AI Glasses Will Stay a Phone Accessory Through 2028

We expect AI glasses to remain a large phone accessory through 2028.

Key points

• We put about 75% odds on every best-selling AI glasses model still depending on a paired phone at the end of 2028.
• We expect no glasses that run without a phone to sell 1 million units in any single year before then.
• Camera law will likely limit where glasses can go faster than optics or batteries improve.
• We put about 60% odds on a US state requiring recording indicators that wearers cannot disable by the end of 2028.
• The companies that own phones and assistants will likely control the app layer on glasses too.

Our forecast

By the end of 2028, every best-selling AI glasses model will still rely on a paired phone for heavy AI work. No glasses that run without a phone will sell 1 million units in a single year. We put this call at about 75%.

A second call follows from the first. At least one US state will require recording indicators that wearers cannot disable by the end of 2028. We put that at about 60%.

Why we think so

Our signal scan found a category growing fast as cameras and microphones. Shipments rose 263% in the first half of 2026¹. Display-less models made up 96% of that volume¹. Growth at Meta's eyewear partner then fell to 20% in the third quarter². That curve fits an accessory reaching its mass market.

Display glasses are the path to replacing the phone, and they remain far from it. Meta raised Ray-Ban Display component orders to just 150,000 units³. Reviewers report about 3.5 to 7 hours of battery, depending on display use³. TrendForce forecasts 950,000 AR glasses worldwide in 2026³. Standalone sales of 1 million a year would need better waveguide supply and all-day battery within three years.

Tethering is a strategy as much as a limit. Samsung's Gemini glasses pass heavier AI work to a paired Galaxy phone⁴. Apple has reportedly moved its headset engineers to glasses⁵. Meta's WebMCP preview places Meta AI between the wearer and every app⁶. Each company gains by tying the face to the platform it already runs. Our open questions piece named independent hardware share as a test. No maker has passed it yet.

Agents pull the same way. MCP SDKs see close to 500 million downloads a month under neutral governance⁷. Gartner expects about 40% of enterprise applications to embed task agents by the end of 2026⁸. Glasses are becoming a voice front end for work that runs on the phone and in the cloud. That role needs a microphone and a camera. It does not need glasses to replace the phone.

Privacy sets the ceiling. The courts of England and Wales and New York's court system bar camera glasses⁹. Lawmakers in more than a dozen US states plan wearable bills for 2027⁹. A Pennsylvania proposal would require recording indicators that wearers cannot disable⁹. Camera-based AI queries can already analyze a scene without lighting the LED¹⁰. Meta's reported Name Tag plan has drawn an EPIC complaint, questions from senators, and a proposed class action¹¹. Our implications piece traced how these rules follow the camera. A device people must remove at the door cannot take over the phone's role.

How this call fails

It is 2029 and we were wrong. Agents are the likeliest cause. METR finds agent task length doubling every four to seven months¹². At that pace a spoken request could hand off hours of work with no screen involved. Glasses could then absorb most phone use while still pairing with a phone. Our measure would hold and our meaning would fail.

China is the second likely cause. A 15% trade-in subsidy and Qwen-linked glasses could push a standalone model past 1 million units in the domestic market¹³. Apple is the third. A first Apple glasses product with an all-day display would reset what buyers expect⁵.

Who loses

• Headset makers lose investment as Apple and Meta shift to glasses⁵.
• Developers who build for one store lose leverage while Meta's review rules stay unwritten⁶.
• Blind and low-vision users lose access where venues ban every camera⁹¹⁴.
• Bystanders in states without consent law lose practical anonymity if face recognition ships¹¹¹⁰.

First moves

• Design for audio and the assistant first, with the display as an optional layer¹⁴.
• Expose app functions through MCP to reach Meta, Android XR, and later platforms with one integration⁶⁷.
• Write a camera-glasses policy for venues and worksites now, with an exemption for assistive devices⁹¹⁴¹⁵.
• Track the 2027 state bills and plan for consent rules that differ by state⁹.
• Expect app store fees and review rules on glasses to mirror the phone stores⁶⁵.

What would change our mind

• A display model with reviewed all-day battery and no phone in the loop passes 1 million orders³.
• A Chinese standalone model sells more than 1 million units in a year under the trade-in subsidy¹³.
• Meta ships Name Tag and no state passes a consent or indicator law by 2028¹¹⁹.
• Survey data show glasses owners cutting phone screen time by a quarter or more.