OpenAI's $30B chip shopping spree 💸🧠

💰 OpenAI Drops $30B on Cerebras (That's Not a Typo)

OpenAI agreed to pay chip startup Cerebras over $20 billion across three years - potentially reaching $30 billion - plus warrants for up to a 10% equity stake. Cerebras is planning a $3B raise at a $35B valuation ahead of its IPO this quarter.
Why it matters: OpenAI is aggressively diversifying away from Nvidia for inference compute. The AI chip market just went from "one-horse race" to "okay maybe two horses and a very expensive pony."

🧬 OpenAI Built a Biology AI That Dunks on Actual Scientists

OpenAI released GPT-Rosalind, a specialized model for drug discovery and biological research that can read papers, query lab databases, design experiments, and generate hypotheses. On a blind RNA test from gene therapy lab Dyno Therapeutics, it scored better than 95% of human scientists. Amgen, Moderna, and the Allen Institute are already using it.
Why it matters: OpenAI is going full domain-specialist mode - first GPT-5.4-Cyber, now this. They're coming directly for AlphaFold's lunch, and they brought receipts.

🟣 Anthropic Ships Opus 4.7 - Coding Benchmarks in Shambles

Anthropic dropped Claude Opus 4.7, hitting 64.3% on SWE-bench Pro (up from 53.4%), beating GPT-5.4 and Gemini 3.1 Pro on agentic coding. But here's the thing - Anthropic's own unreleased Mythos Preview scores 77.8%. New features include an xhigh effort level and an /ultrareview command that flags bugs and design issues.
Why it matters: Anthropic is now running two parallel tracks - a public release cycle and an exclusive preview pipeline for enterprise/government clients. It's giving... tiered access to superintelligence energy.

🖥️ OpenAI Turns Codex Into a "Never Leave This App" Super App

OpenAI overhauled Codex into a unified ChatGPT + Atlas + Codex app with background computer use, parallel agents, an in-app browser, image generation, and persistent memory. It now has 3 million weekly users with 70% month-over-month growth. Background computer use lets it operate any Mac app autonomously - no API needed.
Why it matters: This is OpenAI's answer to Claude Code and Cowork. The vision?? Codex wants to be the app developers never close. Your IDE is now a lifestyle.

🌊 Windsurf 2.0: Your IDE Is Now a Project Management Tool

Windsurf 2.0 shipped an Agent Command Center - a Kanban-style view for managing fleets of parallel cloud and local agents - with Devin (the autonomous coding agent) now embedded directly in the IDE. Windsurf Spaces organize agent sessions, PRs, and files by task with automatic context carryover.
Why it matters: The future developer is less "person who writes code" and more "person who manages a team of AI agents from a Kanban board." We're all project managers now. No one is safe.

🏛️ Google and the Pentagon Are Talking Classified Gemini Deployment

Google is negotiating with the Department of Defense to deploy Gemini models in classified settings, covering all lawful uses. Google proposed contract language to prevent use for mass surveillance or autonomous weapons without human oversight.
Why it matters: Google went from "don't be evil" to "we'll deploy in classified military settings but like... responsibly." The guardrail language around surveillance and weapons is going to get scrutinized HARD.

🛒 AI Shopping Traffic Up 393% - And It Actually Converts

AI-referred traffic to U.S. retail sites rose 393% year-over-year in Q1 2026 (Adobe data, 1 trillion+ visits). AI visitors now convert 42% better than human shoppers - a complete reversal from last year. AI-driven revenue per visit is 37% higher, and shoppers arriving via AI spend 48% longer on sites. But ~34% of product pages still can't be properly accessed by AI.
Why it matters: AI isn't just changing search - it's rewriting e-commerce economics. Retailers who optimize for LLM accessibility win. Everyone else?? Invisible.

🎨 Canva's AI Assistant Can Now Just... Do Everything

Canva launched AI 2.0 - the assistant now calls tools autonomously to create editable, layered designs from text prompts, plus integrations with Slack, Gmail, Google Drive, Calendar, and Zoom. Its image model is 5x faster and 30x cheaper; video is 7x faster and 17x cheaper.
Why it matters: Adobe launched Firefly assistant and Figma shipped MCP support the same week. The design tool wars just went full battle royale. Canva is betting it's the "last mile" where creative work actually gets finished.

🔍 Google Lets You Browse Websites Inside AI Mode

Google rolled out side-by-side browsing in AI Mode on Chrome desktop - open a webpage next to your AI conversation and ask follow-ups without losing context. A new "plus" menu lets you pull in content from your open Chrome tabs to enrich searches.
Why it matters: Google is merging search and browsing into one AI-assisted workflow. The line between "googling something" and "having an AI research assistant" just got very blurry.

🏦 Bank of England Is Stress-Testing AI Before It Breaks Finance

The Bank of England is running scenario analysis and simulations to test how AI could destabilize financial markets - focusing on "herding" behavior where AI agents amplify selloffs during market stress. This comes after Anthropic's Mythos launch raised concerns about AI-boosted cybersecurity vulnerabilities in banking.
Why it matters: Central banks aren't watching from the sidelines anymore - they're actively stress-testing for systemic risk. Turns out "what if all the trading AIs panic at the same time" is a question worth answering before it happens.

