Does the speech recognition computation happen in the browser code, or do they send the voice bits to their Google servers which translates it into text and the

Does the speech recognition computation happen in the browser code, or do they send the voice bits to their Google servers which translates it into text and then send it back to the browser? The latter is how voice recognition works on Android, and why it only works when you have an active internet connection. It seems a little weird to me to build an html5 standard that requires server side computation. How is a non-profit open source browser going to fund the massive server load needed, not to mention the R&D needed to develop a voice-to-text translator? Google’s stuff is proprietary.