Anthropic has released new figures showing that Claude is taking on a steadily growing, increasingly routine role in the company's internal AI research and development. As of August this year, 26% of AI R&D work is led by Claude, a sharp rise from less than 1% at the start of the year, indicating that the company is advancing toward higher levels of automation and collaboration. The data also shows that during August the company was running 30,000 AI Agents simultaneously, each with its own identity and the ability to receive tasks through a shared messaging system, correct one another's errors, and form a large-scale collaborative network. These changes reflect Anthropic's long-term investment in automation tiers and computational efficiency, along with its continued refinement of content review and safety mechanisms.
How Anthropic defines automation tiers — and how they map to reality
The company uses the six-level automation framework proposed by Epoch AI to define the degree of AI involvement, ranging from AL0 (no AI participation at all) to AL5 (full autonomy). As of August, AL4 (AI-led) covered about 26% of R&D work; however, AL5 (fully autonomous) still has no observed real-world deployments, showing that current automation remains subject to strict safety and controllability limits. Compared with earlier figures, Anthropic has strengthened the task allocation and monitoring mechanisms between AI Agents, ensuring that every decision goes through online review, with offline monitoring flagging key conversation transcripts each week; after filtering, only a small number of entries are sent for human review. This workflow demonstrates that while the company is boosting R&D efficiency, it remains committed to rigorous risk control.
On the compute allocation side, figures released by Anthropic show that during the July holiday week, roughly 6% of AI R&D compute was devoted to safety research; calculated only against AI-driven R&D, that share is around 12%. These numbers illustrate that the company is placing growing weight on safety while seeking a balance with model training and testing efficiency. Its longer-term strategy is to rely on a diversified mix of compute sources and alliances to ensure that R&D can continue to scale within a controlled envelope.
Potential IPO scale and industry-chain interactions
At the same time, attention is turning to Anthropic's future capital-market moves. People familiar with the matter said, the company is preparing a potential initial public offering that could set a global record, with fundraising expected to reach into the hundreds of billions of US dollars and a post-listing valuation that could approach 2 trillion US dollars. Nvidia is seen as one of the most promising cornerstone investors and could commit as much as 10 billion US dollars, providing strong backing for the listing. Newly public information also indicates that negotiations are still being adjusted, and investment terms and deal structure could shift depending on market conditions.
Anthropic has also demonstrated deep ties to the chip and cloud supply chain. The company previously announced it would invest more than 1,000 billion US dollars in AWS and has committed to purchasing more than 300 billion US dollars of compute capacity over 10 years, much of it powered by Nvidia chips. Beyond Nvidia, Google, Amazon and Broadcom are also major investors and compute suppliers for the company, forming a cross-platform, diversified compute ecosystem. This arrangement mirrors the tight interplay between capital markets, chip supply and cloud computing that defines frontier AI firms, with significant implications for the value chain of the global technology industry.
Alongside this, Anthropic has built a long-term collaboration framework that spans major corporations. Last year the company announced it would channel more than 1,000 billion US dollars into AWS for long-term cloud computing resources, while pushing partners including Google and Broadcom to expand multi-gigawatt compute capacity. These partnerships and investment arrangements provide stable compute support for the company's Claude models and give third-party cloud and chip vendors room to pursue new experiments and innovations within long-term collaborations. For observers, the direction of travel means AI R&D is no longer a solo effort by any single company, but is being driven by an ecosystem of global tech giants working together.
Taken together, Anthropic's trajectory shows the company centering high levels of automation to lift R&D efficiency, while drawing on vast capital and a diversified compute base to power its push toward the upcoming peak of AI commercialization. Beyond watching the gap between AL4 and AL5, the market's response focuses on whether the company can, within a strict review and safety framework, raise its degree of automation to a level that delivers real commercial value.