The 4Q25 Earnings Season is beginning and we thirsty for some updates!! We will start with our preview for 4Q25 and 2026 and main themes to follow during the Season!
Starting with one of the main tech industries in this century, we will discuss what we are seeing in the Cloud Infrastructure Market!
Cloud Revenues (The Good)
On first notice, cloud revenues are accelerating significantly for the last two quarters, reaching +26,6% for the Big 4 (AWS, Microsoft, Google and Oracle) in 3Q25, it is ~5pp above the rate they were running the last 5 quarters before. These are tangible revenues from offering compute power for the AI revolution.
The growth between the players has been uneven:
- Oracle is growing +50% and accelerated further to 67% on 4Q25/FY2Q26!! Oracle is getting an advantage from its infrastructure design as well as their position in Database Management Software (which companies are accelerating migration to cloud to unleash the power of AI). Oracle also managed to negotiate contracts with the 3 world largest hyperscalers so it can put OCI inside their DataCenters accelerating multi-cloud migrations.
- Azure is growing 40%, the largest acceleration between the peers in the last 3 quarters; Azure is the greatest beneficiary of OpenAI training and inference workloads. The growth on Azure is restrained by supply, given that the Microsoft Applications are also growing more than expected from the demand of Agentic AI in Microsoft Software Suite, as well as cloud migration acceleration from legacy clients (both Oracle and Microsoft get the benefit from large Data Center clients that are finally migrating data massively to cloud so it can feed AI models for training).
- Google also is registering an acceleration in Google Cloud growing 33% in 3Q25 but should accelerate to High 30s going forward. Google benefits from its vertical integration with own infrastructure (TPUs), platform and LLM development, being Gemini3 a major driver of GCP consumption. Google AI for Ads is also a huge application and is transforming the ad ecosystem with AI tools for advertisers.
- AWS is the largest of the players and also the last player to ramp AI capex (see below). AWS acceleration is coming slow, but it is also accelerating, reaching 20% in 3Q25 (+3pp QoQ of acceleration) and we believe this could go further in 4Q25 with the star of Project Rainier.
But besides the uneven distribution of new revenues between the hyperscalers, what we see is that the Delta Revenue is in a new level from what we were seeing previous quarters:
But MORE important than the revenues numbers are the DEMAND SIGNALS.
From Microsoft call, what we could hear:
“…demand again exceeded supply across workloads, even as we brought more capacity online.”
“In Azure, we expect Q2 revenue growth of approximately 37% in constant currency as demand remains significantly ahead of the capacity we have available.”
“Therefore, we now expect to be capacity constrained through at least the end of our fiscal year.”
“We’ve been short now for many quarters. I thought we were going to catch up. We are not. Demand is increasing.”
From Google:
“Cloud had another great quarter of accelerating growth with AI revenue as a key driver.”
“Cloud backlog grew 46% quarter-over-quarter to $155 billion.”
“We’re investing in TPU capacity to meet the tremendous demand we are seeing from customers and partners…”
“We are signing new customers faster… We are signing larger deals. We have signed more deals over $1 billion through Q3 this year than we did in the previous two years combined.”
“We have more demand than we have supply.”
“…we still expect to remain in a tight demand/supply environment in Q4 and 2026.”
“…operating margin benefited from strong revenue growth and continued efficiencies in our expense base…”
And from Amazon:
“AWS is growing at a pace we haven’t seen since 2022, reaccelerating to 20.2% year-over-year, our largest growth rate in 11 quarters.”
“Backlog grew to $200 billion by Q3 quarter end and doesn’t include several unannounced new deals in October, which together are more than our total deal volume for all of Q3.”
“Trainium2 continues to see strong adoption, is fully subscribed is now a multibillion-dollar business that grew 150% quarter-over-quarter.”
“I look at the momentum we have right now, and I believe that we can continue to grow at a clip like this for a while.”
“As fast as we’re adding capacity right now, we’re monetizing it.”
