(Speedwell Members can read a PDF version of this post here)
2Q26 Business Update.
Meta reported 2Q26 and the stock sold off -9% the following day.
While they reported strong revenue growth of +27% (FXN), operating earnings were up only +9% after backing a legal charge of $2.4bn and severance of $1.2bn. With D&A increasing on their large capex build, earnings will continue to be pressured.
Below we can see that capex has continued to grow as a multiple of depreciation. Mathematically this creates a headwind for the business’s margins. (See this past update for a fuller discussion on depreciation). Depreciation reached $6.3b last quarter, +46% y/y.
Revenue growth was driven by ad impression increases of +14% and ad price increases of +12% y/y.
Above we can see that Europe was a bit softer with ad prices increasing just +10% versus the US at +20%. This divergence could partially be attributed to their “less personalized ads offering” in Europe.
3Q26 forward guidance of $61-64bn represents +19-25% growth. While this is still very strong growth at their scale, it is likely a larger potential deceleration than investors were expecting.
The other big news (which really shouldn’t have been news at all), was that Meta started talking about 2027 capex. It shouldn’t be surprising that after committing $130-145bn in capex for 2026 that 2027 would be another large capex year. While they didn’t provide exact figures, they noted their plans are geared towards “maximizing” 2026 and 2027 capacity. A similar, if not higher, amount of capex the following year is likely.
This brings us to a broader discussion of their compute strategy. Whereas in the past they have been clear that most of the capacity they are building out is for their own use, they recently started talking about how they are open to selling excess capacity. This has come alongside announcements of a new cloud business and Business Agent Platform.
On the call Zuckerberg emphasized that they don’t want to make “short-term” decisions, which in this context would mean just reselling compute for a few years at a premium—but they simultaneously didn’t take that off the table. Instead, their focus is on coupling compute together with “intelligence”. We interpret this to mean that they want to productize their compute through their Muse Spark AI models and new Business Agent Platform.
The strategy it seems is to leverage the short-term constrained compute environment to push sales of their new AI products (most of which have yet to be released) in hopes of creating a stickier service that can withstand a more abundant compute environment. This is a significantly different risk than their past strategy of building compute just for their own existing Family of Apps products to drive better recommendations and ad targeting.
For comparison, Google’s cloud service didn’t really start in earnest until around 2018 when Thomas Kurian took over. Prior to this they struggled to win large enterprise deals. 2017 GCP revenues were just $4bn and it would take until 2023 for them to reach a small $1.7bn profit with $33bn in revenues. While their capex spend over those years was much smaller (in the $20bns range) it still is worth noting that building a product and getting broad adoption took them years.
The hope for Meta is that the shortage of compute and demand for AI could perhaps sway customers to adopt Meta Compute (their cloud service) sooner than they would have otherwise. Meta may have decades of experience managing infrastructure, but they are new to client services and still have much to prove in terms of their AI offerings. They also are competing in a market with GCP, Azure, and AWS, who are formidable competitors with a longer history in this market. If adoption is slower than they hope they, their alternative strategy is to just sell access to raw compute like a neocloud.
As it stands today this is a good strategy as it de-risks a lot of the capex investment, but of course the real fear is that industry capacity outstrips demand, which could mean the premiums they expect to garner from selling raw compute are short-lived. Right now would be the opportune moment to strike longer term contracts on their compute for high premiums, but they are reluctant to do that because they want to try to productize their compute first, which would create a more sustainable business.
The net of this is that the “back up” plan of being a neocloud may not be as attractive of a proposition if they wait a few years, making their capex spend more of a bet on this new cloud service working, without a great alternative if capacity is no longer as constrained in a few years.
Their full year 2026 capex is ~$140bn and not all of that is going to Meta Compute, a meaningful (but unknown) portion is still be reserved for their Family of Apps to improve recommendations and targeting—which is already showing signs of working. They noted their new GEM Model drove a 15.7% conversion uplift on Facebook. Instagram global time spent grew double digits y/y and Facebook Video time spent grew 9% y/y globally. This gives Meta real world proof points that their AI spend is yielding an actual return.
With LTM cash flows (after SBC is backed out) is around $105bn, which means they are committing basically at least 2-3 years of capex to this build out for their cloud and AI services. (Free cash flows went negative in 2Q treating SBC as a cash expense. FCF were about neutral not backing out SBC.) While a worst case scenario of them being stuck with excess capacity in an industry-wide compute supply glut could mean a poor ROI on this spend, it won’t risk the company. The upside could be a new cloud and AI business with a multi-decade runway and unique ability to acquire SMB customers through their advertising arm (more on this in a moment).
Investors are likely still shell-shocked from Meta’s >$100b Reality Labs effort, which over half a decade later still shows middling results. They continue to burn >$4.5bn a quarter in that segment with a promise that losses will narrow, but not clear path to meaningful profits that would rationalize that level of investment.
Having said that, most investors have already discounted Reality Labs and they do have a meaningful shot of being the leader in AR/VR, which one day could still be a huge market. Adoption just has been very slow. Better AI could be the unlock though to make glasses (which have limited on platform controls) more useful. AI is potentially saving this product and giving consumers a real use case by having an AI assist always with you without the need to whip out your phone and type or snap photos for context. We still view it as a call option though and consumers’ general distrust of Meta plus their lack of a mobile platform makes it a harder sell.
