OpenAI’s path to sufficient revenue for an AI takeoff in 2027
April 3, 2025
FutureSearch was a key contributor to AI 2027, a scenario centered around a leading AI company, “OpenBrain”, creating an intelligence explosion by building AI that speeds up its own AI research.
That work would be enormously expensive. Could they generate enough revenue to pull it off?
This piece adds color to the scenario we researched and was included in AI 2027. This is our view of the most plausible scenario in which OpenAI reaches $100B ARR by mid-2027.
The plausibility of this scenario can be used to ground one’s estimate of whether the entire AI 2027 piece is plausible.
The key questions we addressed are:
How much faster would OpenAI need to reach $100B, compared to the previous fastest companies to do so?
How would the revenue shift from primarily consumer, to primarily “agents”?
OpenAI Today
Our estimates for OpenAI in March 2025 are approximately $7B ARR growing at nearly 200% annualized. Our simple exponential growth model fits 3 key statements from senior OpenAI executives. (See Appendix B) Note that the revenue for a particular calendar year is substantially less than the ARR at the end of the year due to the rapid growth.
Extending the growth rate leads to $86B ARR by June 2027 and $100B+ by Aug 2027. This is in line with OpenBrain’s revenue in the scenario. Notably, OpenAI’s $1B revenue in 2023 grew roughly 270% to hit $3.7B in 2024, so the 200% growth rate in our model is a slight deceleration.
A key question is whether OpenAI’s growth rate is sustainable. How does it compare to the growth rates of other leading companies?
A New Record For $100B ARR?
The table shows how long it took 5 leading high growth companies to scale from $1B to $100B in revenue.
The time to scale from $1B to $100B has been rapidly decreasing. This makes sense for leading software/platform companies that have close to zero marginal cost and can hyperscale in an increasingly connected world.
If this trend were to continue, then a leading company of today could scale from $1B to $100B in just over 4 years! Applying this to OpenAI would indicate $100B revenue by mid-2027, which is consistent with our simple exponential growth model.
Exponential Growth in the LLM Business
Today, the dominant source of revenue for OpenAI is ChatGPT, a consumer product.
We identify four leading candidates for where $100B ARR by mid-2027 would come from:
Consumers/ChatGPT for personal use
Enterprise applications which we envision as co-pilots or assistants for workers
Replacement workers that are semi-autonomous agents capable of fully replacing humans for some existing jobs
API revenue from companies that build companies on top of foundation models.
We find API the hardest to estimate. API revenue could be the biggest revenue driver, if a large number of companies like FutureSearch are able to build totally new applications to serve a wide variety of industries.
But API revenue could be small if open source models are good enough, or competition heats up. After all, as of this writing neither GPT-4o nor GPT-4.5 are the best models for their cost.
So we reject API revenue as a plausible area of outsized growth, and focus on 1-3.
Revenue Breakdown
The table below presents a model for where OpenBrain’s revenue might reasonably be expected to come from given the details of ai-2027:
| Date | Total ARR ($B) | Consumer | Enterprise | Agents | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| subs (m) | $/mth | ARR ($B) | subs (m) | $/mth | ARR ($B) | subs (m) | $/mth | ARR ($B) | ||
| Apr 2025 | 8 | 23 | 20 | 5.5 | 3.8 | 50 | 2.3 | 0.2 | 200 | 0.5 |
| Aug 2025 | 12 | 30 | 20 | 7.2 | 5.5 | 50 | 3.3 | 0.5 | 200 | 1.2 |
| April 2026 | 22 | 42 | 22 | 11.1 | 9 | 50 | 5.4 | 1.4 | 331 | 5.4 |
| Aug 2026 | 35 | 48 | 22 | 12.7 | 11 | 60 | 7.9 | 2.0 | 634 | 14.9 |
| Dec 2026 | 50 | 54 | 22 | 14.3 | 12.5 | 60 | 9.0 | 3.0 | 760 | 27.2 |
| Mar 2027 | 73 | 58 | 25 | 17.4 | 13.5 | 100 | 16.2 | 4.0 | 812 | 39.4 |
| Jun 2027 | 99 | 62 | 25 | 18.6 | 14.5 | 100 | 17.4 | 5.5 | 956 | 62.6 |
Here are the corresponding growth rates, showing record-breaking CAGR:
| Date | Consumer | Enterprise | Agents | |||
|---|---|---|---|---|---|---|
| subs (m) | CAGR | subs (m) | CAGR | subs (m) | CAGR | |
| Apr 2025 | 23 | 3.8 | 0.2 | |||
| Aug 2025 | 30 | 121% | 5.5 | 202% | 0.5 | 1451% |
| April 2026 | 42 | 66% | 9 | 110% | 1.4 | 353% |
| Aug 2026 | 48 | 49% | 11 | 82% | 2.0 | 192% |
| Dec 2026 | 54 | 42% | 12.5 | 47% | 3.0 | 253% |
| Mar 2027 | 58 | 34% | 13.5 | 37% | 4.0 | 244% |
| Jun 2027 | 62 | 30% | 14.5 | 33% | 5.5 | 230% |
The table shows that in August 2025 Consumer is over half of OpenBrain’s revenue, while revenue from Agents is the smallest.
