Meta 的 AI 资本开支是合理的：AI 基础设施收入应快速增长，而资本开支上升更慢，因此自由现金流应在第三季度 349 亿美元资本开支使其转为负 20 亿美元后重回正值——尽管营收增长了 26%。看空者只看到当下的负自由现金流；预计 AI 成本效率提升 10 倍应会扩大使用量，而非缩减算力支出。

Meta after such a strong earnings dropped about 7%.

What are the 10x engineers at Meta doing? Now, Meta did report and they talk about compute intensive products and capacity constraints and we'll talk a little bit about that later on throughout the episode.

We looked at Meta where it's using AI for actual monetization.

Now that we understand this case and see some stuff Nvidia is doing, I want to jump into Meta because Meta gives us a concrete example why both sides of the AI bull bear debate can be reasonable.

Now revenue was for the most recent earnings 64.6 billions in the third quarter up 26% year-over-year. Operating cash flow was about $ 32.7 billion about 50% of operating cash flow.

Unfortunately, capital expenditure was about 34.9. That left free cash flow at a negative of $2 billion. The business is generating substantial cash, but the investment program was larger than the quarterly operating cash flow.

In my opinion, I believe the answer is yes. I actually believe it's going to be a bit different. I think revenue from these AI infrastructures that are going to be building is going to increase at rapid levels and you're going to see capix go up but not go up at the same rate.

So there is going to be a point where we're still spending insane in capital expenditure, but all the investments that you've made on these GPUs that are still very effective. We see H100 still do extremely well.

We see A100 still being high demand. I'm sure Meta Platform is using them for their products. You're going to see all those still make massive revenue after you're done paying them.

And you're going to be collecting great revenue, great cash flow that that free cash flow is going to go back to positive.

Management in their most recent earnings reported more than 10 million Muse agent setups in the United States and Canada over roughly 6 weeks. It described compute intensity as one factor limiting a broader rollout alongside local product work that gives us product level reason to want more capacity.

Meta has talked a little bit in their most recent earnings of some of the ways they're going to be building revenue off Muse, right? They have the consumption based tier levels.

They have API pricing. Uh they have fee taking on any form of e-commerce atic use cases.

The bull case is Meta has told us where they're having it. The bare case is, yeah, you said it, but we don't see it just yet. What we do see is that negative free cash flow.

Now, Meta did mention that they expect more than 10-fold improvements in cost efficiency in AI next year. The reason that expectation matters to me is wait bears are going to say look 10 times more better more tokens next year in theory that's going to reduce demand dramatically right if you're using if you're able to get 10 times more token I find it the other way the reason that expectations matter to me is that cheaper useful AI can open up more task more ages and more usage a lower cost per task does not automatically tell us what happens to compute bill It depends on how much usage expands and that's the thing right AI needs to continue to grow and that is one of the things that I be talking I keep talking about in my bullish thesis for it to continue AI needs to continue to innovate this is consistent with how I think about small models and agent workloads the useful test is whether customers get enough value to keep paying and whether the product economics improve as usage grow efficiency and growing hardware demand can coexist is but that is a business outcome to watch rather than uncertainty right now.