And net-net, what happens is some of these knowledge worker tools that people have used all the time, right? Because when you think about Teams, when you’re having a meeting, you’re not doing a random meeting, the meeting is in the context of some business process. It could be a supply chain meeting where you’re trying to understand which suppliers to bet on or what terms to do. And so now you can access all that data right in the Team’s context. So that’s I think what’s exciting for us, having built all these horizontal tools, which I would say we’re under underappreciated for the amount of work. How people use those tools to make progress on business process, but we now get to bridge that between the business applications and knowledge worker tools, tomorrow horizontally.
Raimo Lenschow: Okay. Perfect. Thank you.
Brett Iversen: Thanks Raimo. Operator, next question please.
Operator: And the next question comes from the line of Michael Turrin with Wells Fargo. Please proceed.
Michael Turrin: Hey. Great. Appreciate you taking the question. I wanted to go back to Azure. You’ve been hinting at stabilization there for the past couple of quarters, but still very good to see the balance. Maybe you can expand on just what the commercial bookings number, appreