The tech industry loves its garage startup stories. From Hewlett-Packard to Google, the stories of startups that became giants have inspired generations of entrepreneurs.
But the huge sums of money and computing power required for startups trying to make it with today’s hottest technology, the artificial intelligence used in chatbots like ChatGPT and Google Bard, may make those inspirational stories a thing of the past.
In 2019, Aiden Gomez and Nick Frosst left Google to create an AI startup in Toronto called Cohere, which could compete with their former employer. Several months later, they returned to Google and asked if it would sell them the enormous computing power they would need to build their own AI technology. After Google chief executive Sundar Pichai personally approved the arrangement, the tech giant gave them what they wanted.
“It’s ‘Game of Thrones.’ the computing power,” he said. “They choose who gets it.”
Building a pioneering AI startup is difficult without getting the support of “the hyperscalers,” who control the vast data centers capable of running AI systems. And that has put industry giants in the driver’s seat — once again — for what many expect to be the most significant change for the tech industry in decades.
OpenAI, the startup behind ChatGPT, recently raised $10 billion from Microsoft. It will pump most of that money back into Microsoft as it pays for time on the massive clusters of computer servers operated by the larger company. Containing thousands of specialized computer chips, these machines are essential to improving and expanding the capabilities of ChatGPT and similar technologies.
Competitors cannot keep pace with OpenAI unless they are given similar amounts of computing power. Cohere recently raised $270 million, bringing its total funding to more than $440 million. It will use much of that money to buy computing power from the likes of Google.
Other startups have made similar arrangements, notably a Silicon Valley company called Anthropic, which was founded in 2021 by a group of former OpenAI researchers; Character.AI, founded by two top Google researchers; and Inflection AI, founded by a former Google executive. Inflexión raised $1.3 billion in funding last week, bringing its total to $1.5 billion.
At Google, Mr. Gomez was part of a small research team that designed the Transformerthe underlying technology used to create chatbots such as ChatGPT and Google Bard.
The Transformer is a powerful example of what scientists call a neural network — a mathematical system that can learn skills by analyzing data. Neural networks have been around for years, helping to power everything from talking digital assistants like Siri to instant translation services like Google Translate.
The Transformer took the idea into new territory. By going through hundreds or even thousands of computer chips, it could analyze much more data, much faster.
Using this technology, companies like Google and OpenAI began building systems that learned from huge amounts of digital text, including Wikipedia articles, digital books, and chat logs. As these systems analyzed more and more data, they learned to generate text on their own, including terms, blog posts, poetry, and computer code.
These systems – called large language models – now support chatbots such as Google Bard and ChatGPT.
Well before the arrival of ChatGPT, Mr. Gomez left Google to start his own company with Mr. Frosst and another Toronto entrepreneur, Ivan Zhang. The goal was to build large language models rivaling Google.
At Google, he and his fellow researchers had access to nearly unlimited amounts of computing power. After leaving the company, he needed something similar. So he and his co-founders bought it from Google, which sells access to the same chips through cloud computing services.
Over the next three years, Cohere built a large language model that rivals almost any other. Now it sells this technology to other companies. The idea is to provide any company with the technology they need to build and run their own AI applications, from chatbots to search engines to personal tutors.
“The strategy is to build a platform that others can build on and experiment with,” Mr. Gomez said.
OpenAI offers a service along the same lines called GPT-4, which many businesses already use to build chatbots and other applications. This new technology can analyze, generate and edit text. But it will soon handle images and sounds as well. OpenAI is preparing a version of GPT-4 that can examine a photo, instantly describe it and even answer questions about it.
Microsoft Chief Executive Satya Nadella said the company’s arrangement with OpenAI is the kind of mutually beneficial relationship it has long nurtured with smaller competitors. “I grew up in a company that was always doing these kinds of deals with other companies,” he told The New York Times earlier this year.
As the industry races to match GPT-4, entrepreneurs, investors and experts are debating who the eventual winners will be. Most agree that OpenAI leads the field. But Cohere and a small group of other companies are building similar technology.
The tech giants are in a strong position because they have the vast resources needed to push these systems forward more than anyone else. Google too holds a patent on the Transformerthe fundamental technology behind the AI systems that Cohere and many other companies are building.
But there is a wild card: Open source software.
Meta, another giant with the computing power needed to build the next wave of AI, recently open-sourced its latest big language model, meaning anyone can reuse it and build on top of it. Many in the field believe that this type of freely available software will allow anyone to compete.
“Having the collective minds of all the researchers on Earth would beat any company,” said Amr Awadallah, head of AI startup Vectara and a former Google executive. But they will still have to pay for access to a competitor’s much larger data centers.