Welcome back to 4IR. Here's today's lineup:
Oracle stock explodes 36% as Larry Ellison briefly becomes world's richest - AI cloud deals worth $455 billion shock Wall Street
Replit raises $250M at $3B valuation for AI coding platform - Google backs startup that lets anyone code without learning to code
PsiQuantum gets $1B to build quantum computer with a million qubits - BlackRock, NVIDIA bet big on light-powered quantum chips
Scientists say medical AI needs transparency NOW - UW team shows COVID AI models that claimed 99% accuracy were useless
CuspAI raises $100M to find new materials 10x faster - AI will design batteries and chips atom by atom
🔥 TOP STORY: Oracle jumps 36% as founder briefly passes Elon Musk
The story: Oracle had its best stock day since 1992 on September 10, jumping 36% after revealing it has $455 billion in signed AI cloud contracts—up 359% from last year. Larry Ellison's wealth shot up by $90-100 billion in one day, briefly making him richer than Elon Musk at $393 billion. The company says its cloud revenue will grow from $18 billion to $144 billion by 2031. Here's the kicker: Oracle actually missed earnings targets. But investors don't care when you're sitting on half a trillion dollars in future AI deals.
What we know:
Stock up 36% on September 10 (best since 1992)
$455 billion in signed contracts (up 359%)
Ellison gained $90-100 billion in one day
Briefly world's richest at $393 billion
Cloud revenue going from $18B to $144B
Company now worth $922 billion
Why it matters: Oracle just became the grocery store of AI—everyone needs what they're selling, and they've locked in the customers for years.
This is insane. One human being made $100 billion in a day. That's more than the GDP of Morocco. Oracle was supposed to be dead—an old database company that missed the cloud revolution. Instead, they're selling the picks and shovels for the AI gold rush. Every AI company needs massive computing power. Oracle has it, and they've already locked in half a trillion in contracts. The Trump connection helps too—Ellison was just at the White House announcing AI projects. While OpenAI bleeds money and Google scrambles, Oracle is printing cash. Boring beats sexy when boring has guaranteed revenue.
🧠BREAKTHROUGH: Replit makes everyone a coder with $250M raise
The story: Replit raised $250 million on September 10, tripling its value to $3 billion. The startup lets anyone build software by just describing what they want—no coding knowledge required. Their new Agent 3 feature goes even further: it writes code, tests it, fixes bugs, and deploys apps completely on its own. Google's AI Fund joined the round, along with Amex. Everything runs in your browser—no downloads, no setup. Just open a tab and start building.
What we know:
$250 million raised September 10
Worth $3 billion (3x last valuation)
Led by Prysm Capital
Google and Amex invested
Agent 3 builds apps autonomously
Works entirely in web browser
Why it matters: This is the end of "learning to code." Why spend years learning programming when AI can do it instantly?
Replit isn't just disrupting coding—it's deleting it. Agent 3 is wild: you say "build me a recipe app," and it creates the whole thing while you watch YouTube. No Stack Overflow, no debugging, no crying at 3 AM. The browser-based approach is perfect. My grandma could build an app on her iPad. Google's investment shows they get it—this is how normal people will create software. Every business owner who said "I wish I had an app for that" can now build it themselves. The $3 billion valuation seems huge until you realize they're replacing a $500 billion industry. Coding bootcamps are toast. Maybe even computer science degrees.
💰 MOONSHOT: PsiQuantum's $1B plan for a million-qubit computer
The story: PsiQuantum raised a massive $1 billion on September 10, valued at $7 billion, to build a quantum computer with one million qubits. BlackRock led the round with NVIDIA, showing that serious money believes quantum computing is finally real. Instead of using frozen atoms like other quantum companies, PsiQuantum uses light particles on regular silicon chips. They're building huge facilities in Chicago and Brisbane to test their systems. For context: today's best quantum computers have maybe 1,000 qubits. PsiQuantum is shooting for 1,000 times that.
What we know:
$1 billion raised September 10
$7 billion valuation
BlackRock, NVIDIA, Temasek led
Building million-qubit computer
Uses light instead of frozen atoms
New facilities in Chicago and Brisbane
Why it matters: A million-qubit computer would break all current encryption and solve problems that would take normal computers millions of years.
Everyone else is playing with quantum toys. IBM has 433 qubits. Google has 70. PsiQuantum says "screw it, we're building a million." Using light is genius—it works at room temperature while everyone else needs freezers colder than space. They can use normal chip factories too, no exotic equipment needed. BlackRock doesn't throw billions at science experiments. They see something real here. When this works, it changes everything: drug discovery in days not decades, materials that seem impossible today, AI that makes current models look like pocket calculators. Oh, and it breaks all encryption. Sleep tight.
📰 BATTLEGROUND: Medical AI is already in hospitals—and it's failing
The story: University of Washington researchers dropped a bomb on September 10 (when their research was covered in UW News): medical AI is being used on patients right now, and nobody knows if it actually works. Their Nature paper shows hundreds of COVID diagnosis AIs claimed 99% accuracy in tests but completely failed in real hospitals. Half of doctors surveyed are already using AI tools despite constant errors and biases. The team demands total transparency—show your data, admit your failures, explain your limits—before AI touches another patient.
What we know:
UW News coverage on September 10
Published in Nature Reviews Bioengineering
COVID AIs claimed 99% accuracy, failed in practice
Half of 2,206 doctors already use AI
Calling for mandatory transparency
FDA regulations years behind
Why it matters: Your doctor might be using AI to diagnose you right now. Nobody's checking if it works on people like you.
This is terrifying. We're beta-testing AI on sick people. The COVID example is perfect—hundreds of "99% accurate" AI models that were basically coin flips in real hospitals. Why? They trained on data from rich hospitals, then failed at community clinics. They worked on white patients, failed on everyone else. Your radiologist might be using AI to read your scan. Your doctor might be using ChatGPT to diagnose symptoms. Nobody's tracking if it works. The FDA treats AI like it's a medical device you can test once and approve. But AI changes every time it's updated. We need transparency now, before AI kills someone and everyone panics.
🚀 VELOCITY: CuspAI will design new materials like Google searches the web
The story: CuspAI raised $100 million on September 10 to build an AI that discovers new materials 10 times faster than current methods. The company is worth $520 million, with investments from Hyundai, Samsung, and NVIDIA—companies desperate for better batteries, chips, and manufacturing materials. Instead of spending years mixing chemicals hoping something works, CuspAI's AI designs materials atom by atom on computers first. The advisors are insane: Geoffrey Hinton (who basically invented modern AI) and Yann LeCun (Meta's chief AI scientist).
What we know:
$100 million raised September 10
Worth $520 million
Hyundai, Samsung, NVIDIA invested
10x faster than traditional research
Designs materials atom by atom
Hinton and LeCun advising
Why it matters: Every future breakthrough—better batteries, faster chips, carbon capture—needs new materials. CuspAI is building the tool to find them all.
Forget chatbots. This is AI doing actual science. While everyone argues about whether AI can write emails, CuspAI is designing materials that don't exist yet. The investors tell you everything: Hyundai needs miracle batteries for EVs. Samsung needs chips that don't melt. NVIDIA needs materials that can handle the heat of nuclear reactor-level computing. This is like having a search engine for materials that don't exist yet. Type in "battery that charges in 1 minute" and it designs the chemistry. The $520 million valuation is laughably low if this works. They're not building a product; they're building a material discovery machine that every manufacturer on Earth will need.
Note: Commentary sections are editorial interpretation, not factual claims