Microsoft Is Switching Off Oracle’s AI Cloud Capacity as the AI Spending Boom Faces Its First Reality Check

Microsoft is quietly switching off some of the artificial intelligence computing power it rents from Oracle, after deciding not to renew the contract. The move, reported on Friday 6 November 2026, is a small step on its own. But it comes at a moment when investors, politicians and the biggest technology companies are all starting to ask the same question: is the enormous spending on AI going to pay off?

The arrangement between the two companies dates back to 2023, when Oracle agreed to supply Microsoft with servers to run an AI-powered version of its Bing search engine. At the time, the deal was seen as proof of how desperate even the world’s biggest tech companies were for computing power. Microsoft, one of the largest cloud companies on the planet, was renting servers from a rival because it could not get enough of its own.

Three years later, Microsoft is walking away from that capacity. According to Bloomberg News, which first reported the move, Microsoft has been turning off the rented Oracle capacity in recent months after declining to renew the contract.

Oracle’s shares fell by nearly 3% in afternoon trading on Friday, after dropping by as much as 4% earlier in the day. Microsoft’s shares were little changed.

What happened

The core facts are simple. Microsoft rented computing capacity from Oracle. The contract came up for renewal. Microsoft said no, and has been switching the capacity off.

The full size of the contract was never disclosed, so nobody outside the two companies knows exactly how much money is involved. What is known is that the capacity was built around older-generation Nvidia chips, and that over time it stopped meeting Microsoft’s needs.

Bloomberg, citing people familiar with the matter, reported that the hardware “no longer met the company’s evolving needs”. The report did not name a specific chip model. Microsoft did not immediately respond to a request for comment.

Oracle pushed back. A spokesperson said demand for its AI infrastructure remains strong, including for earlier-generation graphics processing units, known as GPUs. The company pointed to comments from its latest earnings call about capacity that came up for renewal during its fiscal first quarter, and noted that in September it had renewed contracts for four-year-old GPUs at a 20% premium to their original prices.

In other words, the two companies disagree about what this means. Microsoft is treating the Oracle capacity as yesterday’s hardware. Oracle says plenty of customers still want it, and are willing to pay more for it.

Why Microsoft is walking away

To understand the decision, it helps to understand what a GPU is and why its age matters so much.

A GPU is a specialised computer chip that does the heavy maths behind AI. Training a modern AI model — teaching it to write, to answer questions, to recognise images — takes thousands of GPUs working together for weeks or months. The newest chips do that work far faster and use less electricity per calculation than the ones from a few years ago.

That speed matters because AI computing is extraordinarily expensive. The electricity bill alone for a large AI data centre can run into tens of millions of dollars a year. If a chip is two or three generations old, it can cost more in power to run than the work it produces is worth — at least for a company like Microsoft, which is building its own newer data centres at the same time.

Microsoft’s AI work has also changed since 2023. Back then, the urgent need was extra capacity for Bing, its search engine, after it added AI-powered answers. Since then, Microsoft has built out its own infrastructure for its Copilot AI assistant, for the Azure cloud platform that rents computing power to other companies, and for OpenAI, the company behind ChatGPT, in which Microsoft is a major investor and partner.

The workloads Microsoft runs today are bigger, more demanding and more specialised than the ones it needed help with three years ago. Older hardware that was good enough for an early AI search experiment is not necessarily good enough for what Microsoft is doing now.

Why this is bigger than one contract

The Oracle deal is only one piece of a much larger picture. Microsoft has committed more than $80 billion to renting computing capacity from outside cloud providers, according to Bloomberg. Those providers include CoreWeave, Nscale and Nebius — a group of newer companies, often called “neoclouds”, that have grown rapidly by building data centres and renting them to the big tech firms.

That spending is now under review. Bloomberg reported that senior figures inside Microsoft have grown frustrated over high costs and delays, tightened oversight of data centre projects, and debated whether to sign any new neocloud deals at all.

