Boz's AI Pep Talk: Can Meta Out-Open OpenAI?

When your CTO has to sit down with a journalist to explain your company's "path to AI relevance," you've already lost the narrative. But that's exactly where Meta finds itself — with Andrew "Boz" Bosworth doing the rounds with Alex Kantrowitz at Big Technology, making the case that the company formerly known as Facebook, then rebranded as a metaverse company, is now, actually, seriously this time, an AI company.

Let's be clear about what's happening. Meta is running a multi-front war and losing on most of them. The metaverse bet has cost Reality Labs north of $50 billion in cumulative operating losses since 2020, with no consumer metaverse to show for it. The Quest 3 launched at $499 in October 2023 — it's a decent headset. "Decent" doesn't justify a $50 billion inferno. And while Mark Zuckerberg was busy renaming his company and populating Horizon Worlds ghost towns, OpenAI, Anthropic, and Google ate the AI lunch, the AI dinner, and the AI midnight snack.

So now Meta is sprinting. Hard. And Boz is the pitchman.

The Llama Play: Commoditize Your Way to Victory

The centerpiece of Meta's AI strategy is Llama — the open-weights model series that's genuinely disrupted the industry. Llama 3.1 405B, dropped July 23, 2024, was the statement piece: 405 billion parameters, competitive with GPT-4o and Claude 3.5 Sonnet on many benchmarks, available for download with minimal restrictions (the 700 million monthly active user threshold is a hilarious tell about who Meta's actually worried about). It was a power move. Zuckerberg essentially declared: we'll commoditize the model layer and win on distribution.

And distribution is where Meta actually has fangs. Meta AI — the assistant shoved into Facebook, Instagram, WhatsApp, and Messenger search bars — reportedly crossed 400 million monthly active users by late 2024. That's a number Sam Altman would sacrifice a board member for. The problem? Engagement depth is a puddle. People don't use Meta AI because it's good; they use it because it's there, occupying real estate in apps they already opened for other reasons. It's the AI equivalent of a gas station hot dog — convenient, not aspirational, and you probably won't remember it tomorrow.

Bosworth's argument to Kantrowitz hinges on this integration advantage. Meta doesn't need the single best model; it needs AI woven into the fabric of daily digital life across billions of touchpoints. Fair thesis. It's basically Google's play with Gemini, except Meta has a social graph that Google would trade its search monopoly for.

The Hardware Wildcard Nobody Saw Coming

Here's where the story gets genuinely interesting. The Ray-Ban Meta smart glasses — launched September 2023 at $299 — have become a quiet sleeper hit. Not because they're revolutionary. Because they're useful for a specific cohort of people who want to record POV video, get AI assistance, and listen to music without staring at a glowing rectangle. The second generation shipped with a better 12MP camera, improved open-ear audio, and Meta AI multimodal features that let the glasses actually see what you're seeing and answer questions about it.

They look like normal Wayfarers. They don't scream "I am wearing a computer on my face" the way Google Glass did back in 2014 when it got you beaten up in San Francisco bars. Meta learned from that failure — make it look like fashion, not firmware.

This matters enormously because if there's a post-smartphone computing paradigm — and that's a planet-sized "if" — Meta wants to own it. Apple's Vision Pro launched at $3,499 in February 2024 and proved that even Apple couldn't make mixed reality mainstream despite industrial design chops and supply chain omnipotence. The mass-market wearable computing device, if it exists, will look more like $299 Ray-Bans than a $3,499 ski goggles simulator. Bosworth knows this. He ran Reality Labs. He oversaw the hardware pivot.

But here's the brutal truth: being the best smart glasses in a market where almost nobody wants smart glasses is a pyrrhic victory. Meta needs a killer AI use case that makes the glasses indispensable. "Hey Meta, what breed is that dog?" is a neat party trick. It's not a reason to wear computers on your face all day.

Open Source: Brilliant Moat or Expensive Charity?

The most fascinating strategic question about Meta's AI play is whether open-sourcing Llama is a genius moat or just corporate philanthropy funded by ad revenue. By releasing weights for free, Meta has seeded an entire ecosystem — Hugging Face is buried in Llama fine-tunes, derivatives, and competitors built on their architecture. This creates dependency. If everyone builds on Llama, Meta sets the agenda.

But it also means competitors can match Meta's model quality by just... downloading Llama. The open-source advantage only works if Meta stays ahead in training the next generation, and that demands compute and talent at a scale very few entities can sustain. Meta's buying hundreds of thousands of Nvidia H100s. They're building a 2-gigawatt data center in Louisiana announced December 2024. This is an arms race funded entirely by advertising cash flow — a war chest no AI pure-play can match.

The question Bosworth should be answering: when does Meta monetize AI directly? OpenAI is reportedly pacing toward $5+ billion in annualized revenue. Anthropic has locked down enterprise contracts across finance and healthcare. Google has Vertex AI and Gemini API revenue. Meta has... engagement metrics? "Time spent in app"? The advertising business will benefit from AI-powered targeting and creative generation, but that's incremental optimization, not a new revenue line.

The Verdict: Coherent But Unproven

Meta's AI strategy is logical but unproven. The distribution advantage is real and massive. The open-source play is strategically sound. The hardware is better than anyone expected. But "path to AI relevance" is a telling, honest phrase — it acknowledges Meta isn't relevant in AI yet, not the way OpenAI or even Anthropic is. Zuckerberg's company missed the first wave of generative AI while building virtual conference rooms nobody visited.

Bosworth is a compelling messenger — he's been inside Meta since 2006, he understands the hardware, and he's refreshingly blunt for a Big Tech executive. The strategy has internal logic. But Meta's path runs through a minefield of competitors who arrived first, built trust, and aren't standing still. For every Instagram-level success in Meta's history, there's a Diem, a Portal, a Facebook Dating, a Horizon Worlds — hyped initiatives that launched to fanfare and fizzled into irrelevance.

Boz can pitch it. Whether Meta can deliver is another question entirely. And the clock is ticking.