Meta: The Next Frontier.
Re-Deep-Dive
This is a re-deep-dive of my original Meta deep dive, written in late November 2022, in which I explain how Meta was a highly asymmetric opportunity at the time.
The stock is up 366% since. This deep dive focuses on what’s next for the company…
Ad // Since 2014, I’ve gotten 5 out of 8 picks right.
The performance of my picks thus far since inception:
1. AMD : ~38X 2. Tesla: ~10X 3. Spotify and Palantir: ~3X 4. Hims: 33%+ ***
My 2 Hour Deep Diver course has been tried and tested by over 60 students to date. It teaches you all the mental models that enabled me to make the above investments in just two hours (although it is highly dense content and it will take you time to integrate it).
I’m selling 10 units of my course for just $150 (instead of $199), so now is a great time to buy. They sell out quick.
*** I liquidated two mistakes (GPRO and BB, both down around 60%) and invested the resulting money into Hims. The course will also help you avoid my past mistakes.
Edited by Brian Birnbaum.
1.0 Reels Are Just the Start
Despite the initial criticism, Reels is now Meta’s growth engine. The infrastructure Meta has built to power Reels is the foundation for Meta’s next frontier, which is artificial intelligence as a service.
With the successful deployment of Reels, Meta has solidified its track record of scaling and monetizing new surfaces. When I wrote the original Meta deep dive, Reels was already scaling well, having gone from zero to 140B daily impressions in just two years.
Reels now represents over 50% of all time spent on Instagram.
AI recommended content, Reels’ underlying engine, now accounts for over 50% of all content consumed on Meta, up from zero just a few years ago.
Meta is again on the rise and monthly active user growth has reaccelerated. More than 3.2B people use the platform every day now, with Meta’s daily active users thus accounting for 39.5% of the world’s total population.
Most importantly, Reels monetization is now coming along too. In Q3 2022, Reels was at a $3B run-rate but it was subtracting $500M from total revenue, since Meta couldn’t monetize it as efficiently as traditional methods. From Q4 2023 on, however, Reels has positively contributed to total revenue.
Meta proves time and again its ability and willingness to find more effective monetization methods by sacrificing profits in the short term for compounded profits in the long term. Consider that, in Q3 2022, Reels was at 140B plays daily with a $3B annual revenue run rate. In Q2 2023, Reels bumped to 200B plays daily, but compounded to a $10B annual revenue run rate. While reproductions were up 42.8% during the period, revenue from Reels was up 333%.
It’s worth noting that non-linear increases in monetization efficiency have paralleled a broadly improved macro environment, with online advertising bouncing back in recent months. The results are nonetheless encouraging, with price per ad increasing across all of the business’s geographies, especially in Europe and Rest of the World.
This is not the first time Meta evolved to thrive. Meta transitioned from desktop to mobile and posts to stories, respectively, to stay relevant. During each transition, the market thought Meta wouldn’t make it and, like this time around, Meta emerged stronger. Zuckerberg has been increasingly vocal since the start of 2023 that they owe their capacity to evolve to culture:
Now that brings me to the last part of our playbook for building leading services, which is our culture of rapid learning and experimentation across our apps.
When we decide that a new technology, like AI-recommended Reels, is going to be an important part of the future, we're not shy about having multiple teams experimenting with different versions across our apps until we get it right. And then we learn what works and we roll it out to everyone.
-Zuck during the Q4 2023 earnings call.
Meta has become a machine that excels at scaling and monetizing networks–so long as culture remains excellent, investors can rely on this innate ability. However, the most recent evolutionary turn has been truly special. Meta exhumed its entire underlying infrastructure to build a new foundation geared towards AI-recommended content, one that lays the groundwork for Meta’s next frontier of AI.
Reels was made possible thanks to Meta’s highly criticized, immense AI datacenter investments. As of Q1 2024, CapEx continues to be driven by Meta’s efforts to build out this infrastructure:
Capital expenditures, including principal payments on finance leases, totaled $6.7 billion, driven by investments in servers, data centers, and network infrastructure.
-Meta CFO Susan Li during the Q1 2024 earnings call.
