Google Cloud turns Gemini into an MLB home run predictor

Baseball fans have always obsessed over numbers, but catching a home run ball has mostly remained pure luck. Google Cloud wants to change that math.

At the MLB All-Star Game in Atlanta, Google Cloud partnered with Major League Baseball's Statcast to launch a generative AI tool that predicts where potential home runs are most likely to land in the stands.

Calling home runs before the pitch

Traditional baseball stats look backward. Google's activation tries to bring that archive into the live pitch.

The system pulls historical Statcast metrics from across the entire All-Star roster, factoring in classic numbers like batting averages alongside historical spray charts from stadiums across the league. It even incorporates real-time game conditions like temperature and wind direction.

Once the system determines the most probable landing section, the data feeds directly into Gemini 2.5 Pro. The model quickly drafts dozens of tailored message options designed for the jumbotron, scoreboards, and concourse screens at Truist Park.

During a pre-game demo featuring Dodgers star Shohei Ohtani, the model calculated that an Ohtani homer would most likely land around sections 152 to 154. Before any message hits the big screen, a human reviewer signs off on the copy to tweak length or phrasing, keeping a human in the loop.

Turning Atlanta traffic into live baseball stats

Google also took this agentic pipeline outside the stadium during the All-Star festivities, outfitting mobile digital billboards across Atlanta.

Orchestrated through Vertex AI, the trucks combined player metrics with live weather, location, and traffic conditions to spit out context-aware billboards in seconds.

For example, drivers stuck in traffic around Buckhead might glance over and see a billboard reading that Ronald Acuña Jr.'s bat sends balls flying at 121 MPH, more than five times the average driving speed around Atlanta.

Why it matters

For fans in the seats, this flips sterile box scores into immediate anticipation. You don't need to be a sabermetrics nerd to enjoy having an AI tell you to get your glove ready because the next batter tends to pull the ball right into your aisle.

For sports entertainment and live venues, it signals a shift from static, pre-rendered graphics toward real-time, context-aware generative content. Orchestrating complex data pipes down to sub-second copywriting opens up an entirely new layer of in-stadium advertising and spectator engagement.

We think Google picked a smart battleground here. Predicting landing zones is playful and carries zero regulatory risk, making it an ideal showcase for Gemini 2.5 Pro without the danger of costly halluncinations.

That said, baseball physics is chaotic, and human-in-the-loop review still acts as a throttle on pure speed. It remains to be seen whether stadium operators will eventually trust agentic systems enough to take the training wheels off during live broadcasts.