Remember when Meta was the good guy in artificial intelligence? The company that gave away powerful models for free while everyone else guarded their tech like gold? Yeah, those days feel pretty distant right now.
Something big is shifting inside one of the world’s most powerful tech companies, and it’s creating waves that go far beyond Menlo Park. What started as an almost idealistic push toward open-source AI has morphed into a frantic, billion-dollar chase that’s leaving employees confused, investors nervous, and competitors watching with interest.
The Quiet Death of the Open-Source Dream
It wasn’t that long ago – honestly, just last year – that Mark Zuckerberg was proudly declaring that his company’s Llama models would be the most advanced in the world while remaining completely open. The vision was clear: democratize AI, accelerate innovation, and build goodwill in a community that had grown skeptical of Big Tech’s motives.
That message resonated. Developers loved having access to state-of-the-art models without paying massive licensing fees. Startups built entire businesses on top of Llama. Researchers pushed boundaries knowing their work would benefit everyone.
But something changed in 2025, and it changed fast.
The Avocado Revelation
Behind closed doors, Meta has been developing a completely new frontier model – internally codenamed Avocado. And here’s the part that would have been unthinkable twelve months ago: many people familiar with the project expect this one won’t be open-source when it finally launches.
Think about that for a second. The company that built its entire AI reputation on openness is now seriously considering going fully proprietary. It’s not just a technical decision – it’s a philosophical U-turn that speaks volumes about how brutally competitive this space has become.
The gap between open and closed models that everyone was talking about last year? It didn’t close the way we hoped. In some cases, it actually got wider.
Sources inside the company say Avocado was originally targeted for release before the end of 2025. That timeline has slipped – now they’re looking at early 2026 – as teams wrestle with performance issues that need solving before this model sees the light of day.
The $14.3 Billion Wake-Up Call
The turning point came in June when Meta made a move that shocked the entire industry: they spent $14.3 billion to bring in Alexandr Wang, the young founder of Scale AI, along with a significant chunk of his top talent.
This wasn’t just hiring. This was a statement.
Suddenly, Meta wasn’t trying to win with philosophy anymore. They were playing the same game as everyone else – throwing enormous amounts of money at the best people, building elite teams, and accepting that winning in AI might require the kind of secrecy and control that open-source fundamentally rejects.
- Wang became Chief AI Officer
- He now runs an elite unit called TBD Lab (literally located near Zuckerberg’s office)
- Former GitHub CEO Nat Friedman came in to lead product and applied research
- Even one of ChatGPT’s co-creators joined the ranks
These aren’t incremental hires. This is Meta admitting they needed a complete overhaul of how they build AI.
Culture Clash: The Human Cost
Perhaps the most fascinating part of this story isn’t the technology itself – it’s what happens when you drop Silicon Valley’s most intense startup operators into a company that spent two decades building its own very specific culture.
The differences are stark. Where Meta traditionally valued broad collaboration and internal transparency (think Workplace posts, open feedback, cross-team coordination), the new guard operates more like the frontier labs they’re competing against.
People familiar with TBD Lab say the team doesn’t even use Workplace – Meta’s internal communication platform. They operate more like an independent startup embedded within a much larger company. Communication is tighter, more controlled, more focused on shipping than sharing.
They have this new mantra: “Demo, don’t memo.” It’s not about writing long documents anymore – it’s about building working prototypes fast.
This approach has delivered results in other companies, no question. But it’s also created friction with teams that built Instagram, Facebook, and WhatsApp using very different processes.
The 70-Hour Weeks and Growing Pressure
The intensity is real. Multiple sources describe 70-hour workweeks becoming the norm in AI organizations. There have been layoffs, restructurings, and significant turnover – including some very high-profile departures.
When you’re spending this kind of money and bringing in this level of talent, the pressure to deliver becomes enormous. Every competitor release – Google’s latest Gemini, OpenAI’s GPT updates, Anthropic’s new models – lands like a gut punch when you’re still working on getting your own house in order.
And the financial markets are watching closely. Meta’s stock has underperformed broader tech indexes this year, and analysts are increasingly asking the same question: when do we see return on all this investment?
The Infrastructure Arms Race
Building frontier models isn’t cheap, and Meta is spending accordingly. The company recently increased its 2025 capital expenditure guidance to $70-72 billion. They’re building massive data centers, signing deals with cloud providers, even entering joint ventures to construct facilities that can handle the computational load required.
It’s worth remembering that every major player is doing this. The difference is that Meta came into 2025 thinking their open-source advantage might reduce some of these costs. The community would help improve models. Innovation would be distributed.
That hasn’t happened at the scale they needed, and now they’re making up for lost time in the most expensive way possible.
What This Means for the Industry
In many ways, Meta’s journey tells the broader story of AI in 2025. The early optimism about open-source closing the gap with proprietary models has given way to a more pragmatic reality: at the very frontier, the advantages of control, coordination, and massive concentrated investment are proving difficult to overcome.
We’ve seen similar shifts before. Companies often start with idealistic approaches when they’re playing catch-up, then pivot to more ruthless efficiency when they decide they want to lead. The question is whether Meta is making this pivot from a position of strength or desperation.
The talent they’ve brought in is undeniably world-class. The resources they’re committing are staggering. But building great models requires more than money and smart people – it requires the kind of focused, coordinated effort that large companies often struggle to maintain.
Early 2026 will be fascinating. When Avocado finally ships – whether open or closed – it will tell us a lot about where Meta sees itself in the AI landscape. More importantly, it will tell us whether the open-source dream can still compete at the highest levels, or whether we’ve entered a new phase where only the most disciplined, best-funded organizations can play.
Either way, the era of Meta as the plucky open-source alternative appears to be over. What comes next could define not just their future, but the future of AI development itself.
The most honest assessment might be this: every company that wants to matter in AI is going through some version of this transformation right now. Meta is just doing it more publicly, more dramatically, and with higher stakes than most.
In a world where artificial intelligence increasingly defines technological leadership, there may simply be no room left for idealism. The question isn’t whether companies can afford to be principled – it’s whether they can afford not to be ruthless.
For now, thousands of employees are working through weekends, new leaders are rewriting decades of process, and billions of dollars are being spent on a bet that might determine whether Meta leads the next era of technology or becomes a cautionary tale about what happens when you bring a philosophy to a money fight.
Whatever happens with Avocado, one thing is clear: the AI wars have entered a new and far less forgiving phase.