There are already signs of that tougher attitude elsewhere. On Thursday 5 November, AI cloud company IREN said it had delayed some infrastructure deliveries after Microsoft asked for changes to its network design. Microsoft is, in effect, telling its suppliers that it wants new chips, on time, and built to its own specifications — and it is increasingly unwilling to pay for anything less.

The week that shook AI stocks

The sell-off on Thursday was not an isolated event. It was the latest in a series of warnings that the AI boom is entering a more sceptical phase.

Earlier in the week, ratings agency S&P cut its outlook on Meta, the owner of Facebook, Instagram and WhatsApp, to negative, saying the company’s AI spending was outrunning its cash. Meta has committed to spending heavily on data centres and AI talent, and S&P’s decision was a signal that even the credit markets are starting to worry about how the bill will be paid.

At the same time, politics is catching up with the data centre boom. In the United States, Democrats who won control of Congress in the midterm elections this week are planning a wave of new legislation aimed at AI systems and data centre development, according to Reuters.

What Oracle says — and why it matters

Oracle’s response deserves attention, because it shows how differently the two sides see the same market.

Oracle argues that demand for its AI infrastructure is still strong. The company says it renewed contracts for four-year-old GPUs in September at prices 20% higher than the original deals. If that is right, it means there are customers who are happy to use older chips — perhaps smaller companies that cannot afford the newest hardware, or businesses running simpler AI tasks that do not need the latest speed.

There is truth in that. Not every AI job needs the newest chip. A company that wants to run a chatbot for customer service, or sort through documents, does not need the same hardware as a company training a giant new AI model from scratch.

But there is also a limit to that argument. The most valuable AI work — the work that justifies the biggest spending — does need the newest chips. And the companies doing that work, like Microsoft, are the ones with the most money to spend. If they stop renting older capacity, the market for it shrinks to smaller buyers, and the premium prices Oracle is charging become harder to sustain.

Oracle’s share price reaction tells its own story. A fall of 3% to 4% in a single day is not a collapse, but it is a clear signal that investors took the news seriously. Oracle has built a large part of its recent growth story on AI infrastructure, and on a huge computing contract with OpenAI that was announced in September 2025. Investors will now be watching closely for any sign that other customers are following Microsoft’s lead.

What it means for ordinary people

Most people will never rent a server from Oracle or build a data centre. So why does this story matter outside the world of tech investors?

First, because the cost of AI eventually reaches everyone who uses it. The price of an AI subscription, the cost of a search engine, the fees a business pays for cloud software — all of these are shaped by how much it costs to run the computers behind them. If the big companies learn to spend more efficiently, AI services could become cheaper and better. If they keep spending wildly, those costs will eventually be passed on to users.

Second, because data centres use enormous amounts of electricity and water. A single large AI data centre can use as much power as a small city. Communities near these facilities — in the United States, in Europe, and increasingly in Asia — are already asking who pays for the extra power lines, the extra water, and the higher prices that can follow. The more carefully companies choose what to build, the less waste there is for everyone else to pay for.

Third, because this story is a useful reminder that technology booms do not move in a straight line. There was a time, not long ago, when the shortage of AI chips was so severe that companies were renting whatever they could get, from whoever had it. Now the biggest buyer in the market is switching some of that capacity off. That is not a sign that AI has failed. It is a sign that the industry is growing up — moving from a phase of buying everything available to a phase of choosing carefully.

For Pakistan, the lesson is a familiar one. The global AI race will shape the prices Pakistanis pay for technology, the jobs available in the tech sector, and the infrastructure the country needs. A more disciplined, more efficient global AI industry is better news for everyone than a bubble that bursts.

What happens next

The most important thing to watch is whether other Microsoft suppliers report similar changes. If CoreWeave, Nscale, Nebius or IREN start talking about delayed deals, renegotiated contracts or softer demand, it will confirm that Microsoft’s decision is part of a broader pullback. If they report business as usual, the Oracle wind-down may turn out to be a one-off, driven by the age of that particular hardware.

The next test will come when Microsoft and Oracle report their next financial results. Investors will listen closely to what Microsoft’s executives say about their cloud spending, and to what Oracle’s executives say about demand for their AI infrastructure. The numbers will show whether this was a small pruning exercise or the start of something bigger.

