Anthropic pulls in another $450 million to build safe AI

The AI research company Anthropic has raised another $450 million to further its development of safe large language models, and to fuel a budding rivalry with generative AI market leader OpenAI.

The funding round is led by Spark Capital, with participation by Google, Salesforce Ventures, Sound Ventures, Zoom Ventures, and others.

Before the new round, Anthropic had raised $1 billion, according to Pitchbook (and $1.3 billion, according to Crunchbase). The company’s last funding round was a $300 million raise March 9, after which it was valued at $4.6 billion, per Pitchbook.

The company says the new money will support its work on
“helpful, harmless, and honest” AI systems. More specifically, it will fund Anthropic’s development of new products, further its safety research, pay for additional head count, and help the company support businesses using its large language models and assistants.

Anthropic aims to raise up to $5 billion in the next two years to take on rival OpenAI, according to a Series C pitch deck seen by TechCrunch. The company wants to enter a dozen industry verticals, the deck said, offering enterprises in those spaces API access to its LLMs and other tools.

OpenAI currently boasts a much higher valuation than Anthropic—around $27 billion as of late April. It’s raised an estimated $300 million from VCs Sequoia Capital, Andreessen Horowitz, Thrive, K2 Global, and others, along with an estimated $10 billion in backing from its strategic partner Microsoft.

Anthropic’s Claude AI assistant is already a capable rival to OpenAI’s ChatGPT. The newest version of Claude, Anthropic says, now has a 100K context window, meaning that it can digest and analyze hundreds of pages of materials to inform its output. ChatGPT can take in only about 3,000 words of context.

Anthropic was founded in 2020 by a brother-sister team of former OpenAI researchers: Dario Amodei led efforts to build the GPT-2 and GPT-3 LLMs at OpenAI, and Daniela Amodei led the company’s safety and policy functions. The other cofounders—Jack Clark, Jared Kaplan, Sam McCandlish, and Tom Brown—also previously worked at OpenAI.