These investments are broadly creating a machinery that enables Meta to capture the intelligence generated by its network. Apart from powering AI recommendations, intelligence will enable Meta to launch irreplicable and high gross margin digital services based on AIs that only Meta can train. These investments stand to multiply Meta’s earning power over the coming decade.
Meta’s network is full of people interacting with each other and brands, yielding utterly proprietary data. Further, Meta’s proliferating compute power is learning correlations between these interactions, with the aim of increasingly emulating and thus assisting any participant in the network.
With Meta’s daily active users soon accounting for over 40% of the world’s population, it’s an attractive business opportunity. During the Q1 2024 call, Zuckerberg shared how he believes Meta’s market share positions it to become the world’s leading AI company:
With the latest models, we're not just building good AI models that are going to be capable of building some new good social and commerce products. I actually think we're in a place where we've shown that we can build leading models and be the leading AI company in the world and that opens up a lot of additional opportunities beyond just ones that are the most obvious ones for us.
So this is what I was trying to refer to in my opening remarks where I just view the success that we've seen with the way that Llama 2 and Meta AI have come together as a real validation technically that we have the talent, the data and the ability to scale infrastructure to do leading work here.
And with Meta AI, I think that we are on our path to having Meta AI be the most used and best AI assistant in the world, which I think is going to be enormously valuable.
Meta AI just launched and, according to Zuckerberg, will be rolled out across 2024. With Meta’s track record of monetization, I believe the odds of success are high. Further, Meta’s prominence in the AI industry gives them priority access to the necessary talent. The rising market share of Python, a deep learning framework created by Meta, best evidences their hiring power.
Further, I agree with Zuckerberg that the way forward for Meta is open source. Meta won’t actually make money by keeping the best models to itself. They need to engage whatever mechanisms iterate the model faster than anyone else and then deploy it on their network. Open sourcing Llama is thus strategically an excellent move, which will lead to higher profits in the future.
2.0 A New Asymmetry
The AI investments are asymmetric because they lead to gains in time spent on the platform and buy Meta an option to lead the AI space–compounding the asymmetry I originally pointed out in my first deep dive.
With CapEx expected to range between $35-40B in FY 2024, capital efficiency is a natural concern for investors. During the Q1 2024 call Zuckerberg explained how, much like previous transitions, Meta will prioritize scale before monetization. This means that as CapEx continues to increase, profits will lag.
Although the investment in AI drags short term profits down, it’s actually immediately accretive to the fundamental value of the network, by increasing time spent on the platform. In turn, investing for AI in the long term also positions the company to build world leading models. Thus, the rising CapEx actually presents an asymmetry.
For example, in the Q3 2022 earnings call Meta management shared how a ‘single AI advancement’ led to a 15% increase in watch time for Reels on Facebook. Similarly, in Q1 2024 CFO Susan Li explained how moving from a multi-model architecture to a single-model led to an additional 8-10% gain. Meta now plans to transition all its apps to a single-model architecture, which should yield similar gains across the board.
Being vertically integrated in the AI business allows Meta to pick the highest leverage opportunities. It increases Meta’s ability to convert the CapEx into more time spent on the platform and better performing models down the line.
Further, Meta is planning on doing the same for the model it uses to automate ad campaigns, AI Lattice. The infrastructure investments compound all throughout Meta’s value chain by keeping people on the platform and monetizing their attention.
This asymmetry further compounds the original asymmetry that I highlight in my first Meta deep dive - that Meta was actually spending most of its money on supercharging Family of Apps and only allocating a portion of profits to Reality Labs, thus buying a long term option on the success of the Metaverse without betting the entire house.
In Q1 2024, Family of Apps expenses came in at $18.4B, representing approximately 81% of total expenses. In Q3 2022, Family of Apps came in at 82% of total expenses, which shows how Zuckerberg & Co.remain committed to their capital allocation framework: reinvesting a portion of profits generated by the apps into the Metaverse.
Since the investments in AI supercharge the earning power of the Family of Apps segments, it ultimately increases the total amount of capital that gets redirected towards the Reality Labs segment, without increasing the capital allocation towards the segment in relative terms. This increases the odds of Reality Labs working out without hindering the evolution of Family of Apps.
