$19 billion. OpenAI sits at $25 billion, so the gap is now $6 billion. A year ago, the difference was more than double. Anthropic has closed to 76% of OpenAI's revenue.
And alongside that number came another headline: Anthropic has signed a deal with Google and Broadcom for multiple gigawatts of TPU capacity, coming online in 2027.
This is no longer a race with a clear winner on the horizon. It's a genuine competition.
TL;DR
- Annualized revenue: $19 billion (76% of OpenAI's $25B)
- Google TPU deal: Broadcom partnership, gigawatt-scale TPU capacity, operational in 2027
- MCP: 97 million installs — becoming the industry standard for AI agent tooling
- Claude Sonnet 4.6: Top-ranked in real-world expert task evals and coding agent workflows
- Security incident: Source code leak on March 26
- AI power ranking: Anthropic is solidly #2 and closing fast
Anthropic's Revenue Milestone
In 2023, Anthropic was the "serious but not quite OpenAI" player. Claude 2 earned solid reviews, but ChatGPT's dominance seemed untouchable. Today, the landscape is different.
Revenue Comparison
| Company | Annualized Revenue | Valuation | Notes |
|---|---|---|---|
| OpenAI | $25B | $852B | $2B/month, IPO in prep |
| Anthropic | $19B | Undisclosed | Gap narrowing fast |
Anthropic's growth rate is outpacing OpenAI's. While OpenAI navigates IPO preparations and the organizational pressures that come with it, Anthropic has been heads-down on product and infrastructure. That focus is showing up in the numbers.
The Google TPU Deal: Why This Is a Game Changer
This isn't a routine cloud contract. Anthropic securing gigawatt-scale TPU capacity with Google and Broadcom is a structural move, not an incremental one.
Here's why it matters:
1. Breaking NVIDIA Dependency
The entire AI industry is bottlenecked on NVIDIA GPUs. Google's TPUs represent a credible alternative that can absorb serious workloads. Meta is pursuing a similar strategy with its MTIA chips to reduce NVIDIA reliance. Anthropic has now secured its own alternative lane — a strategic hedge that gives it pricing leverage and supply independence.
2. Locking In Compute Capacity Before Demand Peaks
As Morgan Stanley has predicted, 2026-2027 will see AI compute demand surge. Pre-securing gigawatt-scale capacity isn't just about cost reduction — it's about raising the physical ceiling on growth. You can't serve customers you don't have compute for.
3. Deepening the Google Strategic Relationship
Google is a major Anthropic investor. This TPU deal elevates that relationship from "investor-investee" to "infrastructure partner." Expect tighter Google Cloud integration, better tooling, and potentially preferential terms across the board.
AI Infrastructure Strategy Comparison
| Company | Infrastructure Strategy | Notes |
|---|---|---|
| OpenAI | Microsoft Azure + proprietary data centers | High Azure dependency |
| Anthropic | Google Cloud + TPU gigawatt deal | 2027 operational |
| Meta | MTIA self-developed chips + NVIDIA | Actively reducing NVIDIA share |
| TPUs for self + licensing to third parties | Supplying Anthropic |
OpenAI vs. Anthropic: The Scoreboard Right Now
When OpenAI reported $2B monthly revenue and $122B in funding, many observers called the AI race over. The $19B figure suggests otherwise. This is round two.
Key Metrics Side by Side
| Metric | OpenAI | Anthropic |
|---|---|---|
| Annualized Revenue | $25B | $19B |
| Total Funding | $122B | Undisclosed |
| Valuation | $852B | Undisclosed |
| Key Partner | Microsoft | |
| Infrastructure | Azure + proprietary | Google TPU |
| Open Standard | — | MCP (97M installs) |
| IPO | Q4 2026 expected | Not announced |
| Security Incident | Safety Bug Bounty | Source code leak (March) |
On raw numbers, OpenAI leads. But growth velocity and infrastructure strategy tell a more competitive story.
