Stripe’s OpenRouter math just got better, but its Anthropic problem just got worse 😬🤖; Revolut opens private markets to 75M users 💸🔓; Kimi K3 just made frontier AI something you can own 🤖🔓
You're missing out big time... Weekly Recap 🔁
👋 Hey, Linas here! Welcome back to a 🔓 weekly free edition 🔓 of my daily newsletter. Each day, I focus on 3 stories that are making a difference in the financial technology space. Coupled with things worth watching & most important money movements, it’s the only newsletter you need for all things when Finance meets Tech.
If you’re not a subscriber, here’s what you missed this week:
How to Build an AI Monopoly: The New Rules of Startup Strategy 🦄 [Anyone can build an AI product now. Almost no one can build an AI company. Here is how to escape competition, build moats that compound, and capture the value you create]
The Ultimate Guide to Kimi K3 🤖 [Moonshot’s 2.8T open model is now #1 in the world for frontend coding and AI agents at a third of Claude’s price. What it changes for founders, builders, and investors, and exactly how to operate it]
Turn Claude Opus 5 Into a Financial Analyst That Never Sleeps 📊 [Claude Opus 5 just took #1 on the leading independent finance benchmark. Here’s the framework that turns it into an AI analyst that works for you 24/7]
DeepSeek’s leaked investor call: inside the AI Playbook that erased $589 billion of Nvidia’s value 📉 [Liang Wenfeng walks investors through DeepSeek’s 6x pricing rule, its 20,000-GPU fleet, the Huawei plan to break CUDA, and an AGI roadmap with dates attached]
How to Build an Agentic OS with Claude Fable 5 🤖 [The complete guide to building an autonomous AI agent operating system with Claude Fable 5 — architecture, working code, cost control, security, and the economics of agentic labor]
Block cut 40% of its workforce and then open-sourced Buzz, the AI Agent Workspace that makes it work 🤖🐝 [how it actually works, why calling it a Slack killer misses the ultimate goal, what Block's internal numbers say about running a company on AI agents + the Full AI OS to Run a Startup with Claude & How to Build and Agentic OS with Claude Fable 5 inside]
Stripe is paying $10 billion for OpenRouter’s 5.5% fee, not its LLM routing 🤖💸 [what this M&A is all about, why it’s very strategic for Stripe, & how it differs from Ramp’s LLM routing approach + bonus deep dive into the AI Monopoly Playbook, and How to Build an Agentic OS with Claude Fable 5 inside]
Revolut is officially worth $115 billion on a loan book it hasn’t built yet 🤯🏦 [why it matters, where’s the biggest risk & opportunity for Revolut + bonus deep dives into Revolut, Monzo, and Starling’s latest financials, and Revolut’s Foundation AI Model inside]
PayPal’s Q2 2026 earnings: Stripe’s $53 billion takeover bid just erased the bargain 🤝💸 [breaking down the most important facts & figures from PayPal's Q2 2026 to see whether PayPal is still worth your time & money + bonus deep dive into Stripe, the $159B giant that now wants to buy it, and how to Turn Claude Opus 5 Into a Financial Analyst That Never Sleeps inside]
Robinhood’s Q2 2026: a 44% EPS beat, a $106M gain that’s already gone & why HOOD stock fell anyway 🏹📉 [breaking down the most important facts & figures from Robinhood's Q2 2026, including the $106M gain that vanished before the earnings call even started, to see whether HOOD 0.00%↑ is still worth your time & money + bonus guide on how to Turn Claude Opus 5 Into a Financial Analyst That Never Sleeps inside]
GLM-5.2: The ChatGPT Moment for Local AI 🤖 [Why the first open-weight model that rivals Claude Opus 4.8 and GPT-5.5 — and runs on a single Mac — changes the game for every founder, builder, and investor]
Loop Engineering: How to Design AI Loops That Build, Ship, and Improve While You Sleep 🔁 [From a three-line bash script to multi-day Claude Fable 5 autonomy — everything founders, builders, and investors need to stop prompting agents and start designing the systems that prompt them]
Anthropic Just Told AI Founders Exactly What to Build in 2026 🦄 [1 million conversations. 9 consumer AI domains. A full founder playbook - plus where Anthropic’s own products will and won’t compete]
As for today, here are the 3 incredible FinTech stories that are transforming the world of financial technology as we know it. This was yet another intense week in the financial technology space, so make sure to check all the above stories.
