# The Week Everything Changed: $30B for Anthropic, Mistral's 20x Surge, The Death of Writing Code & Why a $5B Exit is "Failing" ⚡

👋 Hi, I'm Kenneth - Welcome to The Tech Venture Navigator.

This is your weekly playbook for staying ahead in tech, AI, and venture. We explore how top investors think, how real founders build, and the strategies shaping tomorrow's companies — so you remain essential in a landscape being rewritten by AI.

I've been doing this for years. Writing these newsletters. In the trenches with founders. Raising capital. Building and scaling companies the hard way. And I need to be straight with you — I have never seen a week like this one.

Yesterday, Anthropic closed $30 billion at a $380 billion valuation. Their AI coding tool now writes 4% of every piece of code published on GitHub — and that number is heading to 20% by December. Inside OpenAI, engineers are running fleets of 10 to 20 AI agents simultaneously — and they've started calling themselves "sorcerers." Mistral AI just reported 20x revenue growth in a single year — from $20M to north of $400M — and is now Europe's only frontier model provider at a €12 billion valuation. Meanwhile, Brex — once the darling of fintech at a $12.3 billion valuation — just sold for $5 billion, and half of Silicon Valley is calling it a failure. Neuralink has now implanted brain chips in 21 human beings who control computers with their thoughts for 50 hours a week. And a legal AI company that didn't exist four months ago is already in talks for a $6 billion valuation.

The ground is moving under every company, every fund, and every career in technology.

Let's get into it.

Brought to you by Scripe - AI-powered LinkedIn content workspace.

Yesterday, Anthropic closed a $30 billion Series G — the second-largest private tech round in history. The valuation: $380 billion, more than doubled from $183 billion just five months ago. Total capital raised: $69.1 billion. That number puts Anthropic ahead of OpenAI in total venture funding for the first time.

But the valuation isn't the real story. The real story is that the numbers underneath it are no longer speculative.

Anthropic's run-rate revenue has hit $14 billion. That's 10x annual growth for three consecutive years. Their coding agent, Claude Code, alone generates $2.5 billion in annualised revenue — more than doubled since January. Over 500 enterprise customers now spend more than $1 million per year on Claude. Eight of the ten largest Fortune companies are clients. And here's the number that should stop every software founder in their tracks: Claude Code currently authors 4% of all public commits on GitHub. Analysts project that will hit 20% or more by the end of 2026.

Node.js creator Ryan Dahl put it bluntly: the era of humans writing code syntax directly is over.

The investor syndicate tells its own story. GIC and Coatue led. Co-leads: D. E. Shaw Ventures, Peter Thiel's Founders Fund, Dragoneer, ICONIQ, and Abu Dhabi's MGX. The supporting cast includes Accel, Sequoia, General Catalyst, BlackRock, Blackstone, Temasek, Goldman Sachs, JPMorgan Chase, Fidelity, Jane Street, and the Qatar Investment Authority. Thirty-six investors on a single cap table. That's not a funding round — it's a sovereign consensus that this technology is permanent infrastructure.

Meanwhile, OpenAI is reportedly pursuing $100 billion in new capital at an $830 billion valuation. If both numbers hold, the two leading foundation model companies will represent over $1.2 trillion in private market value. We have never seen anything like this in the history of venture capital. Not in cloud. Not in mobile. Not in the internet itself.

Here's why - and here's why they're wrong.

Brex raised approximately $1.7 billion across 14 funding rounds. They also carried more than half a billion in debt. At their peak in January 2022, they were valued at $12.3 billion. Selling for $5.15 billion is a 58% decline from the peak. In the venture math that dominates this industry's psychology, that's a down round. A loss narrative.

But the math tells two completely different stories depending on when you showed up.

For early investors — Ribbit Capital, Kleiner Perkins, Peter Thiel, Max Levchin — who entered when Brex was worth tens or hundreds of millions, this was a phenomenal outcome. Even at $5 billion, their returns were multiples on multiples. They were protected by the vast distance between their entry price and the exit price.

For late-stage investors who participated at the $12.3 billion peak in 2021? They were buying at the top of a mountain during the most euphoric funding market in history. For them to make money, Brex needed to keep growing forever. It didn't — and their positions went underwater. Employees who joined late faced the same trap: stock options priced at peak valuation that became worthless when the exit came in at less than half.

