Meta's gigawatt chip flex 💪🔌

🛡️ OpenAI vs. Anthropic: The Cybersecurity Cage Match

OpenAI launched GPT-5.4-Cyber, a model fine-tuned for defensive cybersecurity, and shipped it to thousands of verified defenders through its Trusted Access for Cyber program. It can reverse-engineer compiled software to flag malware without needing source code. Meanwhile Anthropic's Mythos is locked down to roughly 40 trusted partners under "Project Glasswing" - and Treasury Secretary Bessent literally summoned Wall Street leaders for an emergency briefing on its hacking capabilities.
Why it matters: Two polar opposite strategies - OpenAI says "arm the defenders," Anthropic says "lock it in a vault." Either way, AI-powered cybersecurity just stopped being theoretical and started being a whole geopolitical chess match.

📈 ASML Raises Forecast Because Chip Demand Is Literally Insatiable

ASML beat Q1 earnings and bumped its 2026 revenue outlook to €36-40 billion. CEO Christophe Fouquet said it plainly: "Demand for chips is outpacing supply." They're shipping 60 low-NA EUV tools this year (25% more than 2025), shares are up 40% YTD, and briefly touched a record above €1,300. The one asterisk?? Potential new U.S. export restrictions on China could drag things toward the low end.
Why it matters: ASML makes the machines that make the chips that make the AI. When that company says supply can't keep up with demand, the AI infrastructure buildout is NOT peaking anytime soon.

🔐 Federal Agencies Sneak Around Trump's Anthropic Ban to Test Mythos

Despite President Trump's ban on working with Anthropic, federal agencies are quietly testing Mythos anyway. The Commerce Department's Center for AI Standards and Innovation is evaluating its hacking capabilities, and staff on at least three congressional committees have held or requested Anthropic briefings. Co-founder Jack Clark confirmed the company is in talks with the Trump administration, calling it a "narrow contracting dispute."
Why it matters: When government agencies go around their own president to get access to an AI model... that tells you everything about how significant Mythos is. Politics vs. national security is creating the messiest AI governance situation we've seen yet.

🔬 Anthropic's Jack Clark: "The Government Has to Know About This Stuff"

Anthropic co-founder Jack Clark confirmed at the Semafor World Economy event that the company briefed the Trump administration on Mythos - while simultaneously suing the Department of Defense over being labeled a supply-chain risk. On AI and jobs, Clark said Anthropic is only seeing "some potential weakness in early graduate employment" so far, and advised students to focus on "synthesis across subjects and analytical thinking" because AI gives you the experts - you just need to know what to ask.
Why it matters: Anthropic is suing the Pentagon AND briefing the White House at the same time. That's not a strategy, that's a situationship with the federal government. But it shows how badly AI companies and governments need each other, even when they're fighting.

🤖 Anthropic's Revenue Went From $9B to $30B and OpenAI Investors Are Sweating

Anthropic's annualized revenue jumped from $9 billion to $30 billion between late 2025 and March 2026, mostly from coding tools. Meanwhile, OpenAI's $852 billion valuation is getting side-eyed - one investor told the Financial Times you'd need to assume an IPO valuation of $1.2 trillion+ to justify the round. On the secondary market, Anthropic shares are "nearly insatiable" while OpenAI shares are trading at a discount. Iconiq Capital's Roy Luo, whose firm put over $1B into Anthropic, said: "There is fundamentally a number one and a number two dynamic… We picked."
Why it matters: The private market is already voting with its dollars, and right now it's voting Anthropic. OpenAI's valuation is starting to look like it needs a reality check - or an IPO miracle.

🏦 Wall Street Has $108B in Private Credit Exposure and AI Is Making Them Nervous

JPMorgan ($50B), Wells Fargo ($36.2B), and Citigroup ($22B) disclosed $108 billion in combined private credit exposure and are stress-testing portfolios. The $3.5 trillion private credit market is getting squeezed by AI disruption threatening software portfolios, fund outflows, and a record 9.2% default rate among U.S. corporate borrowers in 2025. Jamie Dimon says "I don't think it's systemic." BlackRock's Larry Fink calls demand "structural."
Why it matters: AI isn't just building things - it's breaking business models fast enough to crater loan portfolios. The second-order financial effects of the AI revolution are here, and Wall Street is only now starting to stress-test for them. Comforting!!

💊 Amazon Built an AI Drug Discovery Platform That Actually Works

AWS launched Amazon Bio Discovery, an AI platform that lets scientists run drug discovery workflows without writing code. It includes biological foundation models that generate and evaluate potential drug molecules. In a collab with Memorial Sloan Kettering Cancer Center, it generated nearly 300,000 novel antibody molecules, narrowed them to 100,000 candidates for lab testing - compressing months of work into weeks. Early adopters include Bayer, the Broad Institute, and Voyager Therapeutics.
Why it matters: Drug discovery is where AI could deliver the most tangible real-world impact. Amazon just removed the coding bottleneck so any researcher - not just ML wizards - can use AI-driven drug design. FINALLYYYY, a use case worth getting excited about.

💪 Meta Signs Gigawatt Chip Deal with Broadcom Through 2029

Meta expanded its partnership with Broadcom for several generations of custom AI processors, committing to over one gigawatt of computing capacity - enough to power ~750,000 U.S. homes. Broadcom CEO Hock Tan will leave Meta's board and move to an advisory role on chip strategy. Zuckerberg said it'll help "build out the massive computing foundation we need to deliver personal superintelligence to billions of people." Meta's custom chip program (MTIA) already powers its ranking systems, with three more chip generations due through 2027.
Why it matters: This is Meta saying "we're done being dependent on Nvidia" in the loudest way possible. A multi-gigawatt custom silicon rollout is the kind of commitment that doesn't have a Plan B. Zuck is ALL in.

