OpenAI vs. the Titans: How a Startup Made Google Play Catch-Up
ChatGPT made advanced AI something people could learn by using. The American company’s breakthrough combined research, human feedback and a simple invitation: ask another question.
On November 30, 2022, OpenAI put ChatGPT online as a free research preview. A visitor could ask a question, read the answer and ask a follow-up. Using it did not require learning a programming language. OpenAI’s launch announcement
By December 5, it had more than one million registered users, according to a later study of the service’s growth. By July 2025, that study reported more than 700 million weekly users. The questions extended far beyond coding: practical guidance, seeking information and writing accounted for nearly 80 percent of conversation topics. The usage study
Google did not open public access to Bard in the United States and United Kingdom until March 21, 2023. OpenAI had reached a mass audience with a general-purpose conversational product before Google launched that rival. Google’s Bard announcement
Google had contributed research that helped make ChatGPT possible, while Microsoft supplied investment and computing systems. OpenAI built on those contributions and its own research to release a chatbot the public wanted to use.
Why the next question mattered
People could work through an idea with ChatGPT by asking it to explain an answer, revise it or try a different approach.
Most questions were not about work. In the 2025 study, about 73 percent of the consumer messages classified in June concerned life outside work; 27 percent concerned work. The consumer-use research
We think that flexibility helped ChatGPT reach people beyond programmers and other technical users. The usage study documents what people did with it; it does not establish why they started using it.
An easy conversation could still produce an unreliable answer. OpenAI’s launch announcement warned that ChatGPT could produce plausible false answers. A fluent response could feel useful before a person had checked whether it was correct.
From a larger model to a better answer
In 2017, Google researchers introduced the Transformer, a model design for processing sequences such as words. OpenAI used that design in its 2018 research, training a model on large amounts of text before adapting it to specific tasks. Google’s original paper, OpenAI’s 2018 research
In 2020, OpenAI described GPT-3, with 175 billion parameters—the values adjusted as a model learns. Written instructions and examples could guide it through different language tasks without retraining it for each test. The GPT-3 research
There was still a gap between generating text and giving someone the answer they wanted. OpenAI addressed it by having people demonstrate desirable responses and rank the model’s answers. Those judgments guided additional training.
In January 2022, OpenAI reported that evaluators preferred answers from a 1.3-billion-parameter InstructGPT to those from the 175-billion-parameter GPT-3 on the customer requests studied. Evaluators preferred the model trained to follow instructions even though it had more than 100 times fewer parameters. The InstructGPT results
The smaller model won that comparison. Teaching it to follow requests mattered more than size alone. ChatGPT then put that approach into a conversation.
The computers behind the conversation
OpenAI’s December 2015 founding announcement described a nonprofit research organization whose supporters had committed $1 billion. The announced leadership included Sam Altman, Elon Musk, Ilya Sutskever and Greg Brockman. It planned to publish research and collaborate with other institutions. The founding announcement
Microsoft became a crucial partner. In July 2019, it announced a $1 billion investment and work with OpenAI on Azure supercomputing systems. In May 2020, Microsoft described a system built for OpenAI with more than 285,000 processor cores and 10,000 graphics-processing units. The system supplied the computing power needed to train the models. The investment, the supercomputer
Buying chips was only the start. Microsoft’s teams had to connect thousands of processors, keep them running together and supply power, cooling and software for long training runs. Microsoft’s infrastructure reporting
The race continues through rivals and partners
By September 8, 2026, OpenAI said its products reached more than one billion weekly active users and 2.5 million businesses. That statement covered its broader product business. Google had reported more than one billion monthly users for the Gemini app in August. The figures cover different products and time periods, so they cannot be used to rank the two companies’ audiences directly. OpenAI’s update, Google’s update
The two competitors also work together. On October 21, OpenAI announced plans to bring its models and Codex workflows to Google Cloud’s Model Garden, with a wider release expected in the coming weeks. Businesses would be able to buy OpenAI’s technology through Google Cloud Marketplace. The Google Cloud agreement
Microsoft remains OpenAI’s primary cloud partner and a major shareholder. Their April 2026 agreement allows OpenAI to serve products through other clouds while Microsoft retains a license to its models and products through 2032; OpenAI can also license them to other companies. The revised Microsoft agreement
Google would be selling its rival’s models alongside Gemini. Customers could choose either through the same cloud provider.
Expansion has required more investment. In February 2026, OpenAI announced $110 billion in new investment, including backing from SoftBank, Nvidia and Amazon. The financing announcement
ChatGPT’s lasting value will depend more on fewer confident mistakes than on another record audience. People use it for everyday guidance; they need answers they can check and a clear admission when it does not know. OpenAI’s own research acknowledges that its models can still give confident false answers. OpenAI’s reliability research
We see American enterprise in OpenAI’s success: a younger company made advanced AI usable through an ordinary conversation. Published research, access to computing systems and investment—including SoftBank’s backing—helped it compete. We welcome that foreign investment and want those opportunities to remain open to the next challenger.
About the reporting
This analysis draws on research papers, company announcements and the cited usage study. Company-reported reach, study findings and our interpretation are distinguished in the text. Research checked October 26, 2026; no new interviews were conducted.
Sources and measurement notes
The 2025 usage paper includes OpenAI-affiliated and academic authors. Its consumer-message analysis covers Free, Plus and Pro plans and uses automated classification. Registration totals, weekly usage and monthly usage are different measures. The 2026 company figures describe different product scopes. The InstructGPT comparison measures evaluator preference on the tested requests. Financing figures are dated announcements, not profits.
