Google is introducing a lower-cost option to its top-tier generative artificial intelligence service, saying businesses can save up to $1 billion a year by moving the majority of their operations to AI.
The move comes as the US tech giant aims to attract more users in a highly competitive field.
The Alphabet-owned company announced Tuesday at the annual Google I/O conference in Mountain View, Calif., that Gemini-powered Google AI Ultra subscriptions are now accessible for $100 per month.
That’s significantly lower than the $250 price tag for what used to be the only option for that service, but with some features removed. The full version of AI Ultra comes with higher usage limits and more Google Cloud storage, and is currently reduced to $200.
Users and analysts were hesitant about the price of the initial service, especially given that the lower tiers, AI Plus and AI Pro, are available at $8 and $20, respectively. All three tiers are enhanced with additional features.
For comparison, Anthropic’s Claude’s top-of-the-line service starts at $100, while OpenAI pioneer ChatGPT costs $200. DeepSeek, on the other hand, has a pay-as-you-go model, so there is no monthly subscription, and the cost for frequent or professional use is estimated at around $25 per month.
This could also be a strategy to fend off challenges from low-cost generative AI services, particularly China’s DeepSeek. DeepSeek’s rock-bottom price has rocked the market, raised questions about the true cost of developing language models at scale, and even caused the stock price of AI chip darling Nvidia to plummet by $600 billion last year.
However, Google did not explicitly state that this was the reason for the price reduction. Instead, CEO Sundar Pichai said the move capitalizes on Gemini’s momentum, which has more than doubled its user base to about 900 million in the past year, and the potential cost efficiencies that AI can provide.
“If companies were using mixes, [AI] If you are a model, you can save a lot of money. To put this into perspective, top companies process about 1 trillion tokens a day,” he said at a media briefing ahead of Google I/O.
“If you move 80% of your workloads to a combination of Frontier models like 3.5 Flash and 3.5 Pro, you’ll save more than $1 billion a year,” he said, referring to the Gemini model, which boasts speed and “PhD-level reasoning.”
“You’ve probably heard anecdotes from other chief information officers about companies already burning through their annual token budgets… so what we’re talking about is real savings that companies can funnel into themselves,” Pichai added.
AI updates
At Google I/O, the California-based company announced Gemini 3.5 Flash, its most advanced product for coding and agent AI. Google also said it introduced Gemini Omni, a successor to Nano Banana that focuses on images and video and uses more accurate physics to create visuals.
Google’s foundational search has also been updated with Gemini 3.5 Flash, allowing users to create multiple agents within the service.
YouTube has updated its video sharing platform to Ask YouTube, which Google said will allow it to generate results even from complex queries. The company also launched Docs Live, a voice-powered feature that helps you compose your drafts while pulling information from Google services like Drive, Gmail, and the web.
“In the future, you’ll be able to create new documents and edit them all directly with your voice,” Pichai said.
Google is significantly ramping up its investments in AI, aiming to keep pace with other leading AI companies, especially at a time when more businesses and people are adopting AI technology.
The company has already announced that it will increase its AI capital spending to up to $190 billion in 2026, up from about $31 billion in 2022. This was the year that ChatGPT was first released to the public, and the generative AI boom began.
A key part of that overall investment is Google’s custom silicon chips. Pichai said the company is taking a dual-chip approach with specialized architecture for training and other AI operations.
“We took a fundamentally different approach… [that] This will allow us to build the largest training cluster in the world,” he said.
“This means training larger, more capable models in weeks instead of months.”