Cloud Capex (The Bad)
Now, the villains in this real life narrative without a doubt are the Capital Expenditures. You invest too little and you will lose clients, you invest too much, prices go down and your return says bye bye.
And the fact is that Cloud Capex is growing strong, stronger than revenues. This is the epicentre of the discussion we see everywhere: Is AI a Bubble?
In fact, there is little to learn from this question. Off course valuations are stretched; Off course companies don’t know for sure if their investments are going to have the desired returns in the expected timeframe. But we can’t simplify that question to a single yes/no question.
The FACT is that investments will keep coming and will keep growing! According to Dan Ives (link?) we are going to see 10 years of investment happening in 2.
We see this very clear from the main Cloud CEOs:
Microsoft about Capex
“This quarter, roughly half of our spend was on short-lived assets, primarily GPUs and CPUs, to support increasing Azure platform demand… The remaining spend was for long-lived assets… including $11.1 billion of finance leases that are primarily for large data center sites.”
“We will increase our total AI capacity by over 80% this year and roughly double our total data center footprint over the next 2 years, reflecting the demand signals we see.”
“Therefore, total spend will increase sequentially, and we now expect the FY ‘26 growth rate to be higher than FY ‘25.”
“We’re continuing to invest aggressively due to the demand we’re experiencing from Cloud customers as well as the growth opportunities we see across the company.”
“In the third quarter, our CapEx was $24 billion… The vast majority of our CapEx was invested in technical infrastructure, with approximately 60% of that investment in servers and 40% in data center and networking equipment.”
“We now expect CapEx to be in the range of $91 billion to $93 billion in 2025, up from our previous estimate of $85 billion…”
“Looking out to 2026, we expect a significant increase in CapEx…”
Amazon
“To enable customers to do so, we need to have the requisite capacity, and we’ve been focused on accelerating capacity the last several months, adding more than 3.8 gigawatts of power in the past 12 months, more than any other cloud provider.”
“You’re going to see us continue to be very aggressive in investing in capacity because we see the demand.”-
“…we’re now double the power capacity that AWS was in 2022, and we’re on track to double again by 2027.”
“This primarily relates to AWS as we invest to support demand for our AI and core services and in custom silicon, like Trainium as well as tech infrastructure…”
“Looking ahead, we expect our full year cash CapEx to be approximately $125 billion in 2025, and we expect that amount will increase in 2026.”
One thing we would like to observe is that Amazon was the latest of the three hyperscalers to ramp up AI capex. Microsoft accelerated in 2Q23 and Google in 4Q23. We see Amazon ramping capex strongly after 2Q24. So we is believe its natural that AWS is taking a little longer to show the AI acceleration. But we will argue in an future Amazon piece that this acceleration should come faster than we thought!
All these investments are great news for Semiconductors and Hardware companies in 2026. They started soon to see this wave of demand from hyperscalers and startups mainly for their internal workloads.
But now we are seeing the broadening of demand, the start of the ramp up into production of AI Applications, companies achieving tangible efficiencies. This year we believe is the year of the Cloud Wave!
The Cloud Wave
The chart below shows exactly what we are observing!
This is a core Chart. First, we saw Capex accelerating mainly after 2Q24. New Data Center leadtimes runs from 12 to 24 months, so we saw capex growing strong for 6 quarters (18 months) and now, strongly in 2Q25 the capacities are coming online, and the Applications market will demand this capacity!
From a application CEO we hear that he would demand 10x increase in capacity for the price because it would increase the reasoning and the quality of the answers on his platform!
But we will only know when its too much when its too late.
That’s why bubbles happen. But we should see that this is not the end of the world!
A quick tale about that: Mid 2021 we all expected ecommerce to keep stronger growth due to definitive change in the consumer mindset about commerce after the pandemic.
We ended up really wrong, so did Amazon. A lot of Distribution capacity was put in place and was not being fully utilized due to growth deceleration. The result is that Ecommerce Margins went negative territory and the company spend the next 2 years adjusting growth plans and optimizing capacity for demand, which pick up again later in 2023 and in 2024 the company was back again close to breakeven.
