As we draw to the end of 2025, what remains clear is that the world of cloud computing is still in a state of exponential growth.

While the industry seems to agree that this cannot continue indefinitely, no one is quite sure when it will slow down. Unsurprisingly, this is largely due to the growth of AI and the seemingly insatiable demand for it.

Consistently, we have heard from hyperscalers that they are “capacity constrained” and expect to remain so for the foreseeable future. While we are also seeing significant cloud adoption from enterprises, particularly as they wish to gain access to said AI capabilities and face a variety of challenges – the cost of hardware, and sheer inaccessibility of both data centers and the servers needed for AI workloads – to do so without turning to cloud providers, another key trend has emerged, that has yet to be truly solved: data sovereignty, or the lack thereof.

With all of this in mind, the cloud market has had an extremely dynamic year.

Money might literally need to start growing on trees

A shared experience between the big cloud computing providers – including Google, Microsoft, Amazon Web Services (AWS), Oracle, and others – is their skyrocketing capex.

For example, in 2023, Google had a full-year capex of $32.3bn. This jumped significantly in 2024 to $52.5 billion, and current projections from the company suggest that 2025’s capex could fall in the range of $91-93 billion.

Google isn’t alone. While AWS spent around $75bn in 2024, at the start of this year, it estimated that capex would surge to more than $100 billion, and by October had already spent nearly $90bn. Microsoft is certainly in the same boat – the most recent quarter saw it spend $34.9bn, and $11.1bn on data center leases alone.

This astronomical spending is paying off in terms of getting more capacity online (AWS launched 3.8GW in data center capacity in just one year), but all of the companies say that it simply isn’t enough, with all the big players reiterating that they are operating with significant capacity constraints.

This clearly demonstrates how massive the demand for AI is – with revenue also growing in the double digits each quarter for all of the providers – but it is also increasingly fueling the concerns surrounding the AI bubble.

Analysis from JPMorgan Chase & Co. estimates that more than $5 trillion will be spent on data centers and AI infrastructure over the next five years.

While the cloud giants are also sitting on giant piles of money, the bank still thinks that this level of build-out will require “participation from every public capital market as well as private credit, alternative capital providers, and even government involvement.”

A big part of the issue comes from the fact that the numbers just don’t quite add up yet. Deals are frequently made with hypothetical money.

A key example is in the almost Internet-breaking deal signed between OpenAI and Oracle. OpenAI agreed in September to purchase $300bn in compute power from Oracle over five years, an increase from the previously committed $150bn for the same time period.

OpenAI said in June that it had an annual recurring revenue of $10bn, and while this is growing quickly (July’s revenue reached $1bn), it isn’t quite in the $60bn-a-year realm.

Regardless, Oracle has apparently agreed to the deal despite not yet having the compute capacity currently available to fulfill it.

Later on in September, Oracle took on $18bn in debt financing to support the necessary buildout, and was later reported to be near receiving an additional $38 billion in debt.

The OpenAI news did wonders for Oracle’s stock value – and chairman Larry Ellison briefly became the richest man in the world – but a certain degree of skepticism remained.

KeyBanc Capital Markets estimated at the time that Oracle might need to source as much a $100m in debt funding over the next four years to fund the contract.

Big Red shared its latest earnings in December 2025, and while it was an undeniably strong performance – $16.06bn in revenue for the quarter and cloud income growing 34 percent Year-on-Year – the company’s share value still dropped some 10 percent in after-hours trading. During the call, CEO Clay Magouyrk addressed concerns about the debt estimates, telling analysts: “We've read quite a few reports that show an expectation of upwards of $100 billion needed for Oracle to go out and complete this build-out. Based on what we see right now, we expect we will need less, if not substantially less, money than that amount.”

While this was clearly hoped to ease concerns, the fact remains that across the board, we are not seeing spending land below expectations. Instead, it is massively surpassing them, quarter over quarter, and year over year.

Update – Since this piece was first published, Oracle investors have filed a lawsuit against the company, alleging they were misled about the amount of debt needed to fulfill Oracle's OpenAI contract.

Original piece resumes –

Rise of the neocloud

Within the context of massive capacity constraints for the hyperscalers, one subcategory of cloud computing providers has benefited immensely: the neocloud.

Unable to access the power needed to deploy GPUs, hyperscalers have been increasingly turning to neoclouds to lease access to their compute.

