Meta (META.US) goes all-in on Muse AI Agent, building full-scene access via VR headsets, smart glasses, and Charm

According to Zhihu Finance APP, if you only look at download data, Muse has already won. But what Zuckerberg wants is clearly more than that. On September 23 local time, Meta (META.US) held its annual Connect developer conference in Menlo Park, California. The signals sent throughout the event were very clear: Muse, the personal AI agent launched only two weeks ago, will become the soul of all Meta’s future hardware. From lightweight VR glasses to smart glasses without cameras, and even a mini device that hangs on a keychain, Muse is embedded in every new product.

This marks a rare synergy between Meta’s product release and market timing in the AI race. Over the past two years, Meta’s situation in AI has been quite awkward—they spent a lot, built plenty of computing power, but always lacked a product that ordinary people could truly use. The appearance of Muse changes that.

100 Grams, $1299: Meta Redefines VR Devices

The most anticipated new hardware at the conference was the Meta VR Glasses. What stands out about this device is not its functionality, but its weight—about 100 grams, only one-sixth that of Apple’s (AAPL.US) Vision Pro and even just one-fifth of Meta’s own Quest 3S.

The breakthrough was achieved thanks to a complete architectural overhaul. Meta moved computing components such as the processor, battery, and cooling fan into a pocket-sized puck, leaving just the display and sensor modules inside the glasses frame. The trade-off for this design is sacrificing some display resolution—about 2.5K per eye, lower than Vision Pro’s 4K—but gaining three hours of battery life, which is higher than Vision Pro, and supporting full-time plug-in use.

Priced at $1299, roughly one-third of the Vision Pro’s $3699. The pricing strategy is clear: Meta is not trying to confront Apple in the high-end market but instead appeals to those interested in VR but unwilling to spend nearly $4,000 just to try it out by positioning at “affordable high-end.”

The device is equipped with a Qualcomm Snapdragon Reality Elite processor, supports eye tracking and gesture control, and users can interact directly with the built-in Muse to complete tasks. The product is expected to hit the market in spring 2027.

Smart Glasses Without a Camera: Meta’s Sincerity or Strategy?

Privacy controversy is a hurdle Meta’s smart glasses cannot avoid. Over the past year, this product has been labeled as “creepy glasses”—banned from courtrooms in New York, restricted at Australian public swimming pools and London pubs, with reports of people using the glasses to record women without consent and post on social media. In March this year, a 230-page class action lawsuit was filed in federal court in California, where more than 70 consumers accused Meta of misleading users on privacy, revealing in the complaint that Meta’s AI annotation staff had viewed private video clips uploaded by users, including changing room footage.

In this context, Meta’s launch of its first camera-free AI glasses, Ray-Ban Meta Audio, is difficult not to interpret as a response. These glasses start at $349, weigh 43 grams, offer up to 12 hours of single-charge battery life, with an additional 48 hours provided by the charging case. The functions focus on audio playback, calls, voice translation, and voice interaction with Muse. Designed by EssilorLuxottica, the product comes in 23 color and lens combinations, with pre-orders opening October 13.

Meanwhile, the camera-equipped Ray-Ban Meta Gen 3 was released simultaneously, starting at $449, featuring a 12-megapixel camera and a 6-microphone array, which Meta says reduces background noise by over 90%.

Meta CTO Andrew Bosworth previously stated publicly that he does not want a few “bad actors” to influence public perception of smart glasses and reiterated that Meta has set up several privacy safeguards, such as an indicator light automatically turning on when recording, and automatically disabling the camera if the indicator is blocked. The launch of the camera-free version can be seen as providing a “trustless” option for consumers—you don’t have to trust the camera, just buy the version without one.

Charm: A Keychain-Sized Entry Point for AI

Toward the end of the conference, Zuckerberg took from his pocket the most talked-about product of the night—Muse Charm.

This is a palm-sized device with about a 2-inch OLED touchscreen, built-in 5G modem, featuring a brand-new operating system specifically designed for Muse. The device includes speakers, microphones, front and rear cameras, and a USB-C port, supports fingerprint unlocking and music/video playback, and can be worn by wristband, lanyard, or just slipped in a pocket.

Charm is not intended to replace a smartphone, but rather serve as a dedicated portal for Muse. By pressing the fingerprint sensor, users can talk directly to the AI without unlocking their phone or opening any app. As Zuckerberg said at the launch: “We’ve put the full Muse experience—including real-time voice and digital avatars—into a device you can hang on your keychain.”

