How Windsurf Got Carved Up in 72 Hours (And Built Google's Secret Weapon)

In May 2025, OpenAI agreed to pay $3 BILLION for an AI coding startup called Windsurf.

72 hours later, the deal was dead. The company was carved up between Google and Cognition. And Microsoft - who didn't even want to buy it - accidentally killed the whole thing.

Six months later, Google launched Antigravity. Their secret weapon in the AI coding wars.

It was built by the same team that created Windsurf.

This is the craziest M&A story in AI history. And nobody's connecting the dots.

Let's rewind.

2021. Two MIT friends - Varun Mohan and Douglas Chen - start a company called Exafunction. AI infrastructure. Enterprise stuff.

They rebrand to Codeium. Then in April 2024, they launch Windsurf - an AI-powered IDE that directly competes with Cursor.

Windsurf explodes. By early 2025:

$82 million in annual recurring revenue

350+ enterprise customers

Hundreds of thousands of daily active users

Full 'agentic' coding - AI that doesn't just suggest, it executes

They're neck-and-neck with Cursor. Both racing to become THE AI coding tool.

And then OpenAI comes knocking with a $3 billion offer.

May 2025. Bloomberg reports: OpenAI agrees to acquire Windsurf for $3 billion.

It makes perfect sense for OpenAI:

Instant talent - the best agentic coding engineers in the world

Working product with paying customers

Technology that integrates perfectly with ChatGPT

A weapon to compete with Microsoft's GitHub Copilot

Wait. Microsoft?

Here's where it gets complicated. OpenAI has a multi-billion dollar partnership with Microsoft. And buried in that partnership is a clause:

Microsoft has rights to any IP that OpenAI acquires.

Think about what that means. If OpenAI buys Windsurf... Microsoft gets access to the technology. The same Microsoft that owns GitHub Copilot - Windsurf's direct competitor.

OpenAI would be handing their competition a copy of what they just paid $3 billion for.

The negotiations drag on. Microsoft and OpenAI can't resolve the IP issue.

And then... the offer expires.

$3 billion. Gone.

Hours later - HOURS - Google announces they're hiring Windsurf's founding team.

But this isn't a traditional acquisition. It's a talent deal.

Google pays $2.4 billion total:

$1.2 billion to Windsurf's investors

$1.2 billion in compensation packages for ~40 employees

A 'substantial portion' going to the co-founders

Google gets: CEO Varun Mohan, Co-founder Douglas Chen, ~40 senior R&D engineers, Non-exclusive license to Windsurf's technology

Left behind: ~210 employees. The product. The customers. The brand.

Google cherry-picked the brains and left the body.

Friday, 5 PM. The Google deal goes public.

Cognition - the company behind Devin, the AI coding agent - makes a call.

By Monday morning, they've signed a deal.

Cognition acquires: All remaining Windsurf IP, The product and trademark, ~210 employees, The $82M ARR business, 350+ enterprise customers

Price tag: approximately $250 million.

Let that sink in. Google paid $2.4 billion for 40 people. Cognition paid $250 million for everything else.

But here's where Cognition got it right. They gave ALL Windsurf employees equity. Removed the standard vesting cliff. Credited previous work.

The employees Google left behind? They got taken care of.

July to November 2025. Four months of silence from Google.

Behind the scenes, the Windsurf team is integrating into DeepMind. Working on Gemini. Building something.

November 18, 2025. Google announces Gemini 3.

And alongside it... Antigravity.

A brand new AI IDE. Agent-first architecture. Direct browser integration. Multi-agent workflows.

Sound familiar?

It's Windsurf's DNA - reimagined with Google's infrastructure.

The same team. The same philosophy. But now powered by Gemini 3.

Let me show you what they built.

Agent-First Architecture: Antigravity doesn't treat AI as an assistant. It treats AI as a partner.

You give it a task. It plans. It executes. It validates. Autonomously.

Manager View: Run 5 agents in parallel. Each working on different parts of your codebase. All coordinated.

Browser Integration: The agent can control a browser. Test your frontend. Debug in real-time. End-to-end.

The Benchmarks: Antigravity with Gemini 3: 76.2% on SWE-bench, Gemini 3 Flash alone: 78% on SWE-bench Verified, Cursor with Claude: ~77%

They're all essentially tied. But Google has one advantage nobody else has:

Gemini 3 Flash: $0.15 per million input tokens

Compare that to Claude Sonnet at $3 per million

Google can subsidize this forever. They're playing a different game.

Step back and look at what just happened:

Windsurf - one company - is now THREE companies:

Google Antigravity: The founding team + Gemini = Google's AI IDE

Cognition/Windsurf: The product + remaining team = Devin integration

The IP: Licensed non-exclusively to Google, owned by Cognition

And Microsoft? They didn't get anything. Their IP clause - designed to protect them - pushed the talent to their biggest competitor.

Winners: Google: Got the best agentic coding team on the planet, Cognition: Got a profitable business for $250M, Windsurf employees: Got equity from Cognition, Investors: Got $1.2B from Google

Losers: OpenAI: Lost the acquisition, Microsoft: Accidentally helped Google

This is what happens when corporate deals meet startup speed. 72 hours changed the entire AI coding landscape.

So what does this mean if you're a developer?

You now have THREE major players in AI coding:

Cursor: Still the leader. Claude integration. Best community.

Antigravity: Free preview. Google infrastructure. Gemini 3.

Windsurf/Devin: Enterprise focused. Cognition backing.

My take? Try all three.

Cursor is still the best overall experience. But Antigravity in free preview? You'd be crazy not to test it.

Especially if Google starts bundling it with Workspace or Cloud.

The Windsurf team didn't disappear. They just changed jerseys. And now they're playing for a team with infinite resources.

I've put together a complete comparison of Cursor vs Windsurf vs Antigravity at End of Coding.

Real benchmarks. Feature breakdowns. Who should use what.

Link in description.

The craziest part of this story? It all happened in 72 hours.

$3 billion offer. Microsoft kills it. Google swoops. Cognition scrambles.

And six months later, the team that almost joined OpenAI is now Google's secret weapon.

That's the speed of AI in 2026.