By June 2027 the revenue from Agents has grown dramatically and composes the majority of OpenBrain’s revenue.
Are these estimates reasonable? Is there really such a market for Consumer? Given Agent-3 from ai-2027, is it reasonable to expect nearly $60B from Agents by June 2027?
Consumer
To ground the consumer subscription numbers in the table, it is useful to consider that the growth in ChatGPT unpaid users has been explosive, reaching 400M monthly users in Feb 2025 up from 300M only 2 months earlier. We estimate that there are roughly 15M paying ChatGPT users in Feb 2025 paying $20/month based on 10M paying ChatGPT users in September 2024.
If the current growth trajectory of ChatGPT were to continue, it could reach 30M subs by August 2025.
Is there really enough consumer demand to support this? A nice comparison is the biggest consumer subscription company in the world, Netflix. Netflix currently has around 300M subscribers paying roughly $10/month. If consumers really find ChatBots useful, which trends suggest is absolutely the case, 30M paying users (a tenth of Netflix subscribers) seems plausible by August 2025.
Our model has this growing to 62M by mid 2027, or a fifth of current Netflix subscribers. This passes our sniff test of reasonableness.
Enterprise
In the model, Enterprise is listed as having 5.5M subs paying $50/month in Aug 2025.
For comparison, OpenAI has 2 million Enterprise subs in Feb 2025, up from 1 million in Sept 2025 according to Lightcap. Continuing this trajectory gets close to 5M subs by Aug 2025.
It appears that the Enterprise growth rate may be higher than the Consumer growth rate. MenloVC estimates that “AI spending surged to $13.8 billion [in 2024], more than 6x the $2.3 billion spent in 2023” corresponding to a 500% annualized growth rate. Microsoft reported AI revenue of $13B ARR in Feb 2025, up from $10B 3 months earlier, a 185% CAGR.
The growth rate in our table is far below the current industry trends, even after accounting for the increased pricing in our model.
In terms of total addressable market size, a useful reference is that Microsoft’s Productivity and Business Process segment had revenue of $78B in 2024. This is only part of Microsoft’s enterprise revenue, suggesting that our model’s $17B ARR of enterprise revenue in mid-2027 is possible.
Replacement Workers
By mid-2027 the dominant source of OpenBrain’s revenue is from agents that are both augmenting and replacing workers.
To investigate how big this market might be we consider five big areas for agents: Assistants, Customer Service Representatives, Knowledge workers (like analysts), Software Engineers, and R&D Researchers. Each of these is a distinct use case that supports progressively higher price points.
The table below shows how this might break down.
| Date | Total ARR ($B) | Assistants | Customer Service | Knowledge Workers | Software Engineers | R&D Researchers | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| subs | $/mth | ARR | subs | $/mth | ARR | subs | $/mth | ARR | subs | $/mth | ARR | subs | $/mth | ARR | ||
| Apr 2025 | 0.5 | 0.2 | 200 | 0.5 | 0.00 | 0 | 0.0 | 0.00 | 2000 | 0.0 | 0.00 | 10000 | 0.0 | 0.00 | 20000 | 0.0 |
| Aug 2025 | 1.2 | 0.5 | 200 | 1.2 | 0.00 | 0 | 0.0 | 0.00 | 2000 | 0.0 | 0.00 | 10000 | 0.0 | 0.00 | 20000 | 0.0 |
| April 2026 | 5.5 | 1.3 | 200 | 3.1 | 0.03 | 0 | 0.4 | 0.02 | 2000 | 0.6 | 0.01 | 10000 | 1.4 | 0.00 | 20000 | 0.0 |
| Aug 2026 | 15.0 | 1.8 | 250 | 5.4 | 0.06 | 0 | 0.7 | 0.05 | 2000 | 1.2 | 0.02 | 10000 | 2.9 | 0.02 | 20000 | 4.8 |
| Dec 2026 | 27.5 | 2.7 | 300 | 9.7 | 0.12 | 0 | 1.4 | 0.08 | 2000 | 1.9 | 0.04 | 10000 | 4.8 | 0.04 | 20000 | 9.6 |
| Mar 2027 | 39.8 | 3.6 | 300 | 13.0 | 0.20 | 0 | 2.4 | 0.12 | 2000 | 2.9 | 0.06 | 10000 | 7.2 | 0.06 | 20000 | 14.4 |
| Jun 2027 | 63.4 | 4.8 | 400 | 23.0 | 0.30 | 0 | 3.6 | 0.18 | 2000 | 4.3 | 0.09 | 10000 | 10.8 | 0.09 | 20000 | 21.6 |
The growth rates from the table above is here:
| Date | Assistants | Customer Service | Knowledge Workers | Software Engineers | R&D Researchers | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| subs | CAGR | subs | CAGR | subs | CAGR | subs | CAGR | subs | CAGR | |