Increased profitability also stems from the now leaner culture. Regardless, the trend is encouraging. In the graph below you can see how operating income from Family of Apps is trending up, with operating losses from Reality Labs only increasing marginally. As a result, both net income and cash from operations are increasing in absolute terms, as you can see in the next two graphs.
3.0 Solving Attribution
We are yet to see numbers that prove revenue generated by Meta’s new attribution model is growing. On the other hand, progress on helping advertisers do more with AI is pleasing.
Although the AI assistant business opportunity is enticing, at present Meta doesn’t make money unless advertisers know how much they’re getting out of their advertising dollars. Apple’s ATT forced Meta to evolve its attribution infrastructure as depicted below, taking consumers to a messaging interface once they click on an ad.
The last thing we know about the click-to-messaging business is that it was at an annual revenue run rate of $9B in Q3 2022. Unfortunately, Meta has not reported on progress since, other than stating in Q4 2023 and Q1 2024 that Family of Apps ‘other revenue’ grew over 80% YoY, driven by business messaging growth coming in at $380M in Q1 2024. In Q4 2023 they also said “click-to-messaging ads continue to grow.”
The above is confusing because if click-to-messaging is reported within other revenue in Family of Apps, the $380M generated in Q1 2024 only equates to $1.5B annual revenue run rate, which is well short of the $9B run rate prior to Apple's changes.
Further, Meta has made tangible progress in terms of enabling advertisers to obtain better results. In August 2022, Meta launched Advantage+, which enables advertisers to automate the creation and optimization of ad campaigns. The initial launch was motivated by a test in Q3 2022, in which Advantage+ yielded a 32% increase in ad return on investment.
In Q3 2023 Advantage+ was at an impressive annual revenue run rate of $10B, although the metric was not updated in Q1 2024. In Q1 2024 it yielded a 28% YoY decrease in cost per click for advertisers and so it seems that the feature continues to deliver yearly performance increases. Advantage+ is AI driven and thus likely benefits from the aforementioned infrastructure investments.
Since Meta is looking to transition towards using less AI models, it’s likely that Advantage+ is or soon will be powered by Meta Lattice. Further, Shop Ads (an advertising feature which similarly bypasses traditional attribution methods) hit a $2B annual revenue run rate in Q4 2023. Meta mentioned that they were investing to scale this feature up in Q2 2023 and thus the progress is also fast.
4.0 Conclusion
When I wrote my first deep dive on Meta the market was convinced that the platform was going out of style, with TikTok taking over. Mark Zuckerberg has shown once again how culture eats strategy for breakfast. He’s essentially focused on fine tuning Meta’s culture and allocating capital to the right endeavors, namely AI servers, and the company has come out much stronger on the other side.
Although I’ve yet to see numbers on the click-to-messaging business, Meta’s next frontier is selling AI assistants to the world. However, like with any other technology, we do not know for sure whether we are faced with a mirage. That’s why it’s appealing to see AI investments that are immediately accretive to time spent on the platform, which together with getting better monetization is how Meta actually makes money.
The progress on the monetization side, beyond the aforementioned blind spot in the click-to-messaging infrastructure, is also driven by AI advancements. Advantage+ is delivering some notable performance increases and thus, I see the core business as being well positioned for continued growth and profitability.
Per my understanding of AI, I think that the AI investments are going to multiply Meta’s earning power. At present, they have data that no one else has (both in terms of quality and scale) and they also have AI compute infrastructure that begins to be unrivaled. Thus, if the world is to see AI assistants happen, Meta is likely to be one of the few companies behind them.
As mentioned, these assistants are likely to be hard to replicate and to deliver high gross margins, since over time Meta is likely to yield higher capital efficiency on the compute side, merely as the result of scale and the continuous optimization. Over the next five to ten years, this can proliferate Meta’s earning power, sending the stock up accordingly.
Until next time!
⚡ If you enjoyed the post, please feel free to share with friends, drop a like and leave me a comment.
You can also reach me at:
Twitter: @alc2022
LinkedIn: antoniolinaresc