Cash Burn
Open AI (pvt) does not give out detailed financial information to the likes of us, but on occasion, they have to give some indication about how things are going financially to their investors and some of that information leaks out. While one certainly cannot build an accurate model around rumors and hints, to us the issue is more how do the numbers we can see compare to those that the company professed were accurate months ago, sort of a credibility check. We show the difference between revenue estimates (company) made in 1Q of this year against those most recently made (3Q), and while this year remained the same, the out years saw increases, not unlike most outyear projections.
Figure 1 - OpenAI Revenue Estimate Change - 1Q '25 to 3Q '25 - Source: OpenAI, The Information
There is some detail that we cannot show, primarily higher growth from ChatGPT than originally expected, but we question (not shown) the increase in the ‘New products’ category that accounts for much of the incremental growth, as this includes an increase in the monetization rate, essentially converting more free users to paid plans. OpenAI indicated that user monetization is expected to reach $2/user in 2026, increasing to $30 by 2030, against a projected 2 billion active users at a margin between 80% and 85%. This gain offset the decrease in API revenue of $5b over the next 5 years and the decrease of $26b from AI agent revenue.
On the cost side, OpenAI is expected to burn through $115b (cumulative) in cash between 2025 and 2029. This is a mere $80b higher than originally expected, on top of the $2b already spent. As its costs rose, OpenAI decided to develop its own data center processor and infrastructure, rather than be at the mercy of uncontrollable costs, adding substantially to the cash burn, and the more recent competition between AI developers for talent has increased the cost of stock-based compensation by $20b more than originally forecasted.
This year, the cash burn has increased from $6.5b to $8b and is now expected to hit $17b next year, $10b more than originally expected while cash burn in 2027 could hit $35b and $45b the following year. That $45b cash burn estimate is ~4 times the previous $11b estimate. The company seems to have maintained its own expectations for running its models this year, but through 2030 even small increases each year in computing costs totaled an additional $11b increment during the period, with the total computing cost spend exceeding over $150b over the 5 year period. Training costs are also running ahead of plan, with the original $7b now expected to be $9b this year and the $17b for training in 2026 is now $19b.
All in, none of this matters because institutional investors are still foaming at the mouth to buy into OpenAI at a $500b valuation, almost 2x the valuation from 6 months ago, likely as they expect the not-for-profit entity to find its way into the public market, although not if the great and powerful Mr. Musk has his way. I thought I heard someone say “Dot com”?
On the cost side, OpenAI is expected to burn through $115b (cumulative) in cash between 2025 and 2029. This is a mere $80b higher than originally expected, on top of the $2b already spent. As its costs rose, OpenAI decided to develop its own data center processor and infrastructure, rather than be at the mercy of uncontrollable costs, adding substantially to the cash burn, and the more recent competition between AI developers for talent has increased the cost of stock-based compensation by $20b more than originally forecasted.
This year, the cash burn has increased from $6.5b to $8b and is now expected to hit $17b next year, $10b more than originally expected while cash burn in 2027 could hit $35b and $45b the following year. That $45b cash burn estimate is ~4 times the previous $11b estimate. The company seems to have maintained its own expectations for running its models this year, but through 2030 even small increases each year in computing costs totaled an additional $11b increment during the period, with the total computing cost spend exceeding over $150b over the 5 year period. Training costs are also running ahead of plan, with the original $7b now expected to be $9b this year and the $17b for training in 2026 is now $19b.
All in, none of this matters because institutional investors are still foaming at the mouth to buy into OpenAI at a $500b valuation, almost 2x the valuation from 6 months ago, likely as they expect the not-for-profit entity to find its way into the public market, although not if the great and powerful Mr. Musk has his way. I thought I heard someone say “Dot com”?
RSS Feed