The world's biggest AI model is now Chinese
Plus Meta's $10bn compute gamble, an AI that hunts bugs in your browser, and 3 startups worth watching. Four minutes, no jargon.
It’s Sunday, you’ve got 4 minutes and a coffee. Perfect.
This week: China unveiled the biggest open AI model on the planet, Meta is in talks to rent $10 billion of computing power to a rival, and OpenAI’s newest model learned to sniff out security holes in your web browser. One of these means Silicon Valley no longer runs the show. Let’s sort which.
🔷 The big ones
1. China just built the world’s biggest open AI model
Moonshot AI released “Kimi K3” - a 2.8-trillion-parameter model that’s open (free to download and run) and going toe-to-toe with OpenAI and Anthropic on public tests.
Why it matters to you: competition is good for your wallet. When it’s not just two American labs, tools get better and cheaper, faster. It’s also now a US-vs-China race - which shapes prices, rules and what’s available where.
2. Meta is renting $10bn of computing power to Anthropic
Meta, which has its own AI… is in talks to lease up to $10 billion of computing power to Anthropic, a direct rival, over two years.
Why it matters: the real AI battle isn’t chatbots, it’s compute - the raw horsepower to run them. It’s so scarce that even rivals sell each other shovels. This is why your AI tools’ prices wobble: someone’s paying gigantic bills behind the curtain.
3. OpenAI’s new model can find security holes in your browser
OpenAI’s latest - GPT-5.6 “Sol” - can now spot real security flaws in browsers like Chrome and Firefox (it can find them, if not fully exploit them yet).
Why it matters: double-edged. The same skill that helps defenders patch holes helps attackers find them. Expect a cybersecurity arms race - and take it as a sign these tools are getting genuinely capable, not just chatty.
📌 Also worth knowing
Alibaba open-sourced the software for its AI chips, aiming straight at Nvidia’s near-monopoly. The world wants off Nvidia’s leash.
Capital One released “VulnHunter,” a free AI tool that finds software flaws before hackers do… a bank open-sourcing security kit tells you where things are heading.
SpaceX’s “Colossus” supercomputer is in line for a multibillion-dollar Pentagon AI deal.
🧠 Jargon decoder
open-weight (”open” model): you can download it and run it yourself, free - owning the recipe instead of renting the meal. Opposite of the locked OpenAI/Anthropic ones you pay to access.
compute: the raw computing horsepower AI runs on. The oil of the AI economy - whoever controls it wins. Hence the $10bn deals.
✅ The honest verdict
Skip the chip and supercomputer stories… noise for you. The one that matters is Kimi K3. For two years, “the best AI” quietly meant “American AI.” For now, at least, that stopped being true… and more competition means better, cheaper tools landing in your lap. Quietly enormous.
👀 Ones to watch
🌊 Panthalassa - data centres that float in the ocean
What they do: build AI data centres that float in the open sea, powered and cooled by the ocean itself.
Raised: $140M Series B, led by Peter Thiel (with John Doerr, Marc Benioff’s TIME Ventures and Founders Fund). Portland-based, ~$210M raised so far, reportedly valued near $1B.
The goal: each node stands ~85m tall (mostly underwater), uses wave motion to generate its own power and seawater to cool the servers - no engines, no fresh water, data beamed back via Starlink.
Why it matters: AI’s real bottleneck now isn’t smarts, it’s power and water… data centres are draining both. Moving offshore removes both limits at once.
My take: this sounds mad until you realise the mainland physically can’t supply enough power for the AI boom. I’m watching whether “put it in the sea” goes from stunt to standard.
🇦🇺 GridCog - the software behind Australia’s energy switch
What they do: an Australian-founded platform that simulates renewable energy projects - solar, wind, battery.
Raised: A$12.5M (US$10M) Series A, led by ABB’s electrification ventures, with Axpo, DNV and Verbund alongside (plus existing backers CEFC and AlbionVC).
The goal: replace slow spreadsheet modelling - weeks of work - with simulations that run in hours, so developers can plan and green-light projects far faster.
Why it matters: the energy transition lives or dies on how quickly these projects get built, and the modelling bottleneck is real. Four major energy companies now own a piece of the tool their rivals plan with.
My take: unglamorous, but this is the classic picks-and-shovels play that quietly wins. When the incumbents you’d compete with are also your investors, you’re onto something.
📦 Bending Spoons - the company quietly buying the apps you use
What they do: an Italian tech company that buys beloved-but-stalled apps and runs them for profit. They own Evernote, WeTransfer, Vimeo, Eventbrite, StreamYard and Brightcove - and just added AOL.
Status: just IPO’d on Nasdaq - priced at $29, jumped ~40% on day one to a ~$25.7B valuation, raising ~$1.68B. Q1 revenue hit $601M with $27.4M net income - a turnaround from a $112M loss a year earlier.
The goal: buy-and-hold (not flip), pulling every app into one central Milan operation that shares engineering, infrastructure and payments. The founder says AI now writes the bulk of its code.
Why it matters: the AI boom made a whole generation of pre-AI software companies cheap. While everyone panics that AI kills old software, Bending Spoons is betting the opposite - loyal audiences + AI-run efficiency = profit. It’s reportedly eyeing 1,000+ more targets.
My take: the most contrarian bet in tech right now. One heads-up is that when they buy an app you love, expect layoffs and price rises. Watch which of your tools they grab next.
That’s the week. If this saved you from nodding along to a conversation about “the China AI thing,” forward it to the person who’ll need it next.
I’ll keep breaking down what actually matters in AI - no hype, no jargon - every Friday.
Follow along wherever you hang out: LinkedIn · YouTube · TikTok · Instagram
Forward this to someone who keeps saying they can’t keep up with AI
Want to work together? Just hit reply.
Cheers, Jagger