😳 CoreWeave's $21B deal with Zuck

💰 CoreWeave Signs a $21 Billion Deal with Meta (Yes, Billion with a B)

CoreWeave just locked in a $21 billion expanded agreement to provide GPU cloud capacity to Meta through December 2032. Their stock jumped nearly 8% in premarket because... obviously.
Why it matters: This is one of the largest AI infrastructure deals ever announced. CoreWeave went from "who??" to sitting at the table with the hyperscalers. Meta's GPU appetite is insatiable and apparently only getting worse. (source)

📊 Anthropic May Have Just Overtaken OpenAI in Revenue

Anthropic's annualized revenue reportedly hit $30 billion, potentially eclipsing OpenAI's self-reported $24 billion. At the start of 2025, OpenAI led 6-to-1. SIX. TO. ONE.
The surge?? Driven by coding agents like Claude Code and enterprise plug-ins that wiped out $1 trillion from software stocks globally back in February.
Why it matters: Turns out developer token volume > casual chatbot users when it comes to actual money. Both companies are eyeing IPOs in the second half of 2026, and this revenue race is basically their pre-game trash talk. OpenAI is now pivoting hard toward enterprise and coding tools like Codex because... that's where the bag is. (source)

🧠 Meta's Superintelligence Labs Drops Muse Spark (And It's... Proprietary??)

Meta's Superintelligence Labs - led by Alexandr Wang after Zuck's $14.3 billion acquisition of Scale AI - just released Muse Spark, a multimodal reasoning model that handles voice, text, and image inputs with a "contemplating mode" that pits multiple agents against each other. It's giving internal debate club energy.
Benchmarks are competitive with Opus 4.6 and GPT 5.4 on reasoning, lags in coding, but is especially strong in health reasoning - aligning with Meta's "personal superintelligence" mission.
Why it matters: Here's the plot twist - unlike Meta's open-source Llama family, Muse Spark is proprietary. The company that built its entire AI brand on open-source just... didn't. This is the first real output from the team Wang built from scratch, and it signals Meta is serious about competing at the frontier, not just giving models away for free. (source)

🤖 Anthropic Launches Claude Managed Agents (a.k.a. "Deploy an Agent in Days, Not Months")

Anthropic opened a public beta for Claude Managed Agents - a platform that takes you from agent idea to live product in days, handling all the backend plumbing that used to take months of engineering pain.
Early adopters include Notion, Rakuten, Asana, and Sentry. Rakuten reportedly set up agents across five departments in about a week each. Sessions cost just $0.08/hour on top of standard AI usage fees.
Why it matters: Anthropic is eating the traditional software development stack alive. Autonomous agents that work solo for hours, coordinate with each other, and cost eight cents an hour?? That's not a chatbot - that's a full-blown enterprise platform. The SaaS industry is lowkey sweating. (source)

⚖️ Court Refuses to Block Pentagon's Anthropic Blacklisting

A D.C. federal appeals court denied Anthropic's request to pause the Pentagon's national security blacklisting while the case plays out - a win for the Trump administration.
Anthropic says the designation could cost it billions in lost business. The blacklisting came after Anthropic refused to let the military use Claude for surveillance or autonomous weapons due to safety concerns.
Why it matters: This is a landmark clash between AI safety principles and government power. A California judge already blocked a separate order, calling it unlawful retaliation. Split rulings = this fight is far from over. The outcome will set precedent for whether AI companies can actually say "no" to the government... or if that's just a nice idea. (source)

🔥 Musk Amends OpenAI Lawsuit - "Give the Money to the Nonprofit, Not Me"

Musk amended his lawsuit against OpenAI to redirect any of the $150 billion+ in damages to OpenAI's nonprofit arm instead of himself. He's also pushing to remove Sam Altman from the nonprofit's board.
OpenAI called the move part of "a harassment campaign." Trial is expected later this month in Oakland, California.
Why it matters: By redirecting damages to the nonprofit, Musk is trying to cosplay as a principled crusader rather than a billionaire chasing a payday. But with the trial imminent and OpenAI's IPO on the horizon, the legal and PR stakes couldn't be higher for both sides. This soap opera has more seasons than Grey's Anatomy. (source)

📈 OpenAI Will Let Regular People Buy IPO Shares

OpenAI CFO Sarah Friar confirmed the company plans to reserve a portion of its IPO shares for individual investors after seeing "really strong demand" from retail during its latest funding round.
They raised over $3 billion from individual investors in the most recent round - three times the original $1B target - at an $852 billion valuation.
Why it matters: This follows Musk's plan to allocate up to 30% of SpaceX's IPO to retail investors. If OpenAI follows suit, it could reshape how mega-IPOs are structured - giving everyday investors a real seat at what could be a $1 trillion listing. Your Robinhood account just got interesting. (source)

🚀 Perplexity Hits $450M ARR After 50% Monthly Revenue Jump

Perplexity hit an estimated $450 million in annual recurring revenue after a 50% month-over-month jump, driven by its Computer agentic system and usage-based pricing.
The growth marks a clear pivot from search to AI agents, with the company finding a lucrative niche in agentic workflows.
Why it matters: Perplexity is proving you don't need to be OpenAI or Anthropic to build a massive AI business. $450M ARR puts them in serious IPO territory. The "Google killer" meme is becoming... less of a meme?? (source)