While Meta hasn’t had success in moving beyond their Family of Apps business, they currently have a ton of optionality if things start to go right for them. On a long enough time frame (say over the next decade) AR and VR seems inevitable in our opinion, and they already have the most popular consumer product on the market in this space.
Their AI initiatives though seem to be more in their wheelhouse as they can leverage their position with tens of millions of SMBs who already advertise with them to do more for their business. For a lot of businesses Facebook and Instagram are their most important sales channel and by connecting to their budding Business Agent platform they can help them run their business better and increase sales. The Business Agent Platform can connect to existing catalogs, CRMs, inventory management, and talk to customers to answer questions, take orders, or deal with returns. We know that AI is going to be very helpful to running a business, the question is “whose AI will it be?”. Meta is in a good position to take over this functionality for a lot of businesses because it is tied to their most important sales channel. It is a little bit like how Adobe leveraged their creative apps (photoshop, premier, illustrator, etc.) to move into analytics and help deliver the creative copy (the Adobe Experience Platform).
One of the reasons why so many other advertising platforms like Snapchat, Pinterest, and Twitter struggled to gain the advertising inventory that Facebook originally had is because Facebook had tens of millions of businesses that already had a presence on the platform. They used that existing presence to convince them to advertise. Now they are using that same connection to push deeper into a business’s productivity suite, all powered by AI.
That is not to say success will be assured, but they don’t need to win over everyone in order to build a real business here.
The other call option is that with the billions they spent on hiring top AI researchers and hundreds of billions spent on compute, they are able to actually build a leading edge frontier model. If that happens, their cloud business has another real unique selling point as they can couple the sale of the two together.
So, in summary, the core Family of Apps business is already benefiting greatly from AI and in our opinion will continue to improve recommendations and ad targeting for some time, which will help improve the product and their ability to monetize it. Their capex bet is somewhat protected against the fact that they could always resell the compute (even if it isn’t at as high of a premium as exists today). Then investors have call options in Reality Labs, Meta Compute, and the possibility they build a leading-edge frontier model.
In total that is 5 different call options the business has: 1) Meta Compute, 2) Reality Labs, 3) the Business Agenet Platform, 4) a Neocloud, and 5) creating frontier AI model. Of course, whether this is a good risk/ reward will ultimately come down to the price investors pay today and whether some level of success with these new initiatives is already priced in. To better understand what is priced in, we will turn to our Reverse DCF in the next section.
Just before we do though, it is worth surfacing a risk that didn’t get much airtime on the call yesterday. Earlier this year there was a California and New Mexico verdict that found Meta at least partially liable for endangering children and being a factor in causing severe depression and anxiety. The New Mexico verdict ordered damages of $375mn and California awarded damages of $6mn. Of course these are small numbers for Meta, but the risk is that this just the beginning and could open up the flood gates for similar personal injury lawsuits.
Meta’s defense primarily hinges on Section 230 of the Communications Decency Act, which protects platforms from content that causes harm from 3rd parties. Instead of going after Section 230, the plaintiffs successfully argued that the products design is the issue, with algorithm being designed to keep children engaged in their product, even to their detriment. Meta is currently appealing the verdict, noting how many real-world factors contribute to depression and anxiety. Complicating this is that research has suggested teens with zero social media are more often to report lower well-being than those with some. If they are not able to successfully defend themselves, this could be a multi-decade drag of litigation against the company with perhaps a large global settlement.
Valuation.
We estimate normalized 2026 EPS to be around ~$29 for the year (after backing out one-timers), which means they trade at ~18x earnings. If we back out the reality labs losses, that multiple drops to about ~15x earnings. To get better sense of what is priced in though, we re-ran our reverse DCF below for 3 growth scenarios.
The 3 growth scenarios start at 5%, 15%, and 20% and then continue to fade according to the schedule below. For 3Q26, they guided revenues to be 19-25% with a 1% headwind from FX. Since our starting growth rate is for the next 5 years, it makes sense to take a bit of a discount to that in our 20% high growth scenario.
Below we can see that investors are pricing Meta stock at about a high single digit growth rate, assuming they can maintain their margin structure over the long-term. If an investor believes they can grow faster than high single digits for the next ~5 years with growth fading their after, the discount rate implied is an above historical average stock market return. In this math we are not explicitly valuing any of the other bets such as Meta Compute, Reality Labs, the Business Agenet Platform, a Neocloud, or creating frontier AI model. (However, an investor is certainly welcome to assume some of the growth Meta achieves in the future in this DCF are from those initiatives).
As always, it is on each investor to make their own judgement on whether they think the potential returns are worth the risks they may perceive. We named our research firm Speedwell after the ship that helped ferry passengers to the Mayflower. The idea is that we want to help you get into the position to take the journey, but ultimately, we are not going to go with you. Each investor must ultimately make their own decisions.
*At the time of this writing, one or more contributors to this report has a position in Meta. Furthermore, accounts one or more contributors advise on may also have a position in Meta. This may change without notice.