Meta Just Quietly Rewrote Its Whole AI Bet at Connect 2026
In this article
Bottom line: At Connect 2026, Meta stopped selling the metaverse and started selling smart glasses as the delivery vehicle for its AI models — a reversal from the VR-first pitch Zuckerberg gave in 2021.
The company confirmed Reality Labs' VR/MR headset roadmap is being deprioritized in favor of Ray-Ban and Oakley smart glasses paired with Meta Superintelligence Labs' models, and detailed a multi-year capex commitment well above $100 billion for 2027 aimed almost entirely at inference and training infrastructure.
If you build on Llama, expect the next major release to ship glasses-first APIs before it ships anything for the Quest line.
The pivot has been building since Meta Superintelligence Labs formed in mid-2025 — Connect just made it official.
I watched the keynote live, and about twenty minutes in I texted a friend who still owns a Quest 3: "I think they just told you your headset was the trial run."
That's not an exaggeration.
Zuckerberg spent roughly four minutes on mixed reality headsets and forty on smart glasses and AI models. For a conference literally named after the metaverse-era rebrand, that ratio tells you everything about where the money and the roadmap are actually going in 2027.
The Setup: From Metaverse to Superintelligence
Back in October 2021, Meta changed its name and told investors the future was fully immersive virtual worlds. Billions went into Reality Labs.
Billions more went into headsets that sold okay but never became the "next iPhone" moment Zuckerberg promised.
Then in mid-2025, something shifted.
Meta stood up Meta Superintelligence Labs, poached researchers from OpenAI and Google DeepMind with nine-figure offers, and folded most of its existing AI research org underneath the new group.
Reality Labs kept building headsets, but the public narrative quietly moved: less "step into the metaverse," more "talk to an AI that can see what you see."
By the time Connect 2026 rolled around this month, that shift had a name, a budget, and a hardware form factor. Meta didn't announce a new flagship VR headset this year.
It announced three new smart glasses SKUs — a sports variant, a display-equipped upgrade to the existing Ray-Ban line, and a business-focused pair built with Oakley — all running on a new on-device/cloud hybrid model Meta is calling part of its Llama "Frontier" tier.
I've been covering AI infrastructure since before most people could spell "transformer," and this is the cleanest example I've seen of a company admitting its original bet needed a new vehicle without admitting the bet itself was wrong.
The Core Insight: Glasses Are the Model's New Runtime
Why Hardware Strategy Is Model Strategy Now
Here's the part that matters if you're a developer and not just a Meta shareholder: the smart glasses aren't a gadget announcement, they're a distribution decision for the model.
A phone app competes with ChatGPT 5 and Gemini 2.5 on equal footing — same form factor, same friction, same ten other apps open in the background.
Glasses with a camera and a mic sitting on your face are a different surface entirely. They see what you see, hear what you hear, and respond without you pulling out a device.
Meta's bet is that ambient, always-available AI beats app-based AI, and that whoever owns the glasses owns the default assistant.
That's a real strategic insight, and it's not crazy.
Ray-Ban Meta glasses sales have been growing fast enough that Meta now says they're outpacing sales of its most recent VR headset, the Quest 3, over a comparable stretch.
Zuckerberg said onstage that the company now considers glasses "the primary form factor for personal AI," which is about as close to a strategy memo as a company reads aloud on stage.
The Infrastructure Number Nobody's Talking About Enough
The bigger story, buried under the glasses hype, was the capex commitment.
Meta confirmed spending guidance that puts 2027 infrastructure investment — data centers, custom silicon (the next-gen MTIA chips), and power contracts — well above $100 billion, with a chunk earmarked specifically for inference capacity rather than training.
That's a meaningful signal.
Training-heavy capex spending says "we're racing to build a bigger model." Inference-heavy capex spending says "we expect hundreds of millions of devices making real-time requests to that model, constantly." Glasses that listen and respond in real time are an inference-hungry product in a way a chatbot in a browser tab never was.
The infrastructure spend and the hardware pivot are the same decision, told twice.
What Changed for Llama
Meta also confirmed the next major Llama release — expected to ship models across a "Frontier" and "Edge" split — will prioritize low-latency, on-glasses inference for the Edge tier before opening up broader API access for the Frontier tier.
Translation: if you're building on Llama today, the next release is optimized for a form factor most of you aren't building for yet, and the flagship model access you actually want will trail behind it.
That's a meaningfully different roadmap than what Meta AI's competitors are doing. Anthropic's Claude 4.6 and OpenAI's ChatGPT 5 are both still fighting the "best model in a text box" war.
Meta just told you it's not fighting that war anymore — it's fighting the "best model on your face" war, which is a genuinely different product problem with different latency, privacy, and battery constraints.
The Reality Check: Glasses Still Have a Battery Problem
I want to be the guy who says this pivot is obviously brilliant. It isn't, not yet.
Smart glasses with always-on cameras and a persistent AI listening for context burn through battery in a way that phones simply don't have to solve as urgently — you can carry a battery pack for your phone, you can't strap one to your temple.
Meta's own demo units at Connect were reportedly swapped out twice during a 90-minute session backstage, according to a couple of attendees I talked to who weren't supposed to mention that.
Nobody official confirmed it, but it tracks with what anyone who's used the current Ray-Ban Meta glasses for a full day already knows: real-time AI is a battery hog.
There's also the privacy problem, which Meta addressed with exactly the amount of seriousness you'd expect from a company whose last decade includes multiple FTC settlements over data handling.
A pair of glasses that can identify a stranger's face and pull up context about them in real time is either an incredible accessibility tool or a surveillance device, and which one it is depends entirely on defaults most companies get wrong on day one.
And Reality Labs isn't dead — it's just no longer the star of the show.
Employees I've heard from indirectly describe a group that knows it's now the "legacy" division inside its own company, which is a rough place to do your best work from.
The Practical Takeaway: What Developers Should Actually Do
If you build anything on Llama or Meta's AI stack, don't wait for the Frontier tier API access to catch up to the marketing.
Start prototyping against the Edge tier constraints now — low latency, small context windows, camera-plus-audio multimodal input — because that's the surface Meta is optimizing for, and it'll shape what the flagship models eventually inherit.
If you're deciding between platforms for a wearable or ambient AI product, Meta just told you its infrastructure bet, which means its pricing and rate limits over the next 12–18 months will likely favor glasses-native use cases over generic API traffic.
Plan accordingly, and don't assume parity with ChatGPT 5 or Gemini 2.5 API behavior — Meta's optimizing for a different latency profile entirely.
And if you're a product person watching from the outside: the lesson here isn't "AI glasses are the future." It's that a company will happily abandon a five-year, tens-of-billions-of-dollars strategy the moment the data says a cheaper, dumber-looking product is actually winning. Ray-Ban Meta glasses were the unglamorous product next to Quest headsets for years.
They just became the whole company's north star.
Is Meta actually right that glasses beat phones as the default AI interface, or is this just the metaverse pivot wearing a new pair of frames?
I've got a Ray-Ban Meta pair on my desk right now and I'm honestly not sure yet — what's your read?
