Sonos wants “Sonos 27mcp” to sound as boring as it needs to be: a protocol that lets ChatGPT or Gemini reach into your speaker system and actually do something. Sonos announced it at a New York event this week alongside the Ace Ultra headphones and Beam Ultra soundbar, and it’s the company’s biggest announcement in two years. It’s also not really a story about headphones.

The hardware itself is solid, by the numbers Sonos gave out. Ace Ultra ships September 29 at $449, the same price the original Ace launched at back in 2024, with 35 hours of battery life with ANC on (five hours more than before), a three-minute fast charge good for another three hours, and a new Wi-Fi-based “Headphone Engine 2” that lets the headphones talk directly to the home network instead of routing everything through Bluetooth, which Sonos says enables uncompressed, zero-latency spatial audio. Beam Ultra is the bigger jump: a full redesign into a 7.1.2 soundbar with nine drivers, including two up-firing ones, the first time the compact Beam line has done true overhead Atmos. It costs $699, a real $200 over the outgoing Beam Gen 2, and ships the same day as the headphones.

None of that is the interesting part. The interesting part is that Sonos built its own in-house voice assistant, called Sonos 27voice, and then immediately hedged that bet by also shipping Sonos 27mcp, which lets you wire a completely different AI model, ChatGPT or Gemini specifically, into the same system. Eventually you’ll be able to spin up ten custom agents, each with its own voice and its own assigned role: one whispers the morning news, another does the motivational-coach bit Sonos demoed on stage. That part doesn’t land until sometime in 2027, which is company-speak for “we’re not sure yet, but we wanted you to know it’s coming.”

I keep coming back to what’s missing from that list. Google Assistant, the actual first-party voice product from one of the two companies Sonos just named as its launch AI partners, isn’t in Sonos 27mcp. At this stage of a company’s software roadmap, that’s not an oversight; it’s a decision, and it tells you Sonos is building for generative chat assistants specifically, not the smart-home command assistants that already exist in most living rooms. It’s a bet that “ChatGPT in your speaker” is worth more to Sonos customers than “Alexa in your speaker” ever was, which is a real claim about where the category is heading, not just a feature checkbox.

It’s a different pitch than the one Nothing is running in the same broad market. Nothing treats the app as the real product and the hardware as close to disposable, cheap enough that its own $99 Ear (a) buds undercut the company’s own flagship, with most of the actual feature differentiation living inside the Nothing X app rather than the silicon. Sonos is doing the opposite: keep the hardware expensive, and now bolt an AI layer on top that has to justify a $250 premium over the soundbar it replaces. Whether that premium holds is a separate bet from whether Nothing’s own 2027 hardware roadmap holds, but both companies are making the same underlying wager that software, not drivers, is where the next round of audio competition actually gets fought.

It’s also a strange bet for this particular company to be making right now. Sonos spent most of 2024 and 2025 clawing back trust after the app relaunch that became the industry’s go-to cautionary tale: a rewrite shipped so broken the company had to publicly apologize and drip-feed feature restorations for over a year. Handing an early-access AI protocol layer to that same customer base, a customer base that’s still a little gun-shy about Sonos software updates breaking things, is either exactly the redemption arc Sonos needs or the setup for round two of the same story. Early access for Sonos 27mcp lands September 8th. I’d give it a few weeks past that before deciding which one this is.

I won’t pretend this is really an audio story anymore. Sonos used to compete on driver quality and room correction. Now it’s competing on whether third-party language models can reliably control a home theater system without doing something embarrassing on stage at the next launch event, which is the same fight already playing out across the AI infrastructure stack everyone else is building. Sonos just decided it wants that fight in its speakers too.

Sources