The Ceiling for AI Valuations
Does revenue matter?
Over the last year, we’ve seen many funding announcements from various AI companies, some even having multiple rounds in the same year! Not only that, we’ve seen these numbers climb higher and higher, with no signal of stopping yet. It’s likely that 2024 will continue this journey towards the summit.
That summit? Most likely OpenAI’s latest $86B valuation. Now if OpenAI can have two funding rounds valued at 30B then $86B within a year, why not other AI companies?
In a sense, OpenAI will likely keep setting the anchor point where other AI companies, even with a slight chance of gaining revenue, will anchor towards.
Here’s a few examples from this year:
Hugging Face got $235M at a $4.5B valuation
Anthropic raised a massive amount at a $4.1B valuation
Inflection raised $1.3B at a $4B valuation
Mistral got $415M at a $2B valuation
Cohere got $270M for a $2.1B valuation
Runway raised $236M to be valued at $1.5B
Character AI raised $150M at a $1B valuation
X AI is seeking to raise $1B, so it’s likely in the same boat as above
I could go on and on but the trend is clear, the top AI startups are being valued at over a billion dollars, with a ceiling (for now) at ~$4B despite any other metrics. We’re likely to see this ceiling break upward in 2024, I’m betting generalized model startups like Anthropic and Inflection will surpass $10-20B next year. Keep note, traditional startup growth stories don’t apply here. Instead of revenue growth it’s model parameter growth, instead of user count it’s total addressable market. Is this true? We can find a few trends with the few data points we have...
Note with OpenAI reporting $1.3B in revenue, their revenue multiple is 66x, while other AI startups are far from reaching even their first $100M (and some even pre-revenue), this puts many of the startups above well above this metric, so it isn’t a viable one.
In an earlier blog post, “How to hit a home run” we talked about how incremental value should win over the go big or go home mentality, but that doesn’t apply here. This time let’s talk about what types of features are shared between these large fund raises:
These companies train their own models
Using GPT3.5-4 to run your business isn’t going to cut it for these unicorn status startups. Not only would you be dependent on OpenAI, but you’re also not running a zero marginal cost business.
The startups that get away with using another model must have a unique dataset that no one else has access to, but these would be hard pressed to enter the unicorn valuations
Generalized models are more valuable than specific models (for now)
Presumably because they have a greater number of use cases which could provide value to a larger total addressable market, however, one must wonder whether focusing on a specific use case and model could prove out the return on investment better?
Expertise with AI, AI infrastructure, or proven track record
The founders of all these startups have deep AI knowledge or first mover advantage
They’re likely people that are going to be hard to catch up to
If a startup is not full of people who are experts in AI or AI infrastructure, it will be harder to raise funds
If you don’t have the expertise, being a proven founder helps, because proven founders have the ability to hire expertise
Revenue is not important
In these valuations, revenue multiples are all over the place, some are even pre-revenue. As such, in the world of AI, revenue is not as big of a factor as potential reach
Large capital is needed for training foundational models
Another reason for these valuations are the capital needed to get H100s and train bigger LLMs, with smaller valuations, these raises would take a huge percentage of the company
Data might become more guarded as companies start trying to get more data for their models
That doesn’t mean it’s “worth” it for the startup, but it is a factor here
So what really determines what the next round of valuations look like? It seems that if revenue is not important, the other bullet points above are actually not based on progress or milestones. That means that the main factor in determining valuations is actually OpenAI’s valuation itself.
That means the ceilings we see around $1B to $4B are not actually revenue multiples, they are OpenAI multiples. OpenAI at $86B? Why not have something valued at 1/20th of it?
Does this mean that from here on out, the ceilings will be based on OpenAI’s next funding round? If OpenAI hits $200B, will $5-10B be the standard across all these current startups?
While that seems impossible to fathom now, we’re entering a booming stock market and interest rate cuts next year. We’re going to see some insane valuations across the board, so it’s not out of the question. 2024 may be the era of $10B startups with very low user and revenue counts. In the world of AI, what matters of course is your expertise, total addressable market, and how many people you can potentially reach.
At some point, these numbers will need to convert into revenue and user counts, because the expectation is for some liquidity event in the future. It’s going to be hard for a startup to be acquired when faced with a $1B vs. $10B valuation, and it may be that most of these AI startups are shooting for an IPO instead. Even then, different numbers matter for a liquidity event when compared to a VC raise.
One wild card? With more fund raises and more competition, pricing power might be affected as well. For instance, OpenAI has cleverly priced its offerings between cheap and powerful. With GPT3.5 Turbo costing $0.001/$0.002 per thousand tokens and GPT4-Turbo jumping to $0.01/$0.03. Their GPT4 pricing seems odd because at ($0.03/$0.06) there doesn’t seem to be a use with GPT4-Turbo available. Just like their valuation, OpenAI has also set an “anchor” for others to price at as well. Claude has to price at $0.0008/$0.0024 to $0.008/$0.024 (converted to 1k tokens) for their fast to powerful models. Mistral’s models are priced from $0.00015/$0.00046 to $0.00273/$0.00819 on the same scale. See a pattern here? Look for pricing to get disrupted as more and more startups enter the race towards $86B.
Okay, one final shocker to end things with, the OpenAI valuation of $86B is actually after all those other fund raises had already closed, so we’re probably already in the next ceiling phase…let’s see how things evolve as we enter 2024, until then, we have some more model training to do…