🔬 Musk: Seven New Models Training on Colossus 2 (Including a 10T Parameter Monster)

Musk revealed xAI has seven new models currently training on its Colossus 2 supercomputer, including 6 trillion and 10 trillion parameter systems.
These would be among the largest AI models ever built, dwarfing current frontier models in raw parameter count.
Why it matters: Despite losing all 11 original co-founders (sir??), Musk is doubling down on sheer scale as xAI's competitive advantage. Whether bigger = better is still debated, but 10 trillion parameters is the kind of number that makes GPU manufacturers cry tears of joy. (source)

🏗️ Bezos' Secret AI Startup Prometheus Poaches xAI Co-Founder

Kyle Kosic, former xAI co-founder who led infrastructure for Musk's Colossus supercomputer before moving to OpenAI, has now joined Jeff Bezos' Project Prometheus.
Prometheus is building AI that understands the physical world and laws of physics, targeting industries like jet engine design, aviation, and architecture. They've reportedly assembled the largest corpus of engineering data ever.
Why it matters: Bezos isn't building another chatbot - he's going after industrial AI, a largely untapped market. The "Berkshire Hathaway-type holding company" model of acquiring stakes in companies and deploying AI engineers inside them is a completely different playbook from anything else in the AI race. This is lowkey the most fascinating story nobody's talking about. (source)

🎭 HeyGen's Avatar V: 15 Seconds of You → A Digital Twin

HeyGen launched Avatar V, which builds a full video avatar from just a 15-second phone recording - capturing real facial details, gestures, and micro-movements. It reportedly outperformed Google's Veo 3.1 on accuracy and lip sync.
The model separates identity from appearance for the first time - record once, then swap outfits and backgrounds without re-filming.
Why it matters: AI avatars went from janky mouth movements to virtually indistinguishable digital twins. Whether you love it or find it deeply unsettling... this tech is about to transform content creation, marketing, and communications. We love to see it (and also we're terrified). (source)

🎨 Canva Acquires Simtheory and Ortto - Quietly Building an AI Empire

Canva acquired Simtheory (agentic AI workspace platform) and Ortto (marketing automation company with 11,000+ customers across 190 countries).
Simtheory accelerates Canva's evolution from a design platform with AI tools to an AI platform with design tools. Ortto powers the full marketing lifecycle - creation to publishing to optimization.
Why it matters: Canva is quietly building an end-to-end AI-powered work platform. Combined with earlier buys like MagicBrief and MangoAI, they want to own the entire creative and marketing workflow. More coming at Canva Create on April 16. It's giving... monopoly vibes. (source)

🏢 Citigroup Uses AI to Cut Document Review from 1 Hour to 15 Minutes

Citi is using AI to slash document review time from over an hour to just 15 minutes for account openings, and to migrate data from legacy systems.
The bank is also cutting IT contractor reliance from 50% to 20% of its tech workforce, hiring more internal software engineers across a 50,000-person tech team.
Why it matters: Textbook example of AI delivering real, measurable productivity gains in a heavily regulated industry. As Citi automates its first 50 critical processes - including KYC and onboarding - it's showing how enterprise AI adoption actually works at scale. Not just vibes. Actual results. (source)

🇦🇺 Bendigo Bank Posts 12.8% Earnings Jump, Announces AI-Linked Job Cuts

Australia's Bendigo and Adelaide Bank posted a 12.8% jump in quarterly earnings and confirmed job cuts tied to new partnerships with Infosys and Genpact focused on AI-driven automation.
The partnerships are expected to deliver up to A$75 million in annual savings by fiscal 2028, equivalent to about 11.5% of staff costs.
Why it matters: Part of a global wave of AI-linked layoffs hitting banking and tech. The pattern is Dramatic. And. Clear. - AI investment goes up, headcount comes down. Regional banks especially need these efficiency gains to compete against larger players with more scale. (source)

📊 Applied Digital Revenue Surges 139% - But Losses Widen Too

Applied Digital's quarterly revenue soared 139% to $126.6 million, crushing estimates of $76.6M, driven by demand for its AI data center services.
But net loss widened to 36 cents per share due to nearly tripled costs and a $59.7 million impairment charge tied to its cloud business.
Why it matters: Classic AI infrastructure growth story - massive revenue growth paired with even bigger spending. With its 300-megawatt Delta Forge 1 campus under construction, Applied Digital is betting that insatiable demand for AI compute will eventually turn these losses into profits. The "spend now, profit later" era continues... (source)

🛡️ OpenAI Publishes a Child Safety Blueprint

OpenAI released a Child Safety Blueprint pushing for updated U.S. laws on AI-generated CSAM, stronger reporting requirements, and built-in safeguards to prevent exploitation.
The framework was developed with NCMEC, Thorn, and state Attorneys General, focusing on three pillars: modernizing laws, improving provider reporting, and safety-by-design measures in AI systems.
Why it matters: As generative AI makes it easier to create harmful content, this is one of the first comprehensive policy frameworks from a major AI company. Whether it leads to actual legislation remains to be seen, but it sets a benchmark for responsible AI development that others will be measured against. FINALLYYYY, some actual responsibility :) (source)