Software

AI + ML

Microsoft turning away AI training workloads – inferencing makes better money

Azure's acceleration continues, but so do costs


Microsoft has explained that its method of funding the tens of billions it's spending on new datacenters and AI infrastructure is to shun customers who want to rent GPUs to train new AI models.

CEO Satya Nadella revealed that strategy on Wednesday during the software behemoth's Q1 2025 earnings call, during which execs detailed how the biz landed quarterly revenue of $65.6 billion and net income of $24.7 billion, improvements of 16 and 11 percent respectively.

Microsoft's Intelligent Cloud segment, which covers server products and cloud services and is Redmond's biggest single source of cash, saw revenue grow 20 percent year on year to reach $24.1 billion. Azure and other cloud services grew by 33 percent. Azure Arc – the multi-and-hybrid-cloud management tool – was called out for having won 39,000 customers.

AI was of course to the fore on the earnings call, with Nadella pointing out Microsoft's AI biz is on track to track $10 billion annual run rate next quarter. That would make it the fastest new product to do so in Microsoft history.

Investment analysts invited to ask questions during the call were curious to know how Microsoft is paying for the massive infrastructure build – $20 billion this quarter alone, most of it on datacenters and servers – required to deliver AI services. Nadella and CFO Amy Hood told a story of supply chain delays making it sometimes hard to bring new infrastructure online to satisfy demand, and impacting margins. But when new AI capacity comes online, increased revenue follows.

Another way Microsoft is managing things is by turning away training workloads.

"We're not actually selling raw GPUs for other people to train," Nadella said. "In fact, that's sort of a business we turn away because we have so much demand on inference" to power the various Copilots and other AI services.

"We kind of really are not even participating in most of that because we are literally going to the real demand, which is in the enterprise space or our own products like GitHub Copilot or M365 Copilot. So, I feel the quality of our revenue is also pretty superior in that context," Nadella added.

CFO Amy Hood added that Microsoft sees revenue won from inferencing as generating the funds to pay for future model training efforts.

But while focusing on inferencing may be a cunning plan, it's not helping to control costs – despite Microsoft's best efforts they rose 12 percent in the quarter. Even though some of that increase was due to Activision staff coming aboard, overall headcount rose two percent.

Most of Microsoft's business lines all posted solid growth: ten percent for LinkedIn; 14 percent for Dynamics; Xbox by 61 percent (thanks to the Activision acquisition); and search and advertising by 17 percent. Microsoft 365 commercial revenue grew 13 percent and consumer popped up five points.

But Windows OEM and device revenue dipped by two percent – and was forecast to do so again.

If that worries Microsoft's leaders, they didn't show it. And they had plenty of other wins to celebrate, reeling off names of blue chip companies that have put Microsoft's AI offerings to work. CFO Hood also pointed to "growth in the number of $10 million-plus contracts for both Azure and Microsoft 365" and "an increase in the number of $100 million-plus contracts for Azure."

All those wins have left Microsoft with $259 billion of future revenue to which customers have already committed. And the tech leviathan predicted strong growth in most of its major product lines, with Azure expected to grow at 31 or 32 percent, and Productivity and Business Processes wares to grow by ten to eleven percent.

It did, however, warn of some small future dull spots. One is a likely dip in revenue from on-prem server product revenue. But that's not a sign of decline – rather a reflection of the fact that this time last year Redmond was in the last weeks of selling long-term support for Windows Server 2012.

Similarly, a forecast slight fall in Microsoft 365 revenue was explained away as cyclical.

Microsoft's shares spiked from around $435 to $442 in after-hours trading, before settling at around $417 – seemingly a sign that investors hoped for better news on costs, but were comforted that AI is paying off. ®

Send us news
18 Comments

Copilot invades Microsoft 365 Personal and Family for an extra three bucks a month

Many users less than impressed by unexpected arrival of AI assistant in Word

Azure, Microsoft 365 MFA outage locks out users across regions

It's fixed, mostly, after Europeans had a manic Monday

Where does Microsoft's NPU obsession leave Nvidia's AI PC ambitions?

While Microsoft pushes AI PC experiences, Nvidia is busy wooing developers

How Windows got to version 3 – an illustrated history

With added manga and snark. What's not to like?

Microsoft’s latest on-prem Azure is for apps you don’t want in the cloud, but will manage from it

Azure Local is about hybrid management, not hybrid resource pools, and is catching up with virtual rivals

Google snags ex-Microsoft exec to helm cloud in the UK

Maureen Costello hopes to 'empower' businesses with AI

Microsoft tests 45% M365 price hikes in Asia-Pacific to see how much you enjoy AI

Won’t say if other nations will be hit, but will ‘listen, learn, and improve’ as buyers react – so far with anger

Microsoft sues 'foreign-based' cyber-crooks, seizes sites used to abuse AI

Scumbags stole API keys, then started a hacking-as-a-service biz, it is claimed

Intel debuts laptop silicon that doesn't qualify for Microsoft's 'Copilot+ PC' badge

TOPS, SCHMOPS, says Chipzilla, our NPUs may be slow but that doesn't matter

Microsoft fixes under-attack privilege-escalation holes in Hyper-V

Plus: Excel hell, angst for Adobe fans, and life's too Snort for Cisco

Azure networking snafu enters day 2, some services still limping

Struggling to connect to the cloud? You’re not alone

In AI agent push, Microsoft re-orgs to create 'CoreAI – Platform and Tools' team

Nad lad says 30 years of change happening in 3 years ... we're certainly feeling the compression of time