Meta Q3 Earnings: Strong Ad Growth, but Concerns on Overspending are Persistent

Hook:

TradingKey - $16B Tax Bomb Tanked the Stock 7%. Zuckerberg is spending big on AI. At this point, it’s all-or-nothing for Meta. Can they fight back? Stay with us in the coming minutes to understand Meta’s recent earnings and their plan to take over the world of AI.

Summary 

The performance of Meta has been quite satisfactory - up nearly 25% year-to-date on the back of strong earnings from resilient advertising, attractive multiple and AI narrative. The valuation is still somewhat low just because the company is seen as an underdog in the AI race, but that creates a favourable risk-reward scenario, in case Meta starts to execute well and their LLM starts grabbing market share from GPT.

Earnings Overview

The share price of Meta is down 7% after hours. Investors were shocked by $15.93 billion one-time tax charge, making a dramatic miss in the EPS. The charge was a non-cash one, and it was a result of the Trump’s One Big Beautiful Bill Act, and it is also expected to lower the future tax expenses. However, the company beat the revenue guidance.  

But if we ignore this charge, the earnings were not so bad after all.

The App Family daily active people (DAP) were 3.54 billion on average for September 2025, an increase of 8% year-over-year and an increase of 1.5% quarter-over-quarter. This demonstrates the resilience of Meta’s social media ecosystem, despite the competition from other apps.

Revenue growth for Q3 was 26%, which is a further acceleration from the growth in the previous quarters. The 26% growth rate is attributed to more ad impressions (+14%) driven by the fine-tuning of the company’s AI tools, as well as an increase in the average ad price (+10%).

Geographically, the United States and Canada, contributing nearly half of the revenue, were up 18%. Europe and APAC grew even faster, both with above 20%, driven by several factors such as recovery from GDPR (the EU) and increased adoption of Threads and Reels (Asia). LatAm was up just 6% mostly due to forex headwinds.

The operating profit margin was 40.1%, which is lower than the same period last year (42.7%), mostly due to increased R&D expenses from the AI ramp-up and the Ray-Ban AI glasses.

Reality Labs continued to burn cash, $4.4 billion, but this cash burn is well-supported by the cash-cow ad business.

In the balance sheet, we do not see dramatic changes. The sum of the cash held and the marketable securities is nearly $44 billion, greater than the total long-term debt of $29 billion.

The Free Cash Flows are around $11 billion (compared to $15 billion in 2024Q3). However, this quarter the company spent only $3.3 billion on share buybacks, compared to nearly $9 billion last year.

Meta said that its employee headcount was 78,450 as of Sept. 30, representing an 8% year-over-year increase, which gave further fuel to the fears that they are overspending.

Fourth quarter 2025 total revenue is expected to be in the range of $56-59 billion, or roughly 20% up from 2024Q4 – something that most analysts expected.

CapEx in FY2025 is expected to be $70-$72 billion (up $30-ish y/y), and as per Zuckerberg, the increase will be even more dramatic in FY26.

Meta AI Strategy

Moving away from the ad business, we would like to provide some insights into Meta’s AI strategy, or what Meta is actually spending these enormous amounts of capex on.

The defining feature of Meta’s AI strategy is that its models are open-source. The model architecture and the training code are publicly released under licensing. The model is free to download, and you just pay to compute it. Developers can use this model, with unlimited fine-tuning, customisation and modification.

Meta shares all these datasets, tools and resources openly so they can compete with closed models like GPT, where the model is simply a black box that the only access is through API access.

Which model is superior, open or closed, is too early to say. In terms of raw performance and speed, closed models do better so far. But open models are cheaper and easier to customise.

But one thing to know is that open-source models like Llama are not a charity. Meta tries to attract as many developers/companies to its platform as possible, providing Meta with the data flywheel. If everyone uses Meta, there will be no need to pay for the OpenAI API.

With the open-source nature of the model, Meta is not focusing on the consumer side but more on the enterprise side of AI. The flexible nature of the Llama open-source model may give an advantage to Meta in acquiring enterprise customers from industries with more specific data needs, such as finance and healthcare.

Even though the quality of the models is not on par with OpenAI at this moment, they can improve over time with more vendors joining the data flywheel. Also, the newly acquired Scale AI, which handles data-labelling, will contribute to the quality improvement of Llama.

Meta also works on MetaAI, a chatbot that is a peer of GPT. Meta AI is embedded across Facebook, WhatsApp and Messenger – all these platforms all together have billions of active users, providing a solid distribution channel.

In terms of advertising, Meta is using  Advantage+ for AI-powered conversion rates and ad bidding, moderation of content. According to many advertisers, Meta’s ad products increase the ROI, which itself provides a tail for the revenue growth (We already see Meta ad growth rates are higher than Alphabet’s).

Lastly, we have the hardware side, where Meta is pushing towards RayBan smart glasses. 

Risks

Loss in engagement is a risk related to the essential business of Meta, social media. Instagram and Facebook, though holding dominant market share, face competition from several frontlines. We have YouTube shorts that are gaining traction, according to recent Nielsen data. We also have TikTok, which will continue to operate in the US (probably under Oracle). These two players are direct competitors with Meta’s platforms for the attention of the viewers. Even the emerging Threads has to compete with the incumbent X. As Meta’s user base is already in a quite mature stage, the chance to see a drop in engagement is higher than the chance to see a spike in engagement.

Another risk for Meta is the AI battle. Meta’s AI models find applicability on the developer side; however, they are lagging OpenAI’s GPT and Google’s Gemini when it comes to user engagement. We are yet to see how the massive CapEx and investments towards AI headcount will play out. The risk itself is that Meta overspends a lot, but their AI efforts do not deliver the desired results. Competition here is also quite intense, with ChatGPT currently at the top of the AI pyramid, creating partnerships with other players like Shopify, Walmart, Etsy, Spotify and Figma. Further to this, there is Alphabet with its AI cloud business and already established ecosystem. Also, DeepSeek is open-source, too, and it may eat out from Llama’s share.

Increased capex and opex are another major risk. In fact, the same can be said for Alphabet, Microsoft and Amazon too, but compared to them, Meta has the smallest revenue and profit, making it more sensitive to overspending. Currently, the company is projecting  $70-72 billion of capex for the whole of 2025 and in 2026, this number will be significantly larger and within the range of the above-mentioned larger peers. 

Macro sentiment is also a risk, especially when it means slower ad spending. In fact, Meta is more prone to this compared with Google or Amazon, simply because Meta relies heavily on ad dollars, unlike Alphabet, which has cloud and Amazon, having both e-commerce and cloud.

Finally, there is the good old regulatory risk. As Meta is a global firm dealing with vast amounts of data, there is always a risk of a data breach. We already saw that a few years ago. 

Valuation

The valuation of Meta remains attractive at around 24x PE - lower than what we see with other Mag 7 peers. One reason is that Meta is still seen as an underdog in the AI race, despite all the massive spending.

This is actually good for the stock, as the low valuation provides a certain support level, while any positive surprise in AI will re-rate the stock up. A multiple re-rating to 30x PE, combined with a prudent estimate of 10% EPS growth, can put the target price of the company up 30% from what we see now.  

Get Started