OpenAI and Anthropic Revenue Breakdown
Breaking down revenue growth, the consumer subscription businesses and the importance of partnerships to the API business
This is a weekly newsletter about the business of the technology industry. To receive Tanay’s Newsletter in your inbox, subscribe here for free:
Hi friends,
With both OpenAI and Anthropic rumored to be raising new financing rounds, some of their financial metrics have been reported by various publications. In this piece, I’ll discuss what we know about their financials and highlight a few observations from that data.
OpenAI
Revenue: OpenAI is at ~$3.6B in annualized run rate revenue as of August 2024, and was roughly at $1.6B in annualized run rate revenue near the tail end of 2023. Based on their projected total revenue of $3.7B for 2024, I estimate OpenAI ending 2024 at ~$5B-5.2B in annualized run rate revenue. This would represent a growth rate of 225% y/y from $1.6B at the end of last year. OpenAI also further projects to get to $11.6B in revenue in 2025, an increase of 213% y/y next year if they are able to achieve it.
Revenue Breakdown: The split of OpenAI’s revenue, per company projections for the end of 2024 is roughly:
ChatGPT subscriptions: $2.7B (~73%) representing 285% y/y growth
This includes about 10M subscribers on the ChatGPT plus plan and another 1M on higher priced plans
API: $1B (~27%) representing 200-225% y/y growth
Margins: We don’t know much about OpenAI’s margins, but given reports that they are expected to lose about $5B this year on about $3.7B of revenue which is obviously not great. Futuresearch estimates that their API business actually has ~50% gross margins, and that most of the losses come from operating costs (R&D, etc) and their ChatGPT business, which gives basically unlimited usage for ~$20/month.
Anthropic
Revenue: Anthropic is expected to be at $1B annualized run rate revenue at the end of this year, about a 900% increase from the ~$100M or so in annualized run rate revenue they were generating at the end of 2023.
Revenue Breakdown: Anthropic’s revenue breakdown skews much more towards its API, particularly through it’s 3P API which is largely via Amazon. Their breakdown is roughly:
Third Party API (via Amazon, etc): 60-75%
Direct API: 10-25%
Claude chatbot subscriptions: 15%
Professional Services: 2%
Margins: We don’t know much about Anthropic’s margins either, but again, they are expected to lose about ~$2B this year, on revenue in the high hundreds of millions. While a lot of this is largely compute and headcount, their business skew more towards they API, which if it does indeed support higher gross margins in the short to medium term, may be beneficial.
If you don’t yet receive Tanay's newsletter in your email inbox, please join the 9,000+ subscribers who do:
Observations
Putting the above together, we can make a few observations when comparing the two. While OpenAI is 5x larger than Anthropic, OpenAI is much more dominant on the consumer side, while the two are much closer on the API side.
1/ OpenAI is at 5x the Revenue Scale
From a scale perspective, OpenAI is roughly 5x the scale of Anthropic in terms of annualized revenue (~$5B vs ~$1B). Anthropic is growing faster, however. At the start of the year, OpenAI was likely around 15x the scale of Anthropic.
2/ ChatGPT is by far the dominant prosumer product
Despite a lot of buzz around Anthropic’s latest class of models including Sonnet, and around Claude and features such as Artifacts, the vast majority of people who use ChatGPT have likely never even tried or heard of Anthropic’s Claude.
From a revenue perspective, ChatGPT will be bringing in about ~$2.7B at the end of the year for OpenAI, relative to Anthropic’s Claude bringing in about ~$150M. That makes ChatGPT about 18x the size from a revenue perspective.
Anthropic recently launched an enterprise version of Claude to bolster their Claude offering for businesses, but even there is likely far behind, with ChatGPT already boasting over 1M users on higher paid ChatGPT plans (enterprise/teams).
3/ Distribution remains king, even for developer products
Anthropic generating ~65% of their revenue from their third party API highlights how important distribution is. According to many benchmarks, at least until OpenAI launched o-1, Anthropic’s sonnet and similar models have been the best for a number of use cases. But having a great product alone is insufficient, and distribution is equally if not even more important.
For Anthropic, that distribution, in the form of their partnership with Amazon to allow their models to be made available on AWS Bedrock seems to be driving the vast majority of their revenue.
4/ The competition for API market share is quite close
I’m a bit surprised by how close things seem to be on the API front. It would appear that OpenAI is somewhere around ~$1.2-1.5B in run-rate revenue from its API revenue, while Anthropic is at around ~$800M in run-rate revenue. That would put OpenAI only about 50-100% larger than Anthropic on the API side, which is a much smaller gap than in aggregate.
Again, this seems to be largely driven by Anthropic’s success on the AWS platform. But I’m also surprised that OpenAI isn’t seeing even more uptick than they are from the Azure OpenAI usage and it raises the question around whether there is some difference in how Azure vs AWS API revenue is accounted for between OpenAI and Anthropic respectively.
5/ Capital requirements continue to be huge
Based on the losses that both have and the reported rounds they are raising ($6.5B in the case of OpenAI and likely >1B in the case of Anthropic), it’s clear that foundation models will continue to require large amounts of capital, and that the number of players that can compete is quite small. We’re basically down to OpenAI, Anthropic, Meta, Mistral, Google and X.AI as well as various other attempts by Amazon/Microsoft.
In addition, as these companies push towards profitability at some point (even if not immediately as capital is still available), they seem to be betting on a few trends:
A large reduction in inference costs over time to improve margins on the subscription products / API
Potential price increases in consumer subscription plans as model capabiltiy improves. OpenAI has already hinted at increasing prices to $22/mo this year and up to $44/month over the coming years.
A shift from training-time compute to inference-time compute, as highlighted by o1, to somewhat slow the ballooning training costs every 1-2 years.
Sources and Additional Reading
In closing, here are a few of the publications and reports I used to triangulate various datapoints:
As Anthropic’s API gains traction, the competition in AI infrastructure is no longer just about innovation—it’s about strategic alliances.
Thanks for this great analysis!
> here are a few of the publications and reports I used to triangulate various datapoints
It would be great if you could share more details on what data points you used for your triangulation process!