In partnership with
Introducing our new weekly AI newsletter.
There’s been a change in our regular programming. Introducing our new weekly AI newsletter, which will be hitting your inbox every Friday.
We'll be scouring the net for the most important headlines globally, which reflect how fast the world of frontier tech is really moving.
The smartest sector intelligence will land straight to your inbox, for free. Probably without the intelligence. And yeah, not near the smartest. But we promise we’ll make it interesting.
So, without further ado, let's get into it. This week we're covering…
Anthropic’s headline-grabbing antics 🗞️ Musk’s Pentagon deal 🇺🇸 Perplexity Computer arrives 🖥️ OpenClaw runs amok 🦞 Substack’s stock killer ☢️ Nano Banana 2 is here 🍌 Top tools, models, tricks 📲 Major financials 💸
The blocmates. Weekly AI Newsletter will now be dropping every Friday right here. If you still want your daily dose of blocmates’ crypto-focused goodness, head over to our Telegram to stay in the loop.
P.S. There’s a lot in this newsletter – we suggest getting comfy and getting stuck in through your desktop >> hit the READ ONLINE tab at the top of this mailer.
THE WEEK IN AI
Our weekly news run has had a revamp. Get familiar with our news tags, so you can get straight to the point like the simpletons we are.
..THE MAIN EVENT .
#1 Anthropic dominates headlines
If there’s one AI company that broke the news cycle this week, it was Anthropic. Like dog muck on your shoe, it’s been hard to ignore the whiff of melodrama following the Claude developer over the past ten days. To begin with, the Claude Code launch almost killed an American computing institution by nuking IBM's stock by 13% – the worst single-day drop since 2000 – by threatening to make its COBOL consulting arm obsolete overnight.
Then came their very public, very ironic spat with Chinese AI firms, DeepSeek, Moonshot, and MiniMax, who they allege “distilled” aka training their models by querying the shit out of Claude through fake accounts (16 million times, no less) – naturally their Asian counterparts denied this, pointing towards the glaring hypocrisy behind, well, how developers train their AIs in the first place.
Then came the main event: a proper ding-dog with the US government. Defence Secretary Pete Hegseth summoned CEO Dario Amodei to the Pentagon for what officials called a "shit-or-get-off-the-pot meeting," threatening to blacklist Anthropic as a supply chain risk and invoke the Defence Production Act unless it drops its AI safeguards for use in military applications.
Whether Anthropic does indeed shit in the pot, or its pants, remains to be seen – but so far they’re not budging ahead of their Friday 27 deadline.
“Yeah, I met Hegseth. He kept referring to himself as ‘Lord of War’.” Source: AP
.FRONTIER NEWS .
#2 xAI strikes Grok deal with Pentagon
While Dario Amodei took a principled stand against Pentagon pressure, Elon Musk seemingly couldn’t wait to let it sweep over him. xAI has slipped in the back door via Grok and the Pentagon's classified systems. Grok is the next model to join Claude in that space, but the first to agree to the military's "all lawful use" standard, something Anthropic flatly refused. It’s a deal with Musk’s fingerprints all over it: no ethical quibbling, just sign on the dotted line. Whether Grok can actually replace Claude is unclear – defence officials admit it'd be a nightmare to switch – but the Pentagon now has its leverage, and Musk has another government contract.
..MODEL BEHAVIOUR ..
#3 Perplexity Computer: your new super agent
Perplexity has finally resurfaced with its biggest swing yet. CEO Aravind Srinivas unveiled ‘Perplexity Computer’ – a unified AI system he'd been quietly building, calling it "the next big thing." Essentially a super agent, all users need to do describe the end result you want, and it breaks the job into subtasks handled by specialised sub-agents, routing work across 19 models depending on the job – one for reasoning, another for coding, another for research. So less chatbot, more autonomous digital employee. Currently locked to premium subscribers, but the ambition is clear: Perplexity is coming for your entire workflow.
..MODEL NEWS..
