How AI is Draining Google’s Finances

On Wednesday, Google's parent company Alphabet published its financial results for the third quarter of this year. And well, they weren't well received. Spending came in hot, revenue came in soft and the stock duly slumped by 10% in after hours trading, wiping about $400 billion dollars off Alphabet's market cap in barely a few hours. So in this video we're going to explain why Alphabet's stock just slumped and why, shock, it's all to do with AI. Sick of slop? Well TLDR's content is entirely AI-free, with us having strict rules when it comes to the use of AI in our company. So while we might not be perfect, at least you know we're human. Find our full AI policy linked in the description. So simply put, the fundamental problem for Google is that spending came in higher than expected and revenues came in lower than expected. And you don't need an MBA to know that's not good business. Let's start with spending. In the third quarter alone, which is July to September this year, Google spent a whopping $87 billion, which is up 22% from the same period in 2024. This follows a broad trend, with spending in the year to date up 21%, reaching $236 billion in the first three quarters. Obviously the culprit here is AI, but what does that mean? Well, we can break down Google's soaring AI spend into two main categories, capital expenditure and TPU inventory build-up. If those things mean nothing to you, don't worry, we'll explain each in detail. And when we're done, you should have a clear idea of what hyperscalers are actually spending money on as they fund the AI build-out. First, capital expenditure, or CapEx. This refers to money being spent upfront by Google towards its own infrastructure. As you can see on their balance sheet, this value of the company's physical assets has leapt up since last year, which reflects growing capex investments. Right there you can see $51.4 billion spent in the last quarter alone, which we're told is funding Google's AI build-out. But what on earth is that money being spent on? Well, according to CFO Anat Ashkenazi, it's a 65-35% split between active computing hardware and physical data center infrastructure. The computing stuff is essentially GPUs, CPUs and TPUs, along with a tech that allows huge chip clusters to communicate with each other. The data centre's costs go towards physical buildings, their energy supplies and cooling systems. So that's CapEx, next is TPU inventory. Google's AI build out strategy is two pronged. On the one hand they're building out their own computing power so they can reap the rewards of the AI boom. On the other hand they're acting a bit like Nvidia and designing their own specialised AI chips called TPUs or Tensor Processing Units, which they plan to sell on to clients like Anthropic, Apple and Oracle. If you look at that inventory line on the balance sheet, you can see that since last year an additional $10.9 billion has been spent building up Google's inventory. A few years ago, that line would have reflected products like Pixel phones, sitting in warehouses waiting to be sold. Now we're told that the figure is almost entirely accounted for by TPU building. But before you leap to conclusions, there's an important point here to keep in mind, which illustrates that Google's AI spend isn't without justification. Google's cloud CEO, Thomas Kurian, Pointed out in Wednesday's earnings call that the company only produces components and builds TPU inventory after signing contracts with customers. In other words, while this multi-billion dollar figure is sticking out on their books like a sore thumb, those processors are going to be shipped soon and turn into revenue. In broad terms this speaks to a dynamic in the AI build-out which benefits hyperscalers like Google, Amazon and Microsoft over companies further down the food chain like OpenAI and Anthropic. Sure the hyperscalers are spending huge amounts building out America's computing power but they also have significant commitments from enterprise customers which if all goes well means huge revenues will follow. Of course we'll be the first to admit that if all goes well is doing a lot of heavy lifting in that sentence. That brings us to the other side of the ledger – revenues. At first glance, revenues looked pretty good. Total revenue for the quarter came in at $127.6 billion, only a tad below analysts' forecast of $128 billion and up 25% compared to a year ago, i.e. to the third quarter of 2025. The thing that slightly spooked the markets, however, was where that revenue came from. For context, Alphabet's revenues are often broken down into into three segments. Google services, which covers stuff like ad revenues from Google search and YouTube, as well as subscription fees. Google cloud, which basically rents out computing power and software to businesses and other organisations, so they don't have to build and run their own data centres. And other bets, which covers the loss making more experimental businesses, and is usually a relatively small slice of the overall pie. Historically Google services has accounted for the vast majority of all revenues and has been by far and away the most profitable part of the business. The thing that spooked the markets this time around is that Google services revenue came in below forecast at 96 billion dollars. That's up 11% year on year but down slightly compared to the previous quarter. This was in part because revenues from Google Search, historically by far the largest and most profitable part of Alphabet's business, came in at $63.3bn, about $1.5bn below forecast and basically flat on the previous quarter. The only reason that headline revenues came in line with expectations is because of an above forecast print from Google Cloud, which generated about $30bn in revenues for the quarter, roughly $2bn more than forecast. Now, at this point you might be asking, what's wrong with all this? It sounds like Alphabet is essentially diversifying, becoming less singularly reliant on the services part of the business, and offsetting any shortfalls over there with more revenues elsewhere, like Google Cloud. And yet, while diversification is generally a good thing, Google services has been a supremely reliable money printing machine for Alphabet for years now, and markets apparently aren't convinced that other parts of the business, like AI, can truly fill its shoes. Google Search, for instance, has been a money printing machine, in large part because it's had a near monopoly on the search engine market for years, if not decades now. It's nice that Google Cloud is making money, but it only became consistently profitable in around 2023, and it's not obvious, at least to the market, why it would ultimately be as profitable as, say, Google Search, given the lack of any obvious moat. At the same time, the trend lines look like they're squarely headed in the wrong direction for search at the moment, as people move away from traditional search towards AI chatbots. Alphabet themselves admitted on Wednesday that the growing popularity of Google's AI mode was taking a toll on search revenues, because AI queries apparently have lower monetisation rates. All in all this combination of lower revenues and lots more spending meant that Alphabet's free cash flow came in at negative nine billion dollars, down from negative 5.9 billion dollars in the previous quarter and only the second quarter in the red since Alphabet went public in 2004. So that's Alphabet's financial situation right now. They're in a somewhat risky position which explains the stock dip this week but there's certainly room to recover if their bet on AI pays off. Exciting announcement. I'm happy to reveal that the cover of the next issue of Too Long, our 10th. The main story is all about the ascendancy of China, spanning energy, artificial intelligence, global trade and an increasingly assertive posture towards its neighbours. The country's trajectory points towards total dominance. Yet beneath the surface of this meteoric rise, cracks may be forming. You can pre-order your copy right now and if you use code 10, you'll get £3 off any new subscription, including the too-long and TLDR party bundle subscription. Also worth mentioning, as always, there's much more than just the cover story. In fact, and I know you'll love it when I say this, your talk of our longest issue ever is so repetitive. I don't want to hear about your longest issue ever. It's only one or two pages more than the other issues anyway. But this time it truly is longer, at 100 pages, to celebrate our 10th issue. That means that we also have tons of space to discuss the US midterms, the return of company towns, the future of Taiwan, the seeming rise of the AFD, Burnham reversing Thatcherism, Ukraine's corruption scandals and much more. Find out more about the latest issue and subscribe to TooLong by heading to our website, it's linked in the description.