How OpenAI Grew From a Research Lab to 1 Billion Weekly ChatGPT Users
The story of how OpenAI grew ChatGPT to 1 billion weekly users through its API, a free chat product, word of mouth, subscriptions, enterprise, images, and agents
At 7:38 p.m. UTC on November 30, 2022, Sam Altman posted a short tweet:
"today we launched ChatGPT. try talking with it here:"
That was the launch.
OpenAI's own announcement was almost as understated. It called ChatGPT a research preview, made it free, and asked people to use it so the company could learn where the model worked and where it failed.
There was good reason for the modest launch.
A few months earlier, the project had an internal name that sounded more like a prototype: "Chat With GPT-3.5." Sam Altman later wrote that OpenAI was still a quiet research lab at the time. The team wanted to turn an experience people enjoyed in the API playground into a public demo and use the feedback to improve future models.
Nick Turley, who later became Head of ChatGPT, said OpenAI spent much longer on the underlying research but only about 10 days turning the final idea into something it could ship.
Five days after Altman's tweet, ChatGPT had crossed 1 million users.
Within two months, it had reached 100 million.
By August 2026, ChatGPT was available to more than 1 billion weekly active users.
OpenAI had spent seven years getting to the tweet. What happened after it came from a series of decisions that made its research easier to access, easier to understand, and easier for users to show to someone else.
OpenAI's growth timeline
| Year | What happened | Users / scale |
|---|---|---|
| 2015 | OpenAI launched as a research lab | No commercial product |
| 2019 | Microsoft invested $1 billion | OpenAI gained capital and large-scale compute |
| 2020 | OpenAI launched its first commercial API | Developers could build with GPT-3 |
| 2021 | GPT-3 API adoption grew | 300+ apps, tens of thousands of developers |
| 2022 | DALL-E 2 expanded to consumers | 1.5M active users by September |
| Nov. 2022 | ChatGPT launched as a free research preview | 1M users in 5 days |
| Jan. 2023 | ChatGPT continued spreading | 100M users within 2 months |
| Feb. 2023 | ChatGPT Plus launched | $20/month |
| May 2023 | ChatGPT launched on iPhone | ChatGPT moved from browser to mobile |
| Aug. 2023 | ChatGPT Enterprise launched | Consumer adoption moved into companies |
| Nov. 2023 | First OpenAI DevDay | 100M weekly users, 2M developers |
| 2025 | Native image generation launched in ChatGPT | 130M+ people generated 700M+ images in one week |
| Feb. 2026 | OpenAI disclosed new ChatGPT numbers | 900M+ weekly users, 50M+ consumer subscribers |
| Jul. 2026 | Codex adoption expanded | 5M+ weekly users |
| Aug. 2026 | ChatGPT passed another milestone | 1B+ weekly active users |
The first seven years were spent building the technology and reputation
OpenAI started in December 2015 as a nonprofit AI research organization.
Its launch announcement focused on research rather than products. OpenAI wanted to work on artificial intelligence, publish much of what it learned, and attract researchers who cared about building increasingly capable AI.
OpenAI's original announcement
For several years, most people outside technology had little reason to know OpenAI existed.
Researchers and developers did.
OpenAI released projects such as Gym for reinforcement learning, built systems that learned to play games, worked on robotics, and developed increasingly capable language models.
GPT arrived in 2018. GPT-2 followed in 2019.
Each release gave the company more credibility with researchers and developers before it had a mass-market product.
There was one obvious problem: training larger models required enormous amounts of money and computing power.
In 2019, Microsoft invested $1 billion in OpenAI. The companies agreed to build large AI supercomputing systems together, and Microsoft became OpenAI's exclusive cloud provider at the time.
Microsoft invests in and partners with OpenAI
OpenAI now had more compute and a partner with deep enterprise distribution.
The next question was what people would build with the models.
OpenAI let developers find the use cases
GPT-3 arrived in 2020.
OpenAI could have picked a few applications and built them itself. Instead, it launched an API and gave other developers access to the model.
A developer could send GPT-3 some text and receive generated text back.
That small interface covered a strange number of jobs. Developers used it for writing, search, education, support, games, summarization, classification, and products OpenAI had never considered.
Nine months after the API launched, OpenAI said more than 300 applications were using GPT-3. Tens of thousands of developers had access, and those applications were generating an average of 4.5 billion words per day.
GPT-3 powers the next generation of apps
The API gave OpenAI thousands of product experiments without requiring OpenAI to build each product.
If people repeatedly paid to use GPT for writing, coding, search, or support, that behavior told the company something useful.
Developers also became a distribution channel. Every company that built GPT into its own software put OpenAI's models in front of users OpenAI did not have to acquire itself.