Here's what the failure narrative misses entirely: Capital One made one of the smartest strategic acquisitions in banking history. They bought a functioning machine — 25,000 business customers including DoorDash, Zoom, and Anthropic itself — a profitable product, AI-native automation, and a full corporate card and spend management stack. Building that internally would have taken years and billions in development costs. They got all of it at a 58% discount to peak, in a deal their CEO described as building "a payments company at the frontier of the technology revolution."

The real lesson from Brex isn't about failure. It's about a structural truth that every founder raising capital right now needs to tattoo on the inside of their eyelids: a high valuation is not a prize. It's a debt.

Every dollar of inflated valuation creates an expectation that must be paid back with growth. A $20M pre-seed means your Series A needs to justify $60M. A $60M Series A means your Series B needs to hit $180M. And so on, compounding, until the gap between your valuation and your actual business becomes a trap that removes every option except growing into an impossible number.

Brex didn't fail. Brex built a real business, survived a brutal market correction, became profitable, and exited to a strategic acquirer for $5.15 billion. The people who failed were the ones who set a price that the real world could never justify.

Sources: Capital One Official Press Release + Funder Intel

This week, one of the most revealing conversations about the future of engineering dropped. Sherwin Wu, who leads engineering for OpenAI's API platform, pulled back the curtain on what working inside the world's most advanced AI company actually looks like right now. And it should make every technical founder reassess their assumptions.

95% of OpenAI's engineers now use their AI coding tool Codex. Not as a helper on the side. As their primary method of building software. Individual engineers routinely manage 10 to 20 parallel AI agents at once — orchestrating autonomous streams of work simultaneously, reviewing outputs, and directing the next steps. Wu's analogy: engineers are becoming "sorcerers" — directing powerful forces rather than manually casting every spell.

Three things from this conversation that every builder needs to hear:

1. Code review times went from 15 minutes to 3 minutes. That's not a productivity improvement. That's a structural change in how software teams operate, how sprints are planned, and how fast products ship. If the best engineers in the world have already made this transition, it's a matter of months before every competitive engineering org follows.

2. The productivity gap between AI power users and everyone else is widening fast. Engineers who've learned to orchestrate agent fleets are producing at 5-10x the output of those who haven't. This isn't a theoretical concern about the future — it's a split that's happening now, inside real teams, creating a new class divide in engineering.

3. "Models will eat your scaffolding for breakfast." Every workaround, wrapper, and prompt engineering hack you're building today will be obsolete the moment the next model ships. Don't over-engineer for current limitations. Build for where models are heading — not where they are.

Wu framed the next 12 to 24 months as a rare, asymmetric window. Engineers who learn to work with agent fleets now will leap ahead. After the transformation completes, the advantage disappears — it becomes table stakes for everyone.

Source: Sherwin Wu Linkedin + X

While the rest of the tech world argues about software valuations and funding rounds, something extraordinary is happening at the boundary of technology and biology that deserves your attention.

Neuralink has now implanted brain-computer interfaces in 21 human participants across worldwide clinical trials. These individuals use the implant for an average of 50 hours per week — controlling computers, playing chess, browsing the internet, writing messages, and handling daily digital tasks entirely through thought.

The first participant, Noland Arbaugh, received the N1 chip in January 2024. Since then, he's used it daily to regain complete digital independence — navigating the web, learning online, communicating publicly, and playing strategy games — all by thinking.

Here's how the technology works in plain language: 1,024 electrodes on 64 flexible threads are implanted into the motor cortex. Each electrode detects electrical signals from nearby neurons — signals that are roughly 15,000 times weaker than a standard AA battery. The chip amplifies these impossibly faint signals, identifies meaningful patterns, packages the data, and transmits it wirelessly via Bluetooth. After a 20-minute calibration, the system decodes thought patterns into cursor movements dozens of times per second.

The framing that stopped me cold: the human brain receives data at roughly 1 billion bits per second — but it can only express about 10 bits per second through movement, speech, and action. You're connected to fibre-optic broadband. You can only respond through a 1990s dial-up modem.