#4 OpenClaw runs amok on Meta director
In this week's episode of You Can’t Make It Up, Meta’s Director of Safety & Alignment at its Superintelligence Labs – yes the person who makes sure the bots don't go rogue – Summer Yue, watched her OpenClaw agent do exactly that. Highlights include it speedrun deleting her inbox while ignoring every "STOP" command she fired from her phone. "I had to RUN to my Mac Mini like I was defusing a bomb," she posted on X. Yue narrowed down the rogue actor as ‘context window compaction’ – the agent ran out of working memory and quietly dropped her original "confirm before acting" instruction. Her verdict on herself: "Rookie mistake tbh. Turns out alignment researchers aren't immune to misalignment."
..MODEL BEHAVIOUR...
#5 AI will double in intelligence every 3.5 months
Quietly one of the most consequential findings in AI right now. Berkeley-based nonprofit METR devised a metric called the "task-completion time horizon," essentially how long a task takes a human that an AI can complete with 50% reliability and found it has been doubling every seven months for six straight years. If the pace holds to 2030, AI will complete in hours what currently takes a human a full working month. The more alarming footnote: the updated 2026 data suggests the doubling rate has actually accelerated to every 3.5 months meaning the graph everyone's citing may already be outdated.
..FRONTIER NEWS ..
#6 AI “doomsday” blog tanks US markets
If only this note moved markets a 7,000-word Substack has, we’d all be millionaires by this time next year. That Substack article in question: a "doomsday" thought experiment from Citrini Research, helped trigger an 800-point drop in the Dow. Written as a fictional dispatch dated June 2028, the article imagined unemployment above 10% and the S&P down 38% – caused by AI working too well. The mechanism: companies vibe-code their own software in-house, ditch Zendesk and Monday.com, and a wave of white-collar layoffs follows. DoorDash, Mastercard, Amex and Uber all fell 4% or more. Are the markets really that jittery a speculative blog post poses a systemic risk? Pfft.
Source: Cintrini Research
..FRONTIER NEWS .
#7 OpenAI cuts compute spend to $600 billion by 2030
Sam Altman has quietly downgraded his own hype. Months after touting $1.4 trillion in infrastructure commitments, OpenAI is now telling investors it plans to spend $600 billion on compute by 2030 – a 57% cut, reframed as fiscal discipline. To justify it, the company is projecting $280 billion in revenue by 2030, up from $13.1 billion last year. That's an 85% annual growth rate; heroic, if somewhat fantastical, to put it charitably. A $100 billion funding round is being finalized, with Nvidia reportedly in talks to invest $30 billion, valuing OpenAI at $730 billion pre-money. The numbers keep shifting, but the ambition, never. We all know Altman does it for the love of the game.
If you work in fintech or finance, you already have too many tabs open and not enough time.
Fintech Takes is the free newsletter senior leaders actually read. Each week, we break down the trends, deals, and regulatory moves shaping the industry — and explain why they matter — in plain English.
No filler, no PR spin, and no “insights” you already saw on LinkedIn eight times this week. Just clear analysis and the occasional bad joke to make it go down easier.
Get context you can actually use. Subscribe free and see what’s coming before everyone else.
TOOLS OF THE WEEK
The latest tool, model and LLM news, updates, tips and tricks – you name it, we’ll play with it til it breaks.
Apparently this is AI? Source: Google
Nano Banana 2: Google's viral image generator just got a serious upgrade. Nano Banana 2 merges the advanced capabilities of Nano Banana Pro with the speed of Gemini Flash – pro-quality at Flash prices. It handles up to five characters and 14 objects with consistent fidelity, supports resolutions up to 4K, and pulls real-time web knowledge for more accurate renders. Now the default across Search, Ads, Flow and Gemini – Google is cornering this space.
Replit: The OG online coding environment just levelled up hard. Replit's Agent 3 now runs autonomously for up to 200 minutes, builds and tests its own code in a self-healing loop, deploys full-stack web and mobile apps, and, most impressively, can build other AI agents and automations from a plain-English prompt. There’s a good reason it's quickly becoming the platform for non-technical founders wanting to ship software, and fast.