That business was already working before ChatGPT existed.
DALL-E showed what happened when normal people could prompt a model
OpenAI's first big step toward a consumer audience came through images.
DALL-E let someone describe an image using normal language and generate it.
That made AI much easier to demonstrate.
A paragraph generated by GPT could look like any other paragraph. A picture of an astronaut riding a horse or a strange variation of a famous painting immediately made someone ask: How did you make that?
OpenAI gradually expanded access to DALL-E 2 during 2022.
By September, the company said more than 1.5 million active users were generating over 2 million images per day.
OpenAI removes the DALL-E waitlist
Mira Murati later described DALL-E as OpenAI's first real consumer-facing interface.
It also taught OpenAI something about distribution. People shared AI output because the output itself was interesting.
Every shared image doubled as a product demo.
That behavior showed up again a few months later.
OpenAI almost treated ChatGPT like a side project
In 2022, OpenAI already had GPT-3.5 available to developers.
People using the API playground had started talking back and forth with the model. Internally, OpenAI had also been working on instruction-following models that were better at doing what a user asked.
The team experimented with wrapping that work in a chat interface.
Sam Altman later described the project as "Chat With GPT-3.5."
The goal was largely research. OpenAI wanted more people to interact with the model so the company could collect feedback before later releases.
Mira Murati said the team did not anticipate how captivated people would become by talking to it.
Nick Turley remembers the final shipping process moving quickly. Once OpenAI decided to release the product, it took around ten days to get the public version out.
The first version was basic enough that it did not even have chat history.
On November 30, OpenAI put it online for free.
ChatGPT removed the work required to use GPT
Before ChatGPT, someone curious about GPT might need to know what an API was, request access, create a key, understand token pricing, write code, or find another application that had already integrated the model.
ChatGPT asked for a question.
The interface was already familiar because billions of people knew how messaging worked. A user could write normally, get a response, and continue the conversation if the first answer was not good enough.
You could ask:
Explain this like I'm 12.
Then:
Give me an example.
Then:
I still don't understand the second part.
The conversation itself became the interface.
That opened GPT to people who would never have touched an API.
OpenAI also made the research preview free. A new user did not have to understand the value of generative AI before paying for it.
They could find out in a few seconds.
Users figured out the marketing
The blank chat box created another problem.
What should you ask it?
Users answered that question for each other.
Screenshots started appearing on X, Reddit, LinkedIn, TikTok, YouTube, and group chats.
A programmer posted ChatGPT fixing code.
A student asked it to explain a difficult topic.
Someone had it write a poem.
Someone generated an email.
Someone tried to trick it.
Someone posted an answer that was completely wrong.
Each screenshot showed another job the same product could do.
One viral coding example taught programmers how to use ChatGPT. A lesson-plan example did the same for teachers. Marketing examples brought marketers in. Students showed other students.
OpenAI did not need to write onboarding for every profession because its users were already doing it in public.
Five days after launch, Sam Altman said ChatGPT had crossed 1 million users.
OpenAI later reported that it reached 100 million users in two months.
OpenAI data on ChatGPT adoption
Word of mouth worked unusually well because every user's conversation could become a different advertisement for the same product.
OpenAI waited until people wanted more before charging them
Rapid growth created a less exciting problem: serving all those requests cost money.
Altman described ChatGPT's compute costs as "eye-watering."
OpenAI sometimes had to limit access when demand was too high.
In February 2023, a little over two months after launch, OpenAI introduced ChatGPT Plus for $20 per month.
The first pitch was practical. Subscribers received better access during busy periods, faster responses, and priority access to new features.
The free product stayed.
That gave OpenAI a simple path from distribution to revenue. Anyone could try ChatGPT. People who used it enough to care about limits, speed, or new models had a reason to upgrade.
Later, newer models and higher usage limits made the paid plans more valuable.
By February 2026, OpenAI reported more than 50 million consumer subscribers.
The free tier was still bringing people in more than three years after launch.
ChatGPT also made OpenAI's API easier to sell
The API business kept growing alongside ChatGPT.
ChatGPT made the underlying technology familiar.
Before November 2022, a startup using GPT might have needed to explain what a language model was. After ChatGPT, customers already had a reference point.
"This works like ChatGPT for X" became enough to explain a new category of software.
OpenAI also cut API prices.
When it launched the ChatGPT API in March 2023, GPT-3.5 Turbo cost 90% less than the earlier GPT-3.5 models. OpenAI said it had also reduced ChatGPT's serving costs by 90% since December.
Introducing ChatGPT and Whisper APIs
Lower prices made more applications economical to build.
By OpenAI's first DevDay in November 2023, the company reported 2 million developers building on its platform.
