(NEW YORK TIMES)

After several years of criticism that it was lagging behind in artificial intelligence, Google is turning up its competition with smaller startups such as OpenAI and Anthropic.

On Tuesday, Google released a new AI model, Gemini 3, that improves on its predecessor’s ability to create software programs, organise email, and help businesses analyse documents. It can also mix graphics and text together when responding to requests involving travel itineraries, history, or art.

Google said the model would be available in its Gemini app and could be used to fulfill Google Search queries in AI Mode, a conversational search feature it introduced this year.

The impending release of Google’s newest model stoked anxiety inside the companies that jump-started the AI arms race.

At OpenAI and Anthropic, two of Google’s closest AI rivals, workers have speculated that it could be bad for their businesses if Google’s models outperform theirs in tasks such as autonomous coding or image generation, two people with knowledge of the discussions said.

"We’re in a situation where - because of Google’s size and space and their first-mover advantage in search - Gemini could take market share and cause OpenAI and others to fall behind,” said Mike O’Rourke, the chief market strategist at JonesTrading, an institutional trading firm. He said such a turn toward Gemini could ripple through the market, raising questions for companies such as Oracle and Microsoft, which OpenAI has committed to pay billions of dollars for computing.

The AI boom is also facing questions about whether the soaring costs of the technology can be justified by the business opportunities it is creating. AI systems today are primarily used to fulfill traditional internet search queries and help software engineers automate computer programming.

But the technology runs in massive data centers filled with expensive supercomputers, which the industry is expected to spend nearly $7 trillion to build by 2030, according to McKinsey & Co.

Across Wall Street, investors have grown skeptical that companies such as Anthropic, OpenAI, Microsoft, and Google can generate sufficient sales to cover their spending.

In a press briefing, Google said the information produced by Gemini 3 was 72% accurate, according to a standard benchmark test. That is notably high for an AI model of this kind, but it may not be what the average user expects from a technology that will help drive the Google search engine.