MCP at 97 Million Installs: Anthropic's Quiet Platform Play
This is the part I find most strategically interesting.
MCP (Model Context Protocol) is an open standard led by Anthropic. When MCP launched, it was introduced as "the standard for AI agents communicating with external tools." It's now crossed 97 million installs.
Why this matters strategically:
Platform effects. Once developers build MCP-based tools, Claude integration deepens naturally. It's an open standard, but it creates real ecosystem lock-in. Think Android — open source, but firmly in Google's orbit.
Competitor inclusion, Anthropic centrality. MCP is open enough that it works with OpenAI models too. But Anthropic built it, and Anthropic sits at the center of the ecosystem. Owning the standard is worth more than owning the platform.
If OpenAI's moat is the ChatGPT consumer brand, Anthropic's bet is on developer infrastructure — and MCP is the cornerstone of that bet.
What This Means for Developers
I've been using the Claude API for a while, and the trajectory has become more tangible recently.
Claude Sonnet 4.6 is now ranked first in real-world expert task evaluations and coding agent workflows. As covered in Claude Code's latest updates, Claude in agentic contexts is genuinely ahead of GPT-4o and Gemini on tasks that involve long context, precise instruction following, and multi-step reasoning.
Practical Impact Points
1. Context handling and instruction fidelity
Anthropic's strength in long context windows and accurate instruction following is real. On complex codebase analysis or lengthy document processing, the difference is measurable.
2. MCP ecosystem leverage
97 million installs means MCP-based tooling is proliferating fast. Learning MCP now gives you a transferable skill that works across models — it's not a Claude-only investment.
3. Stronger Google Cloud integration ahead
Post-TPU deal, expect tighter Anthropic–Google Cloud integration. Teams running primarily on Google Cloud infrastructure should see meaningful improvements in Claude API convenience and performance.
4. Security hygiene
The March 26 source code leak is not trivial. AI security issues aren't limited to OpenAI. If you're using Anthropic's API in production, now is a good time to audit your key management and access controls.
Honest Assessment
Anthropic is executing well. But clear-eyed evaluation matters.
$19B annualized revenue is impressive. But it's still $6 billion behind OpenAI — and more importantly, it trails OpenAI on consumer brand recognition by a much wider margin. In a world where "AI = ChatGPT" for most users, Anthropic remains primarily a developer and enterprise play.
The Google TPU deal is real, but 2027 is still more than a year away. OpenAI could accelerate its own infrastructure expansion during that window. New players may emerge.
MCP's 97 million installs is a positive signal, but installs don't equal active users. Real ecosystem dominance requires engagement metrics, not just downloads.
And the March source code leak is a reputational hit for a company that has staked its identity on "Constitutional AI" and safety-first development. The short-term damage may be limited, but enterprise customers with stringent security requirements will notice.
The bottom line: Anthropic has become a genuine competitor to OpenAI. Not just a credible alternative — an actual competitor that is growing faster and making structural infrastructure moves. Whether it can close the remaining gap depends on what happens after those Google TPUs come online and whether Claude builds meaningful consumer adoption alongside its developer base.
What's your experience with Claude vs. GPT-4o in production? I'd genuinely like to know where you're seeing the clearest differences.
References
- Anthropic Approaching $19 Billion in Annualized Revenue — The Wall Street Journal, April 2026
- Anthropic Strikes Deal With Google and Broadcom for Gigawatts of TPU Capacity — Bloomberg, April 2026
- MCP Crosses 97 Million Installs — MCP Official Blog, 2026
- Claude Source Code Leak March 2026 — TechCrunch, March 26, 2026
Related Posts:
- OpenAI Hits $2B Monthly Revenue + $122B Funding at $852B Valuation — OpenAI revenue analysis
- AI Data Centers at $1 Trillion: NVIDIA Vera Rubin — The infrastructure competition context
- MCP: The New Standard for AI Agents — Understanding MCP fully
- Claude Code April 2026 Update — Coding agent workflows in practice