Stripe’s OpenRouter math just got better, but its Anthropic problem just got worse 😬🤖
Following the money 💸 The multiple everyone quoted this week when discussing the Stripe/PayPal deal, including me, was off by a factor of three.
Let’s unpack this.
More on this 👉 OpenRouter is running at roughly $140 million in annualized revenue, about $12 million a month, nearly triple its April pace, per The Information’s reporting on the startup’s financials. At close to $10 billion, Stripe would be paying around 70 times revenue, not the 200x that April’s $50 million figure implied. This week’s piece used the stale number.
The rest of the P&L is also better than the price suggests. Cost to serve the routing product runs about $40 million a year, 28.5% of revenue, leaving roughly $100 million of gross profit at a 70% margin. Fewer than 100 employees, and likely profitable before stock compensation. Volume hit 250 trillion tokens a month, up from 50 trillion in February. Not too shabby!
Now let’s set that against the comp that matters. SpaceX is paying $60 billion in stock for Cursor at about 22 times forward revenue, on gross margins that were only just turning positive. OpenRouter costs three times the revenue multiple and earns software-grade margins on it. Measured against gross profit rather than revenue, OpenRouter is the cheaper asset. Hence, the price argument is basically over.
ICYMI:
Zoom out 🔎 What isn’t over is supply. OpenRouter aggregates more than 400 models but routes heavily to closed ones, Anthropic above all. Developers already report friction reaching Anthropic’s models through Cursor and OpenClaw. OpenRouter’s neutrality is a permission granted by the labs it disintermediates, not an asset it owns, and no acquisition changes who controls the keys.
But for Stripe, the conflict runs deeper. We must remember that it also processes payments for Anthropic and OpenAI, and would now own a business whose pitch is moving spend away from both, toward cheaper models including open weights from Chinese labs like Moonshoot AI (Kimi) or DeepSeek. And those two processing relationships are definitely worth more than OpenRouter.
THE TAKEAWAY ✈️
What’s next? 🤔 First and foremost, watch Anthropic’s terms for aggregators, because that one decision sets the floor under this valuation. Of course, the dilution matters too: at February’s $159 billion mark, an all-stock deal costs Stripe shareholders about 6%. And note that OpenRouter now earns under five cents per million tokens routed, which is what happens when volume shifts to cheap open models faster than fees can follow. Growth that fast at a take rate that thin is a bet on volume compounding forever. But maybe that’s the price you should pay if you want to become the Everything Router…
ICYMI:
Revolut opens private markets to 75 million users 💸🔓
The news 🗞️ The same fund managers restricting redemptions are now opening a new retail pipeline, and fintech giant Revolut is betting its IPO narrative on being the front door.
Let’ take a closer look at this.
More on this 👉 On July 27, Revolut started letting EEA customers in France and Spain buy into private equity, credit, and infrastructure funds from Apollo, Ares, Hamilton Lane, and Partners Group, with a minimum investment of €1 through Revolut-managed feeder vehicles. The rollout is expanding to Germany, Italy, Portugal, and other EEA markets through the end of July.
Three days later, Revolut announced a separate deal bundling OpenAI’s ChatGPT Go into every paid subscription tier. Read together, the two moves tell the same story. Revolut is dressing the shop window ahead of a $150 to $200 billion IPO.
The private markets launch runs on ELTIF 2.0, the EU’s overhauled fund wrapper that went live in January 2024 and, for the first time, allows evergreen open-ended structures with retail passporting across the bloc. Without that regulatory plumbing, distributing institutional-grade PE and credit at €1 minimums through a banking app would not be legally possible.