The API gave OpenAI a second way to grow. ChatGPT acquired users directly, while developers put OpenAI models inside thousands of other products.
Mobile put ChatGPT somewhere people could use it every day
The ChatGPT iPhone app launched in May 2023.
It was free, synced conversations between devices, and supported voice input through Whisper.
Introducing the ChatGPT app for iOS
Android followed in July.
The browser had already proved that people wanted ChatGPT. Mobile increased the number of moments when they could use it.
People could ask ChatGPT something while shopping, commuting, traveling, studying away from a desk, or sitting in a meeting.
Over the next few years OpenAI added voice, images, file uploads, data analysis, memory, web search, coding tools, research, connectors, and other features to the same product.
More jobs could now start inside one chat box.
Employees brought ChatGPT into work before OpenAI sold it to their companies
ChatGPT's enterprise motion started with consumer adoption.
Employees were already using the product personally and then started using it for work.
When OpenAI launched ChatGPT Enterprise in August 2023, it said accounts associated with more than 80% of Fortune 500 companies had already used ChatGPT.
That figure did not mean 80% were enterprise customers. Employees at those companies had created ChatGPT accounts.
Introducing ChatGPT Enterprise
OpenAI could sell a company something many of its employees already understood.
Enterprise added controls that businesses needed around security, privacy, administration, limits, and company data.
The adoption path often began with one employee doing useful work, then a team, then a company-wide deployment.
By February 2026, OpenAI said more than 9 million paying business users used ChatGPT for work.
OpenAI later reported that its products reached 2.5 million businesses.
Images gave ChatGPT another huge burst of distribution
OpenAI returned to one of the lessons it had learned from DALL-E in March 2025.
It put much stronger native image generation directly inside ChatGPT.
The feature spread quickly across social media. People generated profile pictures, memes, diagrams, fake product shots, characters, and images in recognizable art styles.
OpenAI later said more than 130 million people generated over 700 million images in the first week.
OpenAI image generation API announcement
Sam Altman compared it with ChatGPT's original launch.
ChatGPT had needed five days to add its first million users.
During the image-generation surge in March 2025, Altman said ChatGPT added one million users in one hour.
The same distribution pattern from DALL-E now had ChatGPT's much larger installed base behind it.
People made something, posted it, and exposed the product to another group of people.
OpenAI kept making one account useful for more jobs
ChatGPT's growth after the initial viral launch depended on giving existing users more reasons to return.
OpenAI kept adding new kinds of work to the product.
Search gave ChatGPT access to fresh information.
Voice made it possible to talk instead of type.
Image generation covered visual work.
File uploads and data analysis brought spreadsheets and documents into the chat.
Deep Research handled longer research tasks.
Codex focused on coding.
Agents began taking actions rather than only generating answers.
One user could now use the same product to research a market, analyze a PDF, debug code, write an email, generate an image, search the web, or work through a spreadsheet.
That breadth makes ChatGPT harder to categorize, but it also gives OpenAI many chances to become part of someone's day.
Codex followed the same playbook
Coding became another large growth surface.
OpenAI turned Codex into a product that could do increasingly long pieces of engineering work.
By July 2026, the company said more than 5 million people used Codex each week.
More than 1 million of them were already using it for work outside software development.
ChatGPT is now a partner for your most ambitious work
That expansion looks familiar.
GPT initially required developers.
ChatGPT made language models accessible to everyone else.
Coding agents began with software engineers, then became useful to people who could describe what they wanted without being professional programmers.
OpenAI keeps widening the group that can use the same underlying capability.
From one tweet to more than a billion weekly users
By February 2026, OpenAI reported:
- 900 million+ weekly ChatGPT users
- 50 million+ consumer subscribers
- 9 million+ paying business users
Six months later, OpenAI said ChatGPT was available to more than 1 billion weekly active users.
A milestone in expanding access to AI
OpenAI now makes money from consumer subscriptions, business products, enterprise deployments, APIs, and advertising.
The path to that scale was less tidy than a normal SaaS growth plan.
Research attracted researchers and developers. The API let developers find commercial use cases. DALL-E showed that normal people would prompt AI and share the results. ChatGPT put the model behind an interface everyone already knew. Free access made experimentation cheap. Screenshots spread use cases. Plus monetized heavy users. Employees carried ChatGPT into companies. Mobile, images, search, voice, and agents gave those users more things to do.
Each step put the technology in front of people who had been excluded by the previous interface.
What OpenAI did especially well
It let people try the product before asking them to understand it
Generative AI was difficult to explain in 2022.
ChatGPT avoided the explanation.
A user could open the site and ask a question.
The product demonstrated itself.
For a new category, that was far more useful than a landing page full of copy about large language models.
It let users discover use cases
OpenAI released a general tool and watched what people did with it.