There is also the question of what happens to the capacity Microsoft is switching off. Oracle says it can resell it at a premium. If it does, that will support Oracle’s argument that older chips still have value. If it cannot, the market will draw the opposite conclusion.

Finally, there is the political angle. The new US Congress is expected to take a closer look at AI and data centres in the coming months. Any new rules on where data centres can be built, how much power they can use, or how their costs are disclosed could change the economics of the whole industry — for Microsoft, for Oracle, and for everyone else in the race.

Frequently asked questions

Why is Microsoft ending its deal with Oracle?

Microsoft is winding down an arrangement to rent AI computing capacity from Oracle after deciding not to renew the contract. The capacity used older-generation Nvidia chips and, according to Bloomberg, no longer met Microsoft’s evolving needs. Microsoft did not immediately respond to a request for comment.

When did the Microsoft–Oracle deal start?

The arrangement was announced in 2023, when Oracle agreed to supply servers to run an AI-powered version of Microsoft’s Bing search engine. The full size of the contract was never disclosed.

How did the stock market react?

Oracle’s shares fell by nearly 3% in afternoon trading on Friday 6 November, after dropping by as much as 4% earlier in the day. Microsoft’s shares were little changed.

Is Microsoft cutting its AI spending?

There is no evidence of that. Microsoft’s decision concerns one set of rented capacity that fell short of its requirements. The company has committed more than $80 billion to renting capacity from other cloud providers, and it continues to build its own data centres. The move looks like a reassessment of what it rents, not a retreat from AI.

What is a GPU, and why does its age matter?

A GPU is a specialised chip that does the heavy calculations behind AI. Newer GPUs work faster and use less electricity per calculation than older ones. For a company running AI at Microsoft’s scale, older chips can cost more to power than the work they produce is worth.

What is a neocloud?

A neocloud is a newer company that builds data centres and rents computing power to bigger tech firms. CoreWeave, Nscale and Nebius are examples. They grew rapidly during the AI chip shortage, when the big companies could not get enough capacity of their own.

Conclusion

Microsoft switching off some rented Oracle capacity is not, on its own, a dramatic event. No one is suggesting that AI is finished, or that Microsoft is in trouble. The company remains one of the biggest spenders on AI in the world, and its shares barely moved on the news.

But small decisions can reveal big changes. Three years ago, Microsoft was so short of computing power that it rented servers from a rival. Today it is switching some of that capacity off because the chips are too old and its needs have moved on. That is what a maturing industry looks like: less panic buying, more careful choices, and harder questions about what each dollar actually buys.

For Oracle, the challenge is to show that the demand it talks about is real, and that its older hardware still has buyers willing to pay a premium. For Microsoft, the challenge is to show that its enormous spending — more than $80 billion committed to rented capacity alone — is being managed wisely.

And for everyone watching the AI boom, the message is simple. The age of buying everything is over. The age of choosing carefully has begun.

Sources

• Reuters, “Microsoft is winding down Oracle deal to rent AI cloud capacity, Bloomberg News reports”, 6 November 2026 — https://duke.fm/2026/11/06/microsoft-is-winding-down-oracle-deal-to-rent-ai-cloud-capacity-bloomberg-news-reports/
• The Hindu BusinessLine, “Microsoft winds down Oracle AI computing rental arrangement”, 6 November 2026 — https://www.thehindubusinessline.com/info-tech/microsoft-to-wind-down-oracle-ai-computing-arrangement-amid-cloud-partnership-review/article71675955.ece
• EuropeSays, “Microsoft winds down Oracle AI cloud capacity agreement”, 6 November 2026 — https://www.europesays.com/3357089/
• Finimize, “Microsoft Starts Turning Off Oracle AI Cloud Capacity”, 6 November 2026 — https://finimize.com/content/microsoft-starts-turning-off-oracle-ai-cloud-capacity

This article is for general information and education only. It is not legal or financial advice.