→ Revolut charges no platform or transaction fees on top of the funds’ own costs (roughly 2.0 to 2.8% annually)
→ Instead, it collects retrocessions from managers, building a revenue line that scales with assets under management rather than trading volume.
For a company whose wealth-segment revenue growth decelerated to 31% in 2025 after a faster 2024, that’s a welcome addition.
Zoom out 🔎 The timing, though, is hard to ignore. Bloomberg’s own reporting on the launch flags the backdrop directly. Apollo and Ares have both restricted redemptions in 2026, and Partners Group gated an $8.6 billion evergreen buyout fund in June after redemption requests hit 9.8% of NAV against a 5% cap. Revolut says it built a nine-person team to vet each manager’s ability to handle redemption pressure, and that any manager who couldn’t demonstrate that history was excluded. Whether a 9-person team is adequate due diligence for products reaching 75 million potential customers is a question Revolut will have to keep answering.
Of course, the strategic logic is sound here even if the risk picture is murkier. Revolut is following Trade Republic (which tied up with Apollo and EQT last year) and Robinhood (which launched a closed-end fund holding private company stakes, including in Revolut itself).
Thus, it’s clear that private markets access is becoming table stakes for any neobank with ambitions beyond payments. For Apollo, Ares, and their peers, fintechs with tens of millions of users are a new capital source now that institutional LP fundraising has gotten more competitive.
THE TAKEAWAY ✈️
What’s next? 🤔 Looking ahead, what matters next is whether Revolut’s ELTIF funds honor their liquidity windows through the current redemption-stress cycle without gating the retail investors who just walked in. If they do, Revolut will have a proof point for its “sophisticated wealth platform” positioning when it eventually roadshows for that IPO. If any of its four partner funds locks up capital from retail customers who expected to access it, the resulting consumer-protection headlines would arrive at exactly the wrong moment. Ultimately, Revolut is now asking retail users to practice the same patience its own cap table has exercised for over a decade. The difference is that Revolut’s early investors chose illiquidity knowingly. Whether every retail customer putting in €1 understands the same bargain is far less certain.
ICYMI: Revolut is officially worth $115 billion on a loan book it hasn’t built yet 🤯🏦 [why it matters, where’s the biggest risk & opportunity for Revolut + bonus deep dives into Revolut, Monzo, and Starling’s latest financials, and Revolut’s Foundation AI Model inside]
Kimi K3 just made frontier AI something you can own 🤖🔓
The BIG News 🔥 The gap between the intelligence you rent from closed labs and the intelligence you can run yourself just narrowed to almost nothing.
On July 27, Moonshot AI, a 300-person Beijing lab last valued at $20B, released the full weights of Kimi K3. The model has 2.8 trillion parameters with 104 billion active per token, a million-token context window, native vision, and performance that independent indexes place just behind Claude Fable 5 and GPT-5.6 Sol. It is the first open-weight model in the 3T-parameter class, and within days, groups were running it on clusters of consumer RTX 5090s over plain Ethernet without HBM 😳
The parameter count earned the headlines, but for anyone building on open models, the architecture underneath and what Moonshot chose to release alongside the weights matter way more.
More on this 👉 Kimi K3 activates only 104B of its 2.8T parameters per token through an 896-expert mixture-of-experts design, with a custom attention mechanism that handles million-token sequences without the KV-cache blowup that normally makes long context prohibitively expensive.
Quantization-aware training bakes MXFP4 precision into the weights from the start instead of bolting it on afterward, and the combined efficiency gain over Moonshot’s prior generation is 2.5x. In other words, you get more intelligence per GPU-hour from a smaller memory footprint.
Moonshot also open-sourced the infrastructure to run and extend the model, including high-performance attention kernels, expert-parallel communication libraries, and AgentEnv, a distributed sandbox for training long-horizon agents with snapshot, fork, and parallel rollouts. vLLM and SGLang had day-zero support, so this is production-grade tooling from launch.
