Since it was introduced late last year, OpenAI's ChatGPT and the GPT family of large language models that power it have been unbeatable. GPT-4, the most recent LLM from OpenAI, is largely considered the most capable generative AI model available.
Alphabet's (GOOG 9.96%) (GOOGL 10.22%) Google offers AI models of its own, including the PaLM family of large language models. PaLM 2 was announced in May, and it's used internally by Google in various applications in addition to being available to Google Cloud customers.
PaLM 2 is less powerful than GPT-4, but Google's next-generation AI model may match or even beat OpenAI's flagship model. Google's Gemini, announced on Wednesday, is the company's most powerful AI model to date. Built from the ground up to understand text, images, audio, and code, Google claims that Gemini beats GPT-4 across a wide range of AI benchmarks.
A massive opportunity
While estimates vary, Bloomberg Intelligence predicts that the generative AI market will balloon to $1.3 trillion over the next decade, up from just $40 billion in 2022. While hardware and infrastructure will account for the bulk of this total, generative AI software is expected to generate nearly $280 billion of revenue by 2032.
An industry-leading AI model benefits Google in a few ways. First, Gemini can drive growth for Google's cloud computing business. Gemini Pro, the second most capable Gemini model, will be available on Google Cloud starting Dec. 13. The more powerful Gemini Ultra will become available after Google completes additional safety checks and fine-tuning.
Second, Gemini will power and improve Google's own applications and services. The company's Bard chat-based AI tool is now using a fine-tuned version of Gemini Pro, and Google plans to bring Gemini to Search, Ads, Chrome, and other products over time. AI represents an existential threat to Google's core search business, so the company needs to make sure that it's leading the way in integrating AI into the search experience.
Third, Google's Android and Pixel smartphone businesses will benefit from Gemini Nano, the smallest of the three Gemini models. Gemini Nano will run directly on smartphones running Android 14, enabling Android developers to integrate AI into their applications. Google's Pixel 8 Pro devices have already been updated to use Gemini Nano for on-device generative AI features.
Google trained the Gemini models on its home-grown Tensor Processing Units, rather than Nvidia GPUs that are widely used across the industry. Given how scarce and expensive Nvidia's most powerful data center GPUs have become as the AI frenzy picks up steam, Google's use of custom AI chips could give it an important cost advantage as it trains increasingly capable generative AI models.
The battle for AI supremacy
With one fell swoop, Google has largely laid to rest any concerns that it was hopelessly behind in the generative AI race. Microsoft-backed OpenAI is an early leader, but Google's Gemini has the potential to supercharge the company's AI business, especially once Gemini Ultra becomes available early next year.
There are still some thorny questions for Google related to whether and how AI changes the economics of the core advertising and search businesses. Running large AI models is not cheap, and the company has little choice but to embrace AI within its search business. Custom AI chips will help the cost equation, but only so much.
Alphabet stock jumped on Thursday following the Gemini reveal, and for good reason. The company is set for a fierce battle with OpenAI and other AI competitors, but it's now in a strong position as demand for generative AI explodes in the years ahead.
