Google just proved that artificial intelligence can translate speech entirely offline on $80 hardware — and the fact that they're calling it a "technology showcase" rather than a product tells you everything about how Big Tech really makes its money.

The Gemma Translator, built with Google's Antigravity team, runs on a Raspberry Pi 5 using the company's lightweight Gemma 4 E2B model. No cloud connection. No data flowing back to Mountain View. No internet required. It's open source on GitHub, the model is free under Apache 2.0, and the whole rig costs under $100 to assemble. And yet Google has no plans to sell it.

Both TNW and Digital Trends covered this as a feel-good story about what's technically "possible" with efficient AI. Neither asked the question that matters: if this works, why isn't Google shipping it?

The answer is simple. Google Translate, like most Google products, exists to collect data. Every translation query, every voice clip, every fragment of conversation that hits Google's servers is grist for the advertising machine that funds the entire enterprise. An offline translator that never sends a packet to the cloud is, from Google's perspective, a translator that doesn't pay rent.

TNW noted that the project is "a technology showcase, not a product announcement" and that "Google has not said it plans to sell the device or integrate it into an existing product." Digital Trends echoed the same framing almost verbatim, calling it "more of a technology showcase than a product announcement." Both outlets presented this as a neutral fact. It isn't. It's the whole story.

The components are all publicly available. TNW reported that the Raspberry Pi 5 costs about $80, the Gemma 4 E2B model is free under Apache 2.0, and the 3D-printed case and code are open source. The prototype includes a microphone and speaker inside a custom enclosure, processing speech locally with no external dependency. TNW correctly identified who this actually serves: travelers in areas with poor connectivity, organizations handling sensitive conversations, and emergency responders operating without infrastructure.

Digital Trends added that on-device processing could reduce latency, making conversations feel more natural, and listed the same potential applications — travel gadgets, accessibility tools, classrooms, disaster response kits. Both outlets painted a picture of a future where AI assistants don't need data centers. Neither outlet connected that future to the present business model that makes it commercially impossible for Google to pursue.

Google launched Gemma 4 in April with four model sizes. The E2B edge variant was specifically designed to run on hardware as small as a Raspberry Pi. This translator is the proof of concept. The technology is here, today, and it's cheap.

Here's what makes this revealing. Google didn't build this to sell to you. They built it to demonstrate that they could — and then walked away. The open-source release means anyone with $80 and a 3D printer can build their own. But Google itself won't put this on a shelf, because a device that processes your speech locally is a device that doesn't feed the data engine. The same company that shows off privacy-preserving AI in a lab runs its actual products on the opposite principle: vacuum everything up, send it to the cloud, monetize it.

The establishment tech press covered this like a novelty item. "Some of AI's most interesting innovations are being squeezed into tiny devices," Digital Trends wrote. That's the kind of breathless framing that lets Google off the hook. The interesting story isn't that the technology exists. It's that the company proving it works has no intention of letting you use it — because your data is worth more to them than your privacy.

The next time Google tells you it needs your data to make its products work, remember the Gemma Translator. They proved the alternative exists. They just won't sell it to you.