On December 2, 2025, Anthropic made its first-ever acquisition: Bun, a JavaScript runtime most people outside developer circles have never heard of. The price wasn't disclosed. The announcement was buried under headlines about Gemini 3 and GPT-5.2.

But this quiet deal may be the most strategically significant move any AI company has made this year.

To understand why, you need to follow the money—and the code.

Buried in Anthropic's acquisition announcement was a number that should have made headlines: Claude Code hit $1 billion in annual run-rate revenue in November 2025—just six months after becoming publicly available.

To put that in perspective: Slack took about four years to reach $1 billion ARR. Zoom took about three. Claude Code did it in six months.

This isn't just impressive growth. It's a signal that something fundamental has shifted in how enterprises are building software. Netflix, Spotify, KPMG, L'Oréal, and Salesforce are all now using Claude Code as a "critical tool" in their development workflows.

When AI writes code at this scale, a new bottleneck emerges: execution. And that's where Bun comes in.

Now through the end of the year is prime streaming time on Roku, with viewers spending 3.5 hours each day streaming content and shopping online. Roku Ads Manager simplifies campaign setup, lets you segment audiences, and provides real-time reporting. And, you can test creative variants and run shoppable ads to drive purchases directly on-screen.

Bonus: we’re gifting you $5K in ad credits when you spend your first $5K on Roku Ads Manager. Just sign up and use code GET5K. Terms apply.

Use code GET5K now

For the non-developers: A JavaScript runtime is the engine that executes JavaScript code on a server. Node.js has dominated this space since 2009. It's stable, battle-tested, and powers much of the modern web.

Bun is the upstart. Founded by Jarred Sumner in 2021, it's built from scratch in Zig (a low-level programming language) and uses Apple's JavaScriptCore engine instead of Google's V8. The result: Bun is dramatically faster.

How much faster?

• 2-4x faster HTTP throughput than Node.js

• CPU-intensive tasks complete in roughly half the time

• Package installation up to 30x faster than npm

• Near-instant startup times (critical for serverless)

Bun also bundles what Node.js spreads across multiple tools: runtime, package manager, bundler, and test runner—all in a single binary. This matters enormously when AI agents are generating, testing, and iterating on code thousands of times per day.

Here's the insight most coverage missed—and the reason this acquisition matters beyond Anthropic.

We're witnessing a category shift. Developer tools are no longer just tools for developers. They're becoming infrastructure for AI agents.

Think about what's changed. A year ago, a developer used a runtime to execute code they wrote. Today, AI agents are generating, testing, debugging, and deploying code autonomously—running thousands of iterations that humans never see. The runtime isn't serving a human anymore. It's serving an AI.

This creates entirely different requirements:

• Humans tolerate a 2-second startup. AI agents running 10,000 iterations don't.

• Humans read error messages. AI agents need machine-readable signals.

• Humans context-switch between tools. AI agents need unified interfaces.

• Humans optimize for flexibility. AI agents optimize for determinism.

Bun wasn't designed for AI agents—but it happens to be nearly perfect for them. Its speed, its all-in-one architecture, its predictable behavior. Anthropic saw this before anyone else.

The implication is uncomfortable for every devtools startup: If your tool isn't optimized for Agent Experience (AX), not just Developer Experience (DX), you may be building for a shrinking market.

Look at what Anthropic has assembled:

1. The model (Claude) — the intelligence that writes code

2. The coding assistant (Claude Code) — the interface developers use

3. The agent SDK — the framework for autonomous coding agents

4. The protocol (MCP) — how agents communicate with tools

5. The runtime (Bun) — the engine that executes the code

This is the AI equivalent of Apple owning both the iPhone hardware and iOS software. When you control the full stack, you can optimize in ways competitors can't.

Claude Code already ships as a Bun executable to millions of users. If Bun breaks, Claude Code breaks. By owning Bun, Anthropic eliminates a critical dependency and gains the ability to build features that standard runtimes can't offer—like native hooks for LLM context management or optimized execution paths for AI-generated code patterns.

Compare this to Microsoft's approach with GitHub Copilot: They've integrated Claude (yes, Anthropic's model) alongside GPT-5 because neither model alone was sufficient. They don't control the runtime, the protocol, or the agent framework. They're assembling parts; Anthropic is building a machine.

No analysis is complete without the risks. Here's what the skeptics are saying—and why some concerns are valid.

1. Bun's ecosystem is still immature.

Node.js has 15 years of battle-testing and the largest package ecosystem in existence. Bun aims for Node compatibility, but edge cases break things. Enterprise customers who need 100% reliability may hesitate.

2. The Zig dependency is a wildcard.

Bun is built on Zig, a language still in version 0.15 with a history of breaking changes. Zig's creator also has a "no LLM/AI" policy for contributions—potential friction now that Bun sits inside an AI company.

3. Open source communities can be fragile.

Anthropic promises Bun will remain MIT-licensed and community-driven. History suggests acquisitions often start this way—then priorities shift. Developer trust takes years to build and moments to lose.

4. Vertical integration can become a cage.

Apple's walled garden is powerful but limiting. If Anthropic optimizes Bun too heavily for Claude, it may alienate the broader developer community that made Bun attractive in the first place.

Context matters. Here's where Anthropic stands:

• January 2025: $1 billion ARR

• August 2025: $5 billion ARR

• October 2025: $7 billion ARR

• 2025 target: $9 billion ARR

• 2026 target: $20-26 billion ARR

Claude Code alone—at $1 billion ARR—represents roughly 14% of Anthropic's current revenue. More importantly, it's growing faster than any other product line. Anthropic has captured 32% market share in enterprise AI usage (vs. OpenAI's 25%), with particular strength in coding—42% share.

The Bun acquisition isn't about today's numbers. It's about ensuring Claude Code can scale to 10x this revenue without hitting infrastructure walls that competitors would exploit.

If you're a developer:

Bun just became a safer long-term bet. The MIT license stays, the team stays, and now there's guaranteed funding. But watch for Claude-specific optimizations that may not benefit general use cases. Hedge by maintaining Node.js compatibility.

If you're building devtools:

The uncomfortable question: Is your tool optimized for humans or AI agents? Tools that don't clearly integrate into a model platform's stack risk becoming commoditized utilities. Consider your acquisition path now, not later.

If you're an enterprise buyer:

Evaluate AI coding tools not just on model quality but on infrastructure ownership. Anthropic can optimize performance in ways GitHub Copilot can't because Microsoft doesn't own the execution layer. That gap will widen.

If you're watching the AI industry:

This won't be Anthropic's last acquisition. They've explicitly stated they'll pursue opportunities that "bolster technical excellence." Watch for moves into IDEs, deployment platforms, or MCP tooling. The playbook is clear: Own the stack, own the future.

Three years ago, buying a JavaScript runtime would have seemed like a distraction for an AI research lab. Today, it's a competitive necessity.

The deeper lesson isn't about Bun specifically. It's about how quickly the definition of "AI infrastructure" is expanding. Yesterday it was GPUs and training clusters. Today it's runtimes and developer tools. Tomorrow it might be IDEs, deployment pipelines, or the protocols that let agents talk to each other.

Anthropic understood something its competitors are still learning: When AI writes most of the code, the runtime becomes the new battleground.

They just claimed the high ground. The question is what they'll acquire next.