| Apr 2025 | 0.2 | - | 0.00 | - | 0.00 | - | 0.00 | - | 0.00 | - |
| Aug 2025 | 0.5 | 1451% | 0.00 | - | 0.00 | - | 0.00 | - | 0.00 | - |
| April 2026 | 1.3 | 320% | 0.03 | 695% | 0.02 | 799% | 0.01 | - | 0.00 | - |
| Aug 2026 | 1.8 | 165% | 0.06 | 695% | 0.05 | 695% | 0.02 | 695% | 0.02 | 695% |
| Dec 2026 | 2.7 | 236% | 0.12 | 695% | 0.08 | 308% | 0.04 | 361% | 0.04 | 695% |
| Mar 2027 | 3.6 | 221% | 0.20 | 694% | 0.12 | 418% | 0.06 | 418% | 0.06 | 418% |
| Jun 2027 | 4.8 | 213% | 0.30 | 400% | 0.18 | 400% | 0.09 | 400% | 0.09 | 400% |
Let’s consider each of these 5 use cases separately to estimate potential subs, price points, and market sizes.
Assistants already exist in the form of OpenAI Deep Research (OAIDR) and comparable tools that help automate the research process for analysis. OpenAI currently charges $200/month for OAIDR. Reports are that OAIDR is generating $300M ARR as of Feb 2025 after launching in early Feb. Moreover, there is rapid adoption of coding co-pilots among software engineers. $1.3B ARR by August seems consistent with trends.
In our model, we have assistants increasing to nearly 5M paid subscribers at $400/month by mid-2027. We imagine these to be dramatically improved versions of OAIDR that serve as indispensable co-pilots for white collar workers in many industries from software to finance to marketing. If a professional with a $100k salary can have their productivity increased by 50% with an AI assistant, it makes business sense to spend $400/month ~ $5k/yr to provide assistants to nearly all white collar workers. From this perspective our 5M subscriber seems conservative.
Customer Service already uses chatbots to answer basic questions. This category imagines AI bots as full replacements for humans. There are currently roughly 3M customer service reps employed at an average compensation of $40k/year. Our model imagines nearly fully autonomous AI agents priced at $800/month = $10k/year starting in April 2026 and growing to 300,000 such AI agents by mid-2027. If these AI agents have comparable performance to humans then this might well be conservative, since it will make economic sense to replace humans with AI agents.
The Knowledge workers category is meant to cover white collar jobs including accounting, administrative, secretarial, and junior analyst jobs. Our model assumes AI agents capable of these higher level jobs would cost $2k/month = $24k/year. Our model imagines 180,000 such AI agents by mid-2027, far lower than the number of humans with these jobs today.
Fully autonomous Software Engineering AI agents are central to the scenario. Our model assumes that these agents cost $120k/year and that by mid-2027 there are 90,000 such AI agents. This salary is comparable to the average US salaries for a software engineer, but we’re assuming that these agents might be 10x as productive as a current human software engineer. Currently there are 1.6M software engineers in the US, getting paid $230B in salaries. The $10B ARR for this category by mid-2027 seems conservative given these assumptions.
OpenAI has already talked about charging $20k/month for AI agents to replace R&D Researchers. There are currently approximately 1.5M R&D researchers in the US with total salaries of $170B. Our model has these agents first appearing in mid-2026 and OpenBrain receiving $22B ARR by mid-2027. We imagine that these agents might be 10x as productive as current human researchers. While the $22B is far less the amount spent on human researchers today, it seems reasonable to expect that as the cost of research declines more money might be spent on this overall. This is Jevon’s paradox. Satya Nadella has speculated that the conditions for Jevon’s paradox are likely to hold for AI.
Discussion
This piece has attempted to illustrate how OpenBrain could reasonably expect to achieve $100B ARR by mid-2027 in the ai-2027 scenario. We believe that we’ve demonstrated this is entirely plausible by considering current financial statistics, growth trends, and market sizes for AI agents.
These are not intended to be forecasts.