The “overinvestment” allowed Amazon to increase the customer service first, because it got closer to consumers and allowed fast delivery. Only then the company optimized for cost and return, and again in 2025 we hear about another wave of growth investments. The company overcame the overcapacity, delivered growth at an enormous scale and maintained its leadership in ecommerce. Even after the overinvestment mistake.
What we mean by that is that in overinvestment cases, if the structural demand is robust enough and the company has an advantage (like economies of scale, distribution platform) and also there is no use of dangerous levels of leverage (which is an really important and not the case here!), this companies will eventually deliver the growth aligned with long term structural market, which in the Cloud and AI we believe that this is the case.
Cloud Depreciation and Margins (The Ugly?)
One important bear argument relates to Cloud Companies Margins, not only their future prospect, but also questioning accountant assumptions made about useful life of investments (notably Mr. Burry).
So far we have not seen this effect. Actually, the data support the notion that chips within 6 years are still being rented. The Nvidia V100, for example, can be rented for USD 3 to USD 30 per hour, depending on other configurations. And can still be used for low critical use cases like generating data for training or inferencing in low-cost applications.
That shows that the demand for compute can be overwhelming that all compute GPUs are being utilized for different types of applications. The new and shine Blackwells and Vera Rubins will be directed to newer and critical workloads as drug discovery, DNA simulation, autonomous vehicles and robots, personalized ads and movies and so on.
The Worst Case: It’s important to note here that a worse scenario, a 4 year depreciation would mean an impact from 5-8pp in Cloud Margins, which is a lot, but even so we believe the IRR of GPU for hyperscalers would be double digits (maybe not the mid-high 20s we would prefer, but mid-teens would still be great from capital allocation perspective)! If you’re curious we could dive deeper in that subject!
So we would conservatively project lower margins from depreciation, but we should not forget that may be some offsetting expenses. We believe the main sources should come from:
- Revenues Surprise: We believe Revenue Growth should surprise market given what we are hearing from enterprises. We believe there should be a boost in AI workloads in 2026 given that companies are piloting projects for enough time to start deploying in large scale, which we believe could translate in a Cloud strong growth surprise in 2026.
- Operational Leverage: A high utilization rate in D0 (given that supply is constrained, utilization tend to max with new capacity being deploy) should contribute to sustaining margins.
- Automation and Efficiency: AI can increase efficiency from automation in operations, research, development and even service and sales activities.
- Fungibility: AI compute capacity can be redeployed for different applications than initially thought, giving an fungibility optionality where companies can increase the quality and scope of their services.
- Custom Chips: AWS and Google have an advantage over Microsoft regarding their own internal developed chip, which is deployed at cost, reducing the share of the expense Nvidia chips. That mix could be a factor in 2026 and going forward in price competitiveness.
*Worth notice that we are hearing about lower prices being offered buy all three clouds mainly for longer term contracts, with hyperscalers trying to increase visibility for capex deployment.
The Unknown: Future Volumes and Prices
To close the loop on this discussion, returns at the end will depend on future revenues, that we can breakdown in volumes and pricing. We don’t have a crystal ball and don’t pretend to know anything for sure, but Future Performance Obligations can be a great indicative of this future demand, once they imply heavy penalties for cancelations.
Despite the incredible numbers above, we are including Oracles 300 bi from OpenAI, which a lot of people do not fell comfortable because of the concentration risk and high levels of debt. We agree, so:
Even if we remove Oracle and not counting USD 250 bi that OpenAI has with Microsoft, even so we are seen a stronger growth in RPO for cloud companies, much stronger than revenues! We don’t believe that the Cloud revenue would converge to this numbers because the tenure of contracts can fluctuate, but is a very strong signal of demand and support our acceleration thesis!
We have being following H100 and A100 on-demand prices to check for deep discounts and we are not seeing that so far as well.
To be continued…
