What are fair valuations of OpenAI and Anthropic?

The other is via the usage-based pricing for their APIs offered to businesses building on top of models like GPT-3 and GPT-4. Pricing differs based on the size of the context window and the particular model—for example, GPT-4 with a smaller 8K context window costs $0.03 per 1K tokens for prompts and $0.06 for completions, while the less powerful gpt-3.5-turbo model optimized for chat dialogue costs just $0.002 per 1K tokens. 

OpenAI also makes money from renting access to their DALL-E image model, Whisper audio model, and fine-tuning and embedding. Enterprise customers spending upwards of $45K/month get discount pricing.

Despite soaring revenues, OpenAI is highly unprofitable. The company lost $540M last year while developing ChatGPT, and those losses are expected to increase dramatically in 2023 with the growth in popularity of their consumer tools. Sam Altman said earlier this month that OpenAI is going to be “the most capital-intensive startup in Silicon Valley history.”

The reason for that is that operating ChatGPT is massively expensive. One analysis of ChatGPT put the running cost at about $700,000 per day taking into account the underlying costs of GPU hours and hardware. That amount—derived from the 175 billion parameter-large architecture of GPT-3—would be even higher with the 100 trillion parameters of GPT-4.

Corporate structure

The sky-high of training and deploying models like GPT-4 helps explain the unusual terms of Microsoft’s roughly $13B of investment in OpenAI over the last few years.

OpenAI’s hybrid corporate structure, with both a for-profit business wing and a non-profit research lab wing, determines how investors in the company will eventually be paid out. A wrinkle in the for-profit wing of the business is that profits are capped: OpenAI’s earliest investors and employees are limited to making 100x their initial investment. The combined organization is run by OpenAI’s non-profit arm. 

Once the for-profit business arm begins to return profits, the first people to get paid out will be the very earliest investors in the company, who will get their principal paid back.

After those early investors are paid out their principal, 25% of OpenAI’s profits will go to employees and to pay early investors (until they hit their profit cap), while 75% will go to Microsoft until it recoups its $13B principal.

After Microsoft recoups its $13B, it will get 50% of all OpenAI profits until it gets to $92B (at which point they’ll hit the profit cap), while 49% will go to early investors and employees and 2% will go to OpenAI’s non-profit arm.

Once $92B in profit is generated and paid to Microsoft—along with that $13B in principal—all equity reverts back to OpenAI, along with 100% of future profits.

This structure operates like a hedge—for OpenAI, it is a way to make sure that the company has the capital and institutional backing that it needs to survive in the short-term given that the profit-making capacity of the company is still unproven, with a large long-term reward in the event that they’re successful in making it work.

Anthropic

Anthropic last raised at a valuation of $4.1B in March 2023. Anthropic is much earlier along than OpenAI with less visibility into its economics, and therefore it’s tricker to go bottom-up to think about their potential fundamental valuation. 

Anthropic, by contrast to OpenAI, is generating minimal revenue—its chatbot Claude has only become available via application within the last few months. Before that, the company was only known to be working with Notion, Quora and DuckDuckGo in a closed alpha program.

It is likely to grow generating the vast majority of its revenue from B2B deals vs. the more consumer-oriented approach of an OpenAI.

One key advantage for Anthropic is their team—the company was founded by Dario Amodei, the former VP of research at OpenAI, along with a number of other OpenAI employees. 

Also, Google invested nearly $400M in the company in late 2022 in exchange for a 10% equity stake—similar to the arrangement created by OpenAI with Microsoft, who invested $1B in the AI lab in the summer of 2019 and followed that up with a $10B investment in early 2023.

Conclusion

OpenAI’s present $29B valuation is the result of the stock’s scarcity, enthusiasm about the potential future earnings of the company, and the prospect that OpenAI—if able to build and maintain a moat in artificial intelligence—could build the next trillion dollar plus company.

For Anthropic, the level of uncertainty around their valuation of $4.1B is similar. It’s not 100% clear whether they can compete with OpenAI, and if they can, it’s not clear what kind of moat they can build in this space. 

That said, both companies enjoy a few key advantages, from teams that have been working in this space since the beginning to partnerships with the biggest cloud providers. The potential for their products to generate revenue through both B2C and B2B methods has been demonstrated. 

Whether these companies can build moats—and whether or not such moats are even possible in this space—is still unknown.