🎮 Roblox's AI Assistant Can Now Build AND Playtest Your Game

Roblox revamped its AI Assistant with Planning Mode - it analyzes code, asks clarifying questions, and creates editable action plans before building. New tools include Mesh Generation for instant 3D objects and autonomous playtesting that captures screenshots, uses inputs, and fixes bugs on its own.
Why it matters: Roblox is making game dev accessible to people who think in ideas, not code. An AI that builds your game and QA tests it?? The junior dev pipeline is... having a moment.

🎬 Runway CEO: Make 50 AI Films for the Price of One Blockbuster

Runway CEO Cristóbal Valenzuela suggested studios take a $100M blockbuster budget and spread it across 50 AI-assisted films to increase hit odds. The first studio-quality AI feature film, "Bitcoin: Killing Satoshi," reportedly cut production costs from $300M to $70M. Amazon, Sony, and Indian studios are also adopting AI.
Why it matters: Whether this is "democratization of filmmaking" or "a race to produce 50 mid movies instead of one good one" depends entirely on your level of optimism. Either way, the economics of content creation are getting rewritten.

💳 AmEx Buys Sam Altman-Backed AI Expense Startup Hyper

American Express is acquiring Hyper, an AI expense management startup backed by Sam Altman, that builds AI agents to categorize expenses, file reports, and check compliance automatically. AmEx CEO Stephen Squeri called AI a "structural shift." Deal expected to close Q2 2026.
Why it matters: Big finance is moving from "AI chatbot on the website" to "AI agents running back-office operations." The Sam Altman investment connection is just chef's kiss for the narrative.

🚗 Stellantis and Microsoft Sign a Five-Year AI Deal

Stellantis (Jeep, Peugeot) and Microsoft agreed to a five-year partnership covering 100+ AI initiatives - product development, predictive maintenance, cybersecurity, digital services. Stellantis will migrate to Azure and targets a 60% reduction in data center footprint by 2029, plus an AI-driven global cyber defense center.
Why it matters: Legacy automakers are outsourcing their tech brains to Big Tech because Chinese EV makers are racing ahead on software. These partnerships aren't innovation plays - they're survival strategies.

📊 Ericsson's Profit Misses Because AI Is Eating All the Chips

Ericsson's Q1 core profit came in below expectations at 5.2 billion Swedish crowns, blaming rising semiconductor costs driven by AI demand and a North America sales slowdown. CEO Börje Ekholm said customers will need to share the cost burden. Shares fell 1.6%.
Why it matters: Here's the flip side of the AI boom nobody talks about - companies outside the AI bubble are paying more for the same chips everyone wants. AI's appetite for semiconductors is creating collateral damage across industries.

🤖 Antioch Raises $8.5M to Be "Cursor for Physical AI"

Startup Antioch raised an $8.5M seed at a $60M valuation to build simulation tools that help robotics companies train autonomous systems in virtual environments instead of expensive physical test arenas. They're focused on closing the "sim-to-real gap" - making virtual physics realistic enough that robots trained in simulation actually work IRL.
Why it matters: Physical AI is bottlenecked by data scarcity and testing costs. If Antioch nails this, it could do for robotics what cloud computing did for software - make it scalable and affordable. Big if, but big payoff.

🔭 InsightFinder Raises $15M to Debug AI Agents When They Break

InsightFinder AI raised a $15M Series B for its platform that monitors, diagnoses, and fixes issues across AI models, data pipelines, and infrastructure in live production. Customers include UBS, NBCUniversal, Lenovo, Dell, and Google Cloud. Revenue grew over 3x in the past year.
Why it matters: Everyone's deploying AI agents. Almost nobody knows what to do when they break in production. InsightFinder is betting that observability for AI means understanding the entire stack - not just staring at the model and hoping.

🌐 Tencent Open-Sources a Physics-Aware 3D World Model

Tencent's Hunyuan team open-sourced HY-World 2.0, a world model that generates editable 3D scenes with physics-aware movement - objects actually behave according to realistic physics. Useful for game dev, simulation, and robotics training.
Why it matters: Free, open-source world models with actual physics?? This could accelerate everything from gaming to physical AI training. Tencent just handed developers a very expensive gift.

🦪 Alibaba Drops "Happy Oyster" - Interactive 3D Worlds From Text

Alibaba's ATH team introduced Happy Oyster in beta, a world model that creates interactive 3D environments on the fly from text, images, and other multimodal inputs - positioning it as a direct competitor to Tencent's HY-World.
Why it matters: The world model race between Chinese tech giants is heating up FAST. Both Tencent and Alibaba are making these tools accessible, and the rest of the industry is watching closely.

🚀 Upscale AI Raising at $2B Valuation - With No Product

AI infrastructure startup Upscale AI is reportedly raising $180M-$200M at a $2 billion valuation - its third round in just seven months. Investors include Tiger Global, Xora Innovation, and Premji Invest. The company focuses on custom chips and infrastructure but has yet to release a product.
Why it matters: $2 billion. No shipping product. Seven months old. This is either visionary investing or the most expensive game of "trust me bro" in Silicon Valley history. Possibly both??

👟 Sneaker Company Rebrands as AI Firm, Stock Surges 600%

Sneaker brand Allbirds surged 600%+ after rebranding as "NewBird AI," while social media firm Myseum jumped 150% just by adding "AI" to its name. This echoes Long Island Iced Tea becoming Long Blockchain in 2017 - though the article notes CoreWeave's pivot from Ethereum miner to $23B cloud company was actually legit.
Why it matters: A name change isn't a business model. But apparently it IS a stock catalyst?? The AI gold rush has its share of fool's gold, and Wall Street keeps buying it. Some things never change...