GPT-3 developers found writing, search, support, coding, and dozens of other applications.
ChatGPT users went further.
OpenAI could then build around behavior that already existed rather than guessing every job in advance.
Its output was easy to show someone else
Some products become less useful when their output leaves the app.
ChatGPT and DALL-E often became more interesting.
A screenshot could show exactly what the product had done for one person.
That gave OpenAI distribution without requiring a traditional referral program.
Free access gave the product room to spread
ChatGPT's first session did not ask for a credit card.
That helped when most people still did not know whether an AI chatbot had any value to them.
Once people had used it enough to build a habit, OpenAI had several ways to charge the ones who wanted more.
Consumer adoption helped enterprise sales
Companies usually have to teach employees how to use new enterprise software.
Many ChatGPT users arrived at work already knowing the product.
OpenAI could focus its enterprise product on security, administration, company data, and scale instead of teaching people how to chat with AI.
OpenAI kept reducing the skill required to use its models
The progression is visible across the company's history:
research papers → API → image prompt → free chat → mobile → voice → enterprise tools → agents
Each interface gave another group of people access to capabilities that previously required more technical skill or setup.
How did OpenAI grow so fast?
OpenAI's growth came from several channels working together.
Its early research built credibility with technical users. The API let developers distribute OpenAI models inside other products. DALL-E created consumer awareness. ChatGPT gave normal users a free and familiar interface. User screenshots and generated content drove word of mouth. Mobile increased usage frequency. Paid plans monetized heavy users. Consumer adoption moved into companies, while the API continued growing underneath thousands of third-party products.
ChatGPT's initial growth was unusually fast because the product took seconds to try and could perform enough different jobs that every community found its own reason to share it.
Frequently asked questions
When did OpenAI launch ChatGPT?
OpenAI launched ChatGPT on November 30, 2022 as a free research preview. Sam Altman announced the launch in a short tweet directing people to chat.openai.com.
How many users did ChatGPT get in its first week?
ChatGPT crossed 1 million users in about five days.
How long did ChatGPT take to reach 100 million users?
OpenAI says ChatGPT reached 100 million users within two months of launch.
How many people use ChatGPT in 2026?
OpenAI said ChatGPT was available to more than 1 billion weekly active users by August 2026.
How many people pay for ChatGPT?
OpenAI reported more than 50 million consumer subscribers in February 2026. It separately reported more than 9 million paying business users.
What was OpenAI's main growth channel?
Word of mouth drove much of ChatGPT's early consumer growth. Users posted conversations, screenshots, images, and use cases, which showed other people why they might want to try the product. OpenAI also grew through its developer API, Microsoft distribution, mobile apps, enterprise products, and partnerships.
Why was ChatGPT more successful than the GPT-3 API?
ChatGPT removed much of the setup required to use GPT. Someone could use normal language in a familiar chat interface for free, while the GPT-3 API was primarily built for developers. The simpler interface expanded the potential audience from technical users to almost anyone with a browser.
How does OpenAI make money?
OpenAI earns revenue from ChatGPT consumer subscriptions, business and enterprise plans, usage-based APIs, and advertising. In February 2026, OpenAI reported more than 50 million consumer subscribers and 9 million paying business users.
Sources
- Sam Altman — Original ChatGPT launch thread — November 30, 2022
- OpenAI — Introducing ChatGPT — November 30, 2022
- Sam Altman — Reflections — discussion of "Chat With GPT-3.5" and the unexpected ChatGPT growth curve
- Lenny's Podcast / Nick Turley — How ChatGPT became a product — Turley's account of the roughly 10-day final productization
- OpenAI — Introducing OpenAI — December 2015
- OpenAI — Microsoft invests in and partners with OpenAI — July 2019
- OpenAI — OpenAI API — June 2020
- OpenAI — GPT-3 powers the next generation of apps — March 2021
- OpenAI — DALL-E now available without waitlist — September 2022
- OpenAI — Introducing the Intelligence Age — data on ChatGPT reaching 100 million users in two months
- OpenAI — Introducing ChatGPT Plus — February 2023
- OpenAI — Introducing ChatGPT and Whisper APIs — March 2023
- OpenAI — Introducing the ChatGPT app for iOS — May 2023
- OpenAI — Introducing ChatGPT Enterprise — August 2023
- OpenAI — Announcing OpenAI DevDay — November 2023
- OpenAI — Introducing our latest image generation model in the API — April 2025
- OpenAI — Scaling AI for everyone — February 2026
- OpenAI — ChatGPT is now a partner for your most ambitious work — July 2026
- OpenAI — A milestone in expanding access to AI — August 2026
- OpenAI — The Work Now Within Reach — September 2026