Zoom out 🔎 The license tells you the strategy.
Research, fine-tuning, internal deployment, and commercial embedding are all free.
Gates kick in only if you resell model access as a service above $20M annual revenue, or if your product tops 100M monthly active users.
In other words, Moonshot is betting that distribution beats secrecy, planning to monetize through its own API and mega-scale licensees.
THE TAKEAWAY ✈️
What’s next? 🤔 Founders locked into closed-model APIs now have a self-hostable alternative they can fine-tune and audit. The immediate beneficiaries are therefore vertical AI companies in compliance-heavy sectors (legal, healthcare, defense) and anyone building long-running coding or research agents where multi-hour token streams at closed-model pricing destroy unit economics. A startup building a DevOps agent that runs four-hour debugging sessions can now do that on its own hardware at inference cost rather than burning through an API budget.
Looking ahead, the longer-term question is about pricing power. When near-frontier intelligence is available at bare inference cost, the value of a proprietary API hence shifts from “access to capability” to “convenience and uptime,” which is a thinner margin to defend. Whether Anthropic and OpenAI respond by opening more weights, cutting prices, or sprinting to widen the gap at the top will shape the model economy for the next two years. Fascinating times for AI 🤖
ICYMI:
🧠 What else I’m watching
Anthropic Joins FCA Sandbox 🤖 Anthropic is supporting the UK Financial Conduct Authority’s Supercharged Sandbox, providing firms with access to Claude Code and Claude Cowork to test AI use cases like safer agent-led payments, fraud detection, and financial inclusion. Among 21 selected firms, the initiative also includes a new Agentic Academy to accelerate responsible AI innovation in finance. It will be interesting to see whether regulated AI sandboxes unlock safer financial innovation. ICYMI: Loop Engineering: How to Design AI Loops That Build, Ship, and Improve While You Sleep 🔁 [From a three-line bash script to multi-day Claude Fable 5 autonomy — everything founders, builders, and investors need to stop prompting agents and start designing the systems that prompt them]
Circle Acquires IBM Patents 💸 Circle has acquired nearly 1,000 blockchain-related patents from IBM, covering foundational technology, banking, and supply chain verification, positioning it as the US leader in blockchain patent holdings. The deal supports Circle’s mission to build an internet-native financial system, including USDC and its growing suite of onchain products, while exploring further commercial collaborations with IBM. It will be interesting to see whether this solidifies Circle’s lead in onchain finance. ICYMI: Visa’s Stablecoin Platform launches with Open USD, sending Circle down 6% 😳🪙 [why Visa followed the same pattern Anthropic set with Claude Managed Agents, & what’s the bigger play here + bonus deep dives into Coinbase, Circle, and the Ultimate List of Stables Resources inside]
Visa Cuts 7% Jobs 💼 Visa is cutting 7% of its workforce, or around 2,600 jobs, primarily in technology and product roles, as it reconfigures operations for the AI era. The move, driven by AI’s ability to streamline repetitive work and accelerate development, will reinvest resources into consumer payments, stablecoin, cross-border, and B2B solutions, following broader industry trends. Yet another great reminder to focus on productizing yourself 😉 ICYMI:
💸 Following the Money
Venezuelan Buy Now, Pay Later (BNPL) fintech Cashea has raised $100M as it looks to branch out beyond its core installment product into broader payments and savings niches.
Corgi, an AI-powered full-stack insurance platform, has reportedly hit a $4B valuation after raising money from investors for the fourth time this year.
Italian wealth management technology provider Objectway is to buy capital markets IT firm Slib from BNP Paribas and Natixis.
👋 That’s it for today! Thank you for reading, and have a relaxing Sunday! And if you enjoyed this newsletter, invite your friends and colleagues to sign up:
Great as always - thx
Super interesting - thanks Linas for putting such a savage issue this week!