Anthropic CEO Dario Amodei on AGI Timelines and AI Scaling

Dario Amodei's AGI Outlook

• 💡 Dario Amodei, CEO of Anthropic, a physicist and former OpenAI VP, provides a detailed AGI roadmap with specific timelines and revenue figures.
• 🎯 He notes that the exponential growth of AI technology has largely met his expectations since 2017, with progress from "smart high school student" to "PhD and professional stuff."
• ⚠️ Amodei expresses surprise at the lack of public recognition of how close humanity is to the "end of the exponential" in AI development.

The "Big Blob of Compute" Hypothesis

• 🧠 Amodei's 2017 hypothesis posits that only a few core factors truly drive AI progress, not "cleverness" or new methods.
• ⚙️ Key factors include raw compute, data quantity and quality, training duration, and a scalable objective function (like pre-training or RL objective functions).
• ✅ The hypothesis also emphasizes numerical stability to ensure smooth, laminar flow of computation.

Visible Path to Superhuman AI

• 🚀 Amodei describes a unified scaling law across the entire AI stack, where both pre-training and Reinforcement Learning (RL) exhibit log-linear gains.
• 📈 This consistent scaling confirms that the path to superhuman performance is now visible and largely undisputed within the field.
• 🔮 He predicts a 90% chance of reaching a "country of geniuses in a data center" within ten years, with a personal hunch it could be one to three years for verifiable tasks like coding.

Anthropic's Rapid Revenue Growth

• 💰 Anthropic's revenue has shown bizarre 10x per year growth, from effectively zero to $9 billion annualized by late 2025, with further billions added in early 2026.
• 📊 This financial acceleration provides real-world evidence backing Amodei's predictions for capability curves, demonstrating the rapid market adoption.

Trillions in AI Revenue by 2030

• 🌐 Amodei predicts trillions of dollars in AI revenue before 2030, driven by the combined force of technical exponential growth and economic diffusion.
• ⏳ While AI diffusion is faster than any previous technology, it is not instant, facing challenges like enterprise change management and security permissions.
• 🎯 The core question is not if this will happen, but how quickly the world adapts to the rapid advancements and economic shifts.