> any that want to do Alexa style voice prompts need it

Not technically true. More modern smart TVs (~post-2021) put speech recognition models for the hotword only onto the DSP, and then have the DSP wake up the CPU for further auditory processing when it detects a hotword or remote button press. There are major power-efficiency benefits to this, because the DSP uses a fraction of a watt and can easily keep running while the TV is sleeping, while the CPU, Internet, and rest of the circuitry is a major power hog. Your smart TV or streaming dongle would basically never sleep if it were forever running the audio capture.

For that matter, it's pretty rare for the device to actually run the speech recognition models on-device. Smart TV manufacturers are very sensitive to BOM costs, which means that their hardware is only powerful enough to run machine-learning models at the extreme high end of their product range. Most of the time the full audio gets streamed back to the Internet and the speech recognition & voice assistant gets run in the cloud, with commands streamed back to the TV.

I won't say that it never has always-on audio listening; a significant number of devices I tested did. But I'll say this comes because they're poorly implemented. The DSP hotwording approach requires coordination between hardware and software developers, oftentimes across companies, and it's more complex to implement and a lot of executives don't care about the marginal consumer benefit it gives. Many smart TVs just guzzle power when they're off instead.