HyperGraph: One SKILL.md file was never going to cut it for complex AI agents. Built by the team at Hyperbrowser, HyperGraph solves the problem by turning any technical topic into a structured skill tree – linked, composable skill nodes that let agents navigate to exactly what they need rather than wading through one bloated context dump. Fully open source and built on the Hyperbrowser SDK.
Rork: Want to build a native mobile app by vibe-coding in plain English? Use Rork to spit-out a production-ready iOS and Android app in minutes. Backed by a16z and built on React Native, it's the mobile-first answer to the vibe-coding wave that's sweeping the industry. Think of it as Lovable, but for the App Store.
FINANCIALS
Covering significant funding stories, deals and partnerships, and investment opportunities globally.
David Silver, the British AI researcher behind AlphaGo, has left Google DeepMind to found Ineffable Intelligence in London – and is raising $1 billion in what would be Europe's largest-ever seed round, led by Sequoia Capital at a $4 billion valuation. Nvidia, Google, and Microsoft are all said to be circling the deal – a rare signal that world-class AI talent and capital can remain in Europe.
OpenAI is finalising a record-breaking funding round set to exceed $100 billion, with the company's total valuation potentially surpassing $850 billion. Nvidia is in discussions to invest up to $30 billion as part of the round, alongside SoftBank and Amazon, with the pre-money valuation pegged at $730 billion.
MatX, an AI chip startup founded by two Google semiconductor alumni, raised over $500 million in a round led by Jane Street and Leopold Aschenbrenner's Situational Awareness fund. Its chips promise roughly 10x the performance of current GPUs for training AI models, with first-generation chips set to fab at TSMC and sample shipments beginning in 2027.
Anthropic closed a $30 billion Series G – the second-largest venture round ever – valuing the company at $380 billion. Investors included Microsoft, Nvidia, Founders Fund, and ICONIQ, cementing what many are calling a two-horse race between Anthropic and OpenAI for frontier AI dominance.
Nuveen’s $13.5 billion all-cash acquisition of Schroders has sent shockwaves through the asset management industry, marking one of the largest consolidations in the sector this year. The deal highlights a massive move by US capital to capture European market share and diversify into UK-managed private and public assets.
Nvidia: Q4 revenue hit $68.1 billion, up 73% year-over-year, beating expectations across all three metrics, with Q1 2027 guidance set at $78 billion. The irony: the stock dropped 5% despite the blowout, as markets have become so used to Nvidia crushing estimates that it barely moves the needle anymore.
CoreWeave: Revenue grew 110% year-over-year in Q4, with full-year 2026 guidance of $12-13 billion and a revenue backlog swelling to $66.8 billion. Shares fell 5% after hours regardless, with Q1 guidance of $1.9-2 billion coming in below the $2.29 billion analyst consensus.
Alphabet: Beat on earnings ($2.82 vs $2.63 expected) and revenue ($113.83 billion), with Google Cloud at $17.66 billion, while guiding 2026 capex at $175-185bn – more than double its 2025 outlay.
Amazon: Delivered a mixed Q4 but sharply raised its spending outlook, with capex guidance exceeding $200 billion for 2026, largely directed at AI data centers.
Meta: Guided $115-135 billion in 2026 capex, continuing its aggressive AI infrastructure push, while quietly working on a new frontier model known internally as Avocado.
OpenAI: Generated $13.1 billion in revenue in 2025, beating its $10 billion target, while burning $8 billion – below its $9 billion spend forecast.
UNTIL NEXT WEEK, FOLKS.
A brighter note to end on.
Some of AI’s biggest players don’t believe the technology will, in fact, strip us of work, purpose or humanity, but instead offer us more.
Read (the excellently/somewhat suspiciously named) Ivan Mehta’s fantastic think-piece, over at TechCrunch.
“Calm down, it’s just a big robot.” Source: India Times
Plus this benevolent use of OpenClaw had the entire team in stiches...
We hope you enjoy the note. Stay locked for the next. 🤝
