AI was once considered a potential threat to Google's dominance. Instead, Google has positioned itself at the center of the AI ecosystem, not just with its own products but as a key provider of computing power and search data for competing labs.
As I wrote two years ago, Google's so-called innovator's dilemma may actually be an innovator's opportunity. The company is not only moving fast on AI models and search integration, but its infrastructure and data power the next generation of AI products from other labs as well.
The Information now reports that Meta has signed a cloud deal with Google worth more than $10 billion. The six-year contract gives Meta access to servers, storage, networking, and Nvidia GPUs in Google's data centers.
It's one of the biggest deals in Google Cloud's history and stands out given the long-running rivalry between Meta CEO Mark Zuckerberg and Google. Meta also uses infrastructure from Amazon Web Services, Microsoft Azure, Oracle, and CoreWeave, in addition to its own data centers.
OpenAI taps Google for compute and search
OpenAI also relies on Google Cloud for computing power. In addition, The Information reports that OpenAI accesses Google's search data through the third-party service SerpAPI, which scrapes search results and supplies them to customers. OpenAI was publicly listed as a SerpAPI customer until May 2024.
This data helps ChatGPT answer current questions, like news or stock market trends. Other AI companies, including Apple, Perplexity, and Meta, also use SerpAPI. Google doesn't offer an official search API and has little incentive to help ChatGPT improve its search capabilities.
The depth and relevance of a search index directly affect the quality of AI answers. For example, Anthropic uses the Brave index for its Claude model, but results from Brave's API still fall short of what Google and ChatGPT can deliver.
OpenAI is working on building its own search index, similar to Perplexity. For now, ChatGPT pulls from a mix of sources—including Microsoft, Bing, and Google—to deliver search results. Its offering is strong, especially with advanced agentic reasoning models like o3 that are trained on web navigation data, but it still relies on infrastructure and data from other major platforms.