⚛️ Nvidia Drops Open-Source AI Models for Quantum Computers

Nvidia released Ising - the first family of open-source AI models designed to work with quantum computers. One model turns a days-long manual calibration job into hours, another fixes errors at 2.5x the speed and 3x the accuracy of the best open-source alternative. Jensen Huang called it "the operating system of quantum machines." Over 20 institutions are using it at launch, including Harvard, Cornell, Fermilab, and IonQ.
Why it matters: Nvidia is running the same playbook it used for self-driving - build the software layer that makes the hardware useful, then become impossible to replace. If quantum computing is going to escape the lab, AI might be the key... and Nvidia just made sure it's holding that key.

🧑‍💻 Claude Code Gets a Full Redesign - It's Giving Command Center

Anthropic overhauled Claude Code's desktop app with a new sidebar for managing multiple AI sessions, drag-and-drop panes, and an integrated editor and terminal. Devs can run tests, edit files, review changes, and preview outputs all without switching tools. They also launched "routines" - AI tasks that run on a schedule, via API, or triggered by GitHub events like pull requests.
Why it matters: This isn't an IDE anymore - it's a command center for managing a half-human, half-AI dev team. Anthropic is betting that the future of coding is less "writing code" and more "managing agents that write code for you." Honestly?? They might be right.

🌐 Google Turns Chrome Into an AI Operating System with "Skills"

Google launched Skills, a Chrome feature that lets you save and reuse AI prompts across web pages. Save a prompt once, trigger it with / or + anywhere. Early uses include calculating protein macros in recipes, shopping comparisons, and summarizing long documents. A Skills library with pre-built workflows launches today for desktop Chrome users signed into their Google account.
Why it matters: OpenAI has Atlas. Perplexity has Comet. Google's response?? Turn Chrome itself into a personalized AI workflow engine. The browser wars just became the AI browser wars, and Google is not about to lose its home turf.

🇮🇳 Google Brings "Personal Intelligence" Gemini to India

Google launched Personal Intelligence for Gemini in India - users can connect Gmail, Photos, and YouTube and ask things like "What are my travel plans for Jaipur?" Available first to AI Pro and AI Ultra subscribers, with free users coming soon. Google did note that Gemini doesn't always get context right - it might assume you love golf because you're in hundreds of golf course photos (when really, you just love your son who plays golf).
Why it matters: India is one of Google's biggest markets, and seeding advanced personalized AI there is a play to make Gemini indispensable globally. A chatbot that actually knows you is powerful. Also a little terrifying. But mostly powerful :)

📱 Apple Keeps Nuking Vibe-Coding Apps from the App Store

Apple has been removing or blocking updates for vibe-coding apps - tools that let users build apps with AI - citing rules against apps that download or execute code. Anything has been removed twice, while Replit and Vibecode have had updates paused. Anything is now pivoting to desktop companion apps and iMessage-based building, and may shift focus to Android entirely. Meanwhile Apple saw an 84% jump in app submissions in one quarter thanks to AI coding tools. Epic's Tim Sweeney called the crackdown "abhorrent to the founding principles of Apple."
Why it matters: AI is making it possible for anyone to build apps, and Apple's walled garden is fighting it every step of the way. But when your own App Store submissions jump 84% because of the thing you're banning... something's gotta give.

🇰🇷 Korean Startup DEEPX Is Coming for Nvidia's Robotics Chip Crown

South Korean startup DEEPX is expanding its partnership with Hyundai Motor Group to build a computing platform for generative AI-powered robots using next-gen low-power chips. DEEPX's current chips are reportedly 20x more power-efficient and much cheaper than Nvidia's Jetson Orin. Its upcoming DX-M2 chips, built on Samsung's 2-nanometer process, go into volume production next year. Hyundai plans to build a factory manufacturing 30,000 robot units annually by 2028. DEEPX is also raising over $408 million ahead of a potential IPO.
Why it matters: The race for on-device AI in robotics isn't just a Silicon Valley thing anymore. DEEPX's ultra-efficient chips could solve one of humanoid robotics' biggest problems - keeping energy-hungry AI from literally overheating the machines - while giving Nvidia real competition at the edge.

💰 AI Hedging Platform Pillar Raises $20M to Democratize Financial Risk

Pillar, a platform that automates hedging for commodity-driven businesses like metals traders and airlines, raised a $20 million seed round led by Andreessen Horowitz, with participation from Uber CEO Dara Khosrowshahi. The platform uses AI to ingest data from contracts, cash flows, inventories, ERP software, and even WhatsApp messages to analyze exposure across commodities, FX, and freight - then builds and manages hedge portfolios automatically.
Why it matters: Hedging has always been a luxury for big institutions with expensive tools. Pillar wants to make it as accessible as accounting software for small and mid-sized businesses. In a world of wild geopolitical volatility, that's genuinely useful - not just another AI wrapper.

📰 Quick Bites

AI personal finance startup Hiro is winding down - its staff is joining OpenAI. Acqui-hire speedrun complete.
UK AI safety evaluators said Claude Mythos Preview is the first AI to complete their 32-step corporate hack simulation, showing big jumps in cyber-attack capability over Opus 4.6. Cool cool cool.
Baidu released ERNIE-Image, an 8B open-weight text-to-image model that nears top rivals on benchmarks despite being tiny. Small but scrappy.
OpenAI's Greg Brockman framed AI as the shift to a "compute-powered economy," claiming roughly 1 billion weekly ChatGPT and Codex users on the company's 10th anniversary. One. Billion. Weekly.