Why hot AI startup Anthropic wanted a lower valuation

An unusual financial structure has led hot AI startup Anthropic to downplay its own value, highlighting the way some of the top companies in the industry are following new playbooks on how to build and fund companies.

Instead of widely opening up itself to new backers as it seeks to raise $750 million, which would be typical for a buzzy upstart, Anthropic went with existing investor Menlo Ventures as its lead financial support, people familiar with the effort said.

It also aimed to peg its worth somewhere between $15 billion and $20 billion, compared to the previous $5 billion. (In any other world, this would be considered a very big valuation jump, but this is the supercharged world of AI).

What has held it back is partly the terms of previous investments, which allow financial backers to gain more shares if Anthropic’s valuation reached certain thresholds, the people said. That incentivized the startup founded by former OpenAI employees to raise its market value only by so much.

From the outside, the fundraising looked a little strange to some people in the tech industry. But that’s partly because of how companies like Anthropic and OpenAI are structured. Along with traditional investments, they’ve also sought funding from cloud computing companies like Microsoft, Amazon, and Google.

Some of those investments come in the form of “cloud credits” or contracts. In essence, they got funding under the condition that a large portion would be spent on compute power. That has muddied the waters a bit, making it appear as if the cloud companies own a larger percentage of the companies than they do in reality, according to people familiar with the deals.

Some people in Silicon Valley wondered whether Anthropic was having trouble raising funds, given the amount of money it had raised compared to its valuation. OpenAI, by comparison, is reportedly raising at a $100 billion valuation. Anthropic is also one of the hottest tech companies in the industry with some of the top AI researchers in the business, though it has only recently begun building up its sales operation.

What’s also unusual is that Menlo structured its investment as a special purpose vehicle, with some of the other backers coming from the venture capital firm’s limited partners, who wanted to put money directly into Anthropic.

Startups normally would rather have a direct investment than an SPV because it’s the cleanest option. Anthropic was ultimately ok with the structure, partly because Menlo has been a key driver of new customers for the startup’s foundation models, according to two people familiar with the matter.

The Information first reported Anthropic’s latest fundraising round.

Anthropic’s investment history is complicated. Heavily influenced by the Effective Altruism movement, the company took on $580 million in 2022 in an investment round led by Sam Bankman-Fried, the founder of FTX who was recently convicted of fraud charges. The shares, paid for by FTX funds, are now in the hands of a bankruptcy trustee.

The trustee had at first attempted to sell the shares but later decided not to, hoping their value would rise to help pay back more victims of the FTX collapse.

Other investors in that round included The Center for Emerging Risk Research, now called Polaris Ventures, an EA-led organization aimed at reducing the risks of AI that could threaten humanity.

The Anthropic funding round, which is expected to close some time in the coming days, is a win for the company. But that’s not why it’s significant.

There’s a raging debate in the tech industry about where the value creation in AI will come from. Will it be the big, foundation model companies like OpenAI, Anthropic, Cohere, and Inflection? Or will it be the big tech players in the space such as Google, Amazon, Microsoft, and Meta. Or could it be the myriad “pick-and-shovel” players building a layer of infrastructure enabling this all to happen? Maybe it will be an application layer that doesn’t yet exist.

With the latest funding round in Anthropic, Menlo Ventures, which has some of the smartest investors in the AI space, is placing bets on how they see the future unfolding. They see OpenAI and Anthropic continuing to duke it out for supremacy and may both grow into massive companies as the AI pie grows larger.

But today, the biggest and most powerful AI models are most useful for consumers, who want one product that’s helpful for a myriad of purposes. In that space, OpenAI is dominant in two ways. It has the most advanced model and it was first to market with a consumer play, ChatGPT, that is quickly becoming the dominant brand.

Enterprise customers largely aren’t using OpenAI’s most advanced model, GPT-4, because it’s too expensive or slow. They’re seeking out more specialized models that are leaner, faster, and cheaper. Meta’s Llama series of models is quickly gaining traction with those customers.

Where, exactly, Anthropic fits in is an open question and plenty of people in the tech industry question its ability to compete with OpenAI and others. But people close to Anthropic insist that it is beginning to see hockey-stick growth in revenue.

That’s key for Anthropic because, unlike social media companies whose valuations were based almost entirely on user counts, the market sees foundation models as both consumer and enterprise software products that are more akin to business tools. Anthropic doesn’t appear to have much of a consumer following -— at least not yet.

It doesn’t have to be even close to profitable, but it needs to show that businesses are clamoring for its product in order to justify ever growing valuations.

That said, the products these companies are selling are really in their infancy and not enough infrastructure has been built to fully utilize them. This could be the AI equivalent of debating whether Ask Jeeves will beat Lycos in search supremacy.

Anthropic says it has deliberately moved more slowly than OpenAI in order to prioritize AI safety, an area that is a special focus at the EA-influenced company.

And that may not be a bad thing. While OpenAI has the first-mover advantage, it also has a first-mistake disadvantage. For instance, the recent lawsuit filed by the New York Times takes aim at OpenAI’s use of the news outlet’s vast amount of text to train its models.

The technology is so early that there may be an opportunity for come-from-behind wins.

• The Information broke the news on the latest fundraising round on Dec. 20, in what was supposed to be a quiet money raise over the holidays. The news stirred up even more interest in the investment.