The targets ahead: 10,000 implants per year by 2030 at a cost of $40,000-$50,000 per patient. Future capabilities include restoring tactile sensation, enabling natural speech for people who've lost it, and a product called Blindsight that would restore vision by directly stimulating the visual cortex with camera input.

Source: Neuralink official Site + Pew Research (brain chip attitudes)

A lot of people wrote off Mistral last year. The French AI lab was seen as a scrappy underdog burning through capital in a fight it couldn't win against OpenAI and Anthropic's war chests. Those people are very quiet right now.

Mistral's revenue has gone from $20 million to north of $400 million in twelve months. A 20x increase. They're on track to hit $1 billion in run-rate revenue this year. The company has raised €1.7 billion at a €12 billion valuation and just announced a €1.2 billion investment in new AI data centres in Sweden.

CEO Arthur Mensch framed the moment in terms that should resonate with every European founder and investor: "Europe has realised that its dependency on US digital services was excessive and at breaking point today."

That's not posturing. It's the demand signal behind Mistral's numbers. European governments, enterprises, and regulated industries are actively seeking alternatives to US-based AI providers. Data sovereignty, regulatory compliance (GDPR, the EU AI Act), and geopolitical risk are converting from abstract policy concerns into real procurement budgets. Mistral is the only European frontier LLM provider positioned to capture that demand, and the revenue proves the market is buying.

Here's what makes Mistral's position structurally interesting: they've consistently argued that fine-tuned smaller models outperform larger general-purpose models on enterprise benchmarks. In a market where cost-per-token matters as much as raw intelligence, this thesis is gaining traction. Their Gemini 2.5 Flash competitor comes in at a 63% cost advantage over comparable models, and enterprises are noticing.

Source: Financial Times

Pair that with OpenAI's internal data: 95% of their engineers use Codex daily. Code review times cut from 15 minutes to 3. Engineers managing fleets of 10-20 parallel agents.

For founders: your moat is not your code anymore. It hasn't been for a while, but now the numbers make it undeniable. The companies that win from here are the ones with proprietary data, deep distribution, and an understanding of specific workflows that no general-purpose model can replicate. Code is becoming a commodity input. What you know about your customer's problem, that's the defensible asset.
Source: TechStartups: Feb 5, 2026

Source: Company announcements, TechCrunch, Bloomberg

First: legal AI is no longer a niche — it's a category. Harvey at $8B+. Legora tripling to $6B in four months. Lawhive raising $60M. Checkbox at $100M valuation. These are venture-scale outcomes in a vertical that didn't exist three years ago.

Second: Monaco is the stealth deal of the week. A former Founders Fund VC left venture after 18 months to build an AI-native CRM with the former CPO of Apollo/Qualtrics and the former SVP of Engineering at Clari. Backed by the Collison brothers and Y Combinator's Garry Tan. If you're watching the "replace Salesforce" thesis play out in real time, this is the team betting their careers on it.

AI is eating engineering | Sherwin Wu (OpenAI) on the next 12-24 months of AI

We sit at the centre of 350k+ founders, operators, and LPs. Our goal: eliminate the "random walk" of fundraising.

Every day, 4.5 million readers turn to 1440 for their factual news fix. We sift through 100+ sources to bring you a complete summary of politics, global events, business, and culture — all in a brief 5-minute email. No spin. No slant. Just clarity.

Join for free today!

This was a week that will be referenced for years. $30 billion for one company. Code writing itself. Brain chips in 21 humans. A $5 billion exit called a failure. A French startup 20x-ing revenue and proving European AI sovereignty is real. The pace has changed permanently and it's not slowing down.
Thanks for reading this edition. If you got value from it, share it with a founder, teammate, or friend who’d use it too it keeps this community sharp and growing.

Join the crew below and follow me on LinkedIn to never miss an update.

If you’d like tailored help you can book time with me on hubble here.

— Kenneth Kelly
Founder @Tech Venture Navigator

^{What do you think about Tech Venture Navigator?}
💡 ^{Love it |} 🔥 ^{Useful | 👍 Solid | 🤷‍♂️ Okay |} 🚫 ^{Skip it or Leave feedback here.}

^{Got ideas, topics, or frameworks you’d love to see covered? Just hit reply or drop me a message, we read every note.}