany the release.
🌐 Dia Browser Centers AI in Web Experience
Dia, the new browser from The Browser Company, puts AI at the heart of web interaction. Built for personalization, it uses chat to navigate tabs, recall history, and act as a contextual assistant. With cookie access and smart “skills,” it adapts to users’ habits and tasks. Dia isn’t just a browser—it’s a prototype for how AI could become your digital co-pilot.
🏗️ NVIDIA to Build AI Cloud Hub in Germany
NVIDIA will launch its first industrial AI cloud in Germany, targeting sectors like automotive design and logistics. CEO Jensen Huang also revealed plans for 20 AI factories across Europe and a 10x boost in regional AI compute by 2027. Partnering with Mistral, NVIDIA aims to accelerate Europe's sovereignty in AI and deepen its integration into critical industries.
Enjoying our newsletter? Forward it to a colleague—
it’s one of the best ways to support us.
☝️ POWERED BY RECALL
Never Forget Again with Your AI Encyclopedia
A self-organizing knowledge base that grows smarter the more you use it. Recall is your own personal AI encyclopedia — helping you transform scattered content into an interconnected knowledge base that grows smarter the more you use it.
Features include:
High-quality summaries & chat
Automatic categorization and smart linking
Personalized review schedule
(new) Augmented Browsing, which surfaces related content from your knowledge base as you browse - turning stored knowledge into active intelligence.
Experience Recall now and get 30% off with code MB30 valid until 1 July 2025
👾 FORWARD FUTURE ORIGINAL
Inside Our Chat with Satya Nadella
Microsoft’s CEO isn’t just thinking about AI agents as assistants. He believes they’re becoming the new operating system for how businesses run.
In our conversation with Satya Nadella, he shares:
– Why traditional apps are collapsing into agentic workflows
– How Excel is turning into a development environment
– What the “AI factory” of the future looks like
– And why intelligence, not software, will be the architecture of everything
🪖 TALENT WARS
Meta’s $2M Pay Can’t Stop the AI Talent Drain
Meta is struggling to retain top AI talent, even with compensation packages exceeding $2 million annually, as researchers increasingly jump ship to rivals like OpenAI and Anthropic. The exodus isn’t just about pay—Anthropic, in particular, is drawing talent with a culture that values autonomy, intellectual freedom, and minimal bureaucracy. It boasts the highest two-year retention rate in the sector (around 80%), outpacing even DeepMind.
Meanwhile, nearly 20% of new hires at leading AI labs now come from big tech firms like Meta, Google, and Microsoft. As AI startups build influence and credibility, the old guard’s allure is fading. Today’s elite engineers aren’t just looking for cash—they want creative control and purpose. → Read the full article here.
💊 POLICY & REGULATION
FDA Bets on AI to Fast-Track Drug Approvals Amid Staff Cuts
With nearly 2,000 employees cut under Trump, the FDA is turning to artificial intelligence to speed up drug and device approvals, aiming to match the warp-speed pace seen during the pandemic. In a JAMA article, Commissioner Marty Makary unveiled a plan centered on the agency’s new AI model “Elsa,” which helps summarize data and prioritize inspections. Officials claim AI could trim review timelines from months to weeks, but early use shows technical limits and accuracy concerns.
Critics warn that the FDA’s push for faster reviews may echo pharma’s agenda more than public health interests, especially as leaders meet privately with industry CEOs. The effort walks a tightrope between innovation and industry influence, raising questions about regulatory rigor. → Read the full article here.
🛰️ NEWS
What Else is Happening
🕹️ Gripen E Tests AI Co-Pilot: Saab’s fighter jet flew autonomously using Helsing’s Centaur AI, pulling off combat maneuvers in real airspace—no test-range limits, just real-time decision-making.
⚖️ Disney & Universal Sue Midjourney: Hollywood's biggest studios accuse the AI firm of mass copyright theft, citing images of Shrek, Spider-Man, and more—calling it “a bottomless pit of plagiarism.”
🔌 ChatGPT Outage Sparks Panic: OpenAI’s tool briefly went down Tuesday, revealing just how deeply millions of users rely on it—some saying they “no longer know how to work.”
♟️ ChatGPT Loses to 1970s Chess AI: A 1 MHz Atari 2600 emulator beat ChatGPT at chess, exposing how LLMs still flounder with game logic and board awareness—no strategy, just wordplay.
📉 US Tech Unemployment Hits 5.5%: IT joblessness rose for a fifth straight month as AI wipes out entry-level roles, especially in telecoms and compliance—demand now tilts hard toward LLM expertise.
🔬 RESEARCH
Why Smart Models Fail at Hard Problems: New Study Dissects Limits of AI “Reasoning”
A new study explores the limits of Large Reasoning Models (LRMs), AI systems designed to emulate human-like step-by-step thinking. Using precisely controlled puzzles, researchers showed that LRMs’ performance collapses on high-complexity tasks—despite having sufficient computational resources. Interestingly, their reasoning effort increases with problem difficulty only up to a point, then unexpectedly declines. → Read the full paper here.
📽️ VIDEO
This New Browser Agent Is Insane…
Runner H, an open-source web agent, automates online tasks—browsing, clicking, and creating docs—using AI that mimics human behavior, only faster. Get the full scoop in Matt’s latest video! 👇
🧰 TOOLBOX
Stronger Compliance, Smarter Farming, and Sharper Math Skills
🦾 Quantivate: Simplify GRC with tools for risk management, compliance tracking, and business continuity planning.
🌾 LEAF Agriculture: Connect farm data across systems with a secure API to boost decision-making, precision farming, and operations.
🧮 Woot Math: Boost K-12 math skills with interactive, adaptive tools that track progress and tailor lessons to student needs.
