Meta announced this week it's cutting 5% of its lowest performers - roughly 3,600 people - then backfilling those roles with new hires. Add the standard attrition Meta expects, and you're looking at around 10% of the workforce gone by year's end. That's 21,000 people.
The timing is what makes this interesting. Meta spent $65 billion on AI infrastructure in 2024. Data centres, chips, the works. Then they built Llama models that anyone can use for free. No API fees. No revenue model beyond "it makes our other products better, probably."
Now they're restructuring. Mark Zuckerberg's memo talks about raising the bar, increasing the ratio of engineers to other roles, and removing layers of management. The language is careful - this isn't a panic move, it's about "excellence" and "getting leaner." But the subtext is harder to ignore: we spent billions, we're not sure what we're getting back, and we need to tighten up.
The Build-First, Figure-It-Out-Later Bet
Meta isn't alone in this. Every major tech company is pouring money into AI infrastructure right now. The logic goes: build the capability first, find the applications later. It's not a terrible strategy when you're sitting on massive cash reserves and the technology genuinely feels like it could unlock new markets.
But there's a catch. Building AI infrastructure is expensive in ways that are hard to communicate to shareholders. You can't point to a new product line and say "this will generate X revenue." You're betting that having the best models and the fastest inference will matter in three years. Maybe it will. Maybe it won't.
What makes Meta's situation particularly tricky is that Llama is open source. They're not selling access. They're not charging per token. The bet is entirely indirect: better AI makes Instagram's recommendations smarter, makes WhatsApp more useful, makes their ad targeting more effective. It's a reasonable bet. It's also one that's hard to prove in a quarterly earnings call.
What This Means for Builders
If you're building on top of Meta's infrastructure - using Llama models, integrating with their APIs - this isn't necessarily bad news. The models themselves aren't going anywhere. Meta's too deep into this to pull back on the core AI work. But it does signal something about the broader environment.
The era of "spend now, figure out revenue later" is tightening up. Even companies with Meta's resources are starting to ask harder questions about return on investment. That means if you're pitching AI features to clients or stakeholders, you need a clearer story about value. "It uses AI" isn't enough anymore. "It saves us 10 hours a week on customer support" is a different conversation.
For developers and small business owners, there's a practical angle here too. Meta's cost-cutting doesn't hurt you directly - their models are still free, their infrastructure is still accessible. But it's a reminder that even the biggest players are navigating uncertainty right now. The companies that win in this environment are the ones building tools that solve specific, measurable problems. Not the ones building "AI platforms" with no clear use case.
The Bigger Pattern
This isn't the first time Meta has done large-scale layoffs. They cut 11,000 people in 2022, then another 10,000 in 2023. Zuckerberg called 2023 the "year of efficiency." Now we're back here again, and the efficiency language is returning. That suggests the AI spending spree didn't deliver the clarity Meta was hoping for.
It also suggests something about the state of the AI market more broadly. We're past the initial hype phase where just having a model was impressive. Now the question is: what can you actually do with it that makes money or saves money? Meta is asking that question internally. Every business should be asking it too.
The infrastructure buildout will continue - too much momentum, too much capital already deployed. But the companies that survive the next phase will be the ones that can draw a straight line from their AI investment to a real outcome. Meta is making that bet. The workforce cuts are the cost of recalibrating.