The product was designed by a lab led by former Apple interface design director Alan Dye. Its appearance is reminiscent of the 1990s electronic pet Tamagotchi, with a transparent shell revealing internal circuitry. The screen center displays a customizable virtual Muse avatar with different looks, clothing, and voices. When two Charm users are near, the devices can recognize and interact with each other.

Notably, Meta originally planned to release the device in 2027, but Zuckerberg personally pushed the team to bring the timeline forward to late 2026 to catch the holiday shopping season. However, Meta CTO Bosworth admitted that Charm is more of an “exploration,” a typical “V0” product, and that even in the future, users are still expected to interact with Muse more via phones and smart glasses than through Charm.

It’s worth noting that the track record for independent AI hardware isn’t great—both Humane AI Pin and Rabbit R1 attempted to make AI standalone hardware apart from smartphones but both saw poor market performance. The significance of Charm may not be in sales, but in the signal it sends to developers and the market: Muse’s form factor is not limited to smartphone screens, and Meta is willing to explore any possible hardware carriers for it.

Muse’s Trump Card: From 2.5M Downloads to a Pool of 3.6 Billion Users

The reason why Connect’s hardware offensive works is that Muse itself has already taken off.

Since its launch on September 8, Muse’s download data has consistently exceeded market expectations. According to Sensor Tower, the total number of downloads across all platforms reached about 2.6 million in 13 days, with 642,000 daily active U.S. mobile users, far higher than ChatGPT’s 231,000 in the same period. Meta’s share price jumped 11.4% in a single day, adding nearly $200 billion in market value, on Muse’s 13th day after launch.

But what really excites Wall Street is not the download number per se, but Muse’s distribution logic. Over 95% of Muse users are also Facebook users, and 63% are Instagram users. This means Meta does not need to acquire users from scratch—it has a social empire with 3.6 billion daily active users, and Muse just needs to convert from this existing pool.

The commercialization path is already emerging. Muse currently offers a free version, with a weekly limit of 100 million tokens; paid plans are $20/month (500 million tokens) and $100/month (3 billion tokens). Jefferies analysts estimate that if Muse reaches 1 billion users by the end of 2027 and 3% convert to paid users, annualized revenue could hit $10.8 billion. Evercore’s Mark Mahaney puts it more simply: just 1% of Meta’s 3.6 billion users paying $20 per month would generate about $8 billion in annual revenue.

On the e-commerce front, Meta is also speeding up. Muse has already integrated with Walmart (WMT.US), Best Buy (BEST.US), Gap (GAP.S), Sephora, Wayfair (W.US), and payment via PayPal (PYPL.US), with new Expedia (EXPE.US) travel booking and Instacart grocery shopping functions. Zuckerberg made it clear that Meta’s business model is “to extract a small fee from transactions.” Notably, Amazon (AMZN.US) has blocked Muse from browsing its store, a move that, to some extent, highlights the strategic value of partners like Walmart.

Muse will also support Mac operations, with users able to leave file management, email processing, and calendar scheduling tasks to the AI agent, which continues to execute queued tasks after the user leaves the computer. Bain Consulting predicts that by 2030, the U.S. agent-driven e-commerce market will reach $300 to $500 billion, accounting for 15% to 25% of the entire U.S. e-commerce market. Meta is clearly positioning itself for this market.

Looking at all the products released at Connect, Meta’s strategic logic is clear: Muse is the center, hardware is the portal, and the social empire’s user base is the distribution channel. VR glasses target immersive entertainment scenarios, audio glasses are aimed at daily commutes and light interactions, and Charm serves on-the-go quick access—the three scenarios have their own roles, all sharing the common goal of making Muse the most frequent AI contact point in users’ daily lives.

The road, however, will not be easy. Meta’s capital expenditure guidance for 2026 is as high as $130 to $145 billion, mostly focused on data centers and AI infrastructure. At this scale of investment, the market’s expectations for Muse are no longer just for a user-friendly AI product, but as the growth engine to power Meta for the next decade.

Privacy controversy will not fade soon. But the product cadence and strategic clarity exhibited by Meta at this conference at least give investors a sense of possibility: at this inflection point where AI moves from “can chat” to “can act,” Meta has not fallen behind.