ChatGPT sparked an AI arms race. 3 years later, nobody’s winning

Three years ago on Sunday, OpenAI released ChatGPT and accidentally started an arms race. The chatbot reached a million users in five days, forced Google to declare a “code red” and convinced Silicon Valley that artificial general intelligence was upon us and that whoever got there first would win big. Microsoft has invested $10 billion in OpenAI. Google rushed to release Bard. Chinese tech giants have been rushing to create their own chatbots.

This looked like a race to lose for OpenAI. Now, as we approach ChatGPT’s third anniversary, the lead has changed hands so much that the race is starting to feel like musical chairs.

The leader has changed hands several times this year alone

OpenAI entered 2025 as the undisputed leader. At the end of January, she was already under pressure. Chinese startup DeepSeek released models that matched OpenAI’s performance at a fraction of the cost, sending Nvidia shares tumbling and raising questions about whether U.S. tech giants had spent too much on the wrong approach. DeepSeek claimed to have trained its model for less than $6 million. Although underestimated, the lower-cost approach had disrupted assumptions about what it took to be competitive.

Then came summer. GPT-5, released in August after nearly two years of development, was supposed to provide what CEO Sam Altman called “PhD-level intelligence.” Instead, users got a model that labeled Oklahoma as “Gelahbrin” on maps and couldn’t solve basic algebra. Prediction markets that gave OpenAI a 75% chance of having the best AI model collapsed to 14% in a single hour.

Now Google is emphasizing its advantage. Gemini 3, released this month, has garnered rave reviews. Salesforce CEO Marc Benioff said after two hours of testing that he would “not go back.” Analysts say integrating the model into Google’s search engine gives it distribution advantages that OpenAI can’t match.

But Google’s triumph could be just as temporary. Meta’s Llama models power countless startups. Alibaba has just announced a major upgrade to its Qwen chatbot. Claude from Anthropic competes with ChatGPT on coding tests.

Google itself just published a paper on “Nested Learning” that some researchers compared to its 2017 Transformer breakthrough that powers all the big language model technology. If the new approach takes hold, it could reshuffle the cards again.

3 years is an eternity and an instant in AI

The pace of change makes predictions absurd. Three years ago, ChatGPT couldn’t browse the web, remember previous conversations, or generate images. Now he can do all three, as well as analyze spreadsheets, write code, and have voice conversations that sound remarkably human. Eight hundred million people use it every week.

But that same pace means the current leader is perpetually in danger and even analysts can’t keep up. In October, CNBC declared OpenAI’s dominance “unprecedented in Silicon Valley,” with one veteran investor telling CNBC that it was “the fastest period of startup creation and disruption” he had seen in nearly two decades. In November, Altman told staff to prepare for “difficult vibrations” and “temporary economic headwinds” after the Gemini release put OpenAI on its back.

Yet amid all the frenzy, at least one person warned that this was going to happen. In May 2023, six months after ChatGPT launched, an internal Google document was leaked that predicted the ups and downs to come. “We don’t have a gap, and neither does OpenAI,” one researcher wrote.

Unlike search algorithms or social networks – where data, infrastructure and network effects create lasting advantages – AI models could be copied, improved and shared freely. The researcher argued that open source developers would eventually eclipse the big players, that the industry’s obsession with scale was misguided, and that trying to control AI development was a losing game. Google ignored it. OpenAI too.

After what seems like a thousand news cycles later, the memo seems less like internal dissent and more like a prophecy. ChatGPT changed the world. But when advances can be made in a matter of months, every advantage is temporary. Three years of musical chairs later, OpenAI may have no room left when the music stops.