Microsoft has been particularly prolific in this area. In November, it was estimated that the company had signed some $60bn in GPU leasing deals in recent months. Nscale was the biggest beneficiary, netting $23bn across several contracts, but Microsoft has also tapped Lambda, Iren, Nebius Group, and CoreWeave.

While the neoclouds are in many ways competitors to the traditional hyperscalers, they are increasingly seeing hyperscalers as some of their largest customers, bringing in billions in long-term contracts. The neoclouds, however, are no strangers to the debt monster.

CoreWeave has perhaps seen some of the biggest successes this year. Born as a cryptomining company (as were many of the neocloud companies), CoreWeave made headlines in March when it made its initial public offering (IPO) on the Nasdaq market. The first of the large neoclouds to manage this, CoreWeave has grown dramatically in the months since, and in November had $50bn in returning performance obligations (RPO), or contracted revenue, "faster than any cloud in history," with $25bn added in Q3 alone.

The neocloud has signed significant deals with Meta, OpenAI, IBM, Microsoft, and Google over the year so far. At the time of its IPO, Microsoft represented more than 60 percent of its revenue, but today, CoreWeave says that no one customer represents more than 35 percent. In the background of this success, however, CoreWeave has been building up a hefty pile of debt.

Update – Since this piece was first published, CoreWeave shareholders have filed a class action against the neocloud, claiming that they had been misled about the company's ability to meet customer demand for compute capacity.

Original story resumes –

In November, the company revealed it had accrued $14 billion in debt so far this year. The most recent quarter saw CoreWeave paying $311m in interest expenses alone – just over 20 percent of its total revenue. The debt question does leave a degree of uncertainty over the entire industry – neo and not.

IBM’s CEO Arvind Krishna has said that he remains unconvinced that it will pay off – literally. Speaking on the Decoder podcast, Krishna said: "If I look at the total commits in the world in this space, in chasing AGI, it seems to be like 100GW with these announcements, which would cost around $8 trillion.

"It's my view that there's no way you're going to get a return on that, because $8 trillion of capex means you need roughly $800 billion of profit just to pay for the interest."

Sovereignty gets serious

Outside of the economics, politics and international relations have also played a key role in shaping the cloud market in 2026. US President Donald Trump took office again at the start of the year, and brought with him a new sense of political uncertainty.

Out of this was born a renewed fervor for digital sovereignty.

With the whole cloud market reliant on US-based firms, sovereignty has been talked about for a long time, but not with the same degree of anxiety and urgency that we have witnessed in the last year.

With US regulations enabling the US government to access data hosted by US companies (the CLOUD Act, in particular), it suddenly became apparent that any European company or government entity being hosted on Google Cloud, AWS, Microsoft Azure, OCI – and the neoclouds - could have their data exposed should the president press the hypothetical red button.

While the conversation heated up almost immediately after the new administration was instated, its urgency was reiterated after a real-life demonstration of the power of the US government in May.

The chief prosecutor of the International Criminal Court (ICC), Karim Khan, had his Microsoft email account blocked after President Trump sanctioned the court in The Hague because it issued arrest warrants against Israel’s Prime Minister, Benjamin Netanyahu, and his former deputy, Yoav Gallant, for war crimes in Gaza. The court said at the time that it had been “virtually paralyzed” as Microsoft had restricted services out of fear of being targeted by US authorities. Microsoft has denied restricting services to the ICC.

While it is hoped that friendly relations can be maintained across the Atlantic, it isn’t always a guarantee. With many governments literally dependent on the US clouds for operations, the importance of sovereignty becomes clear, if a difficult issue to resolve.

Over the course of 2025, the US cloud providers have all announced various initiatives to help guard against such issues – from on-prem and air-gapped offerings, to AWS’ launch of a European subsidiary – European countries remain uncertain about the status quo.

In October 2025, the EU Commission launched a €180 million ($209.1m) tender for sovereign cloud services, and said that it hoped the contract would serve as "a reference point for cloud providers and a catalyst for the growth of the EU cloud market, especially in the public sector."

While the desire is there for European providers, the reality is that there currently isn’t an EU-based firm that can really compete at the scale of the US companies. For now, the issue remains at an inflection point, and only time will tell if Europe successfully catches up.

There is so much more that could be said about 2025. Frankly, it has been one of the most exciting years to date – from cloud providers developing custom chips, to the growing adoption of hybrid models, and the massive competition and antitrust investigations, several of which are ongoing.

Should things continue following the same trajectory, we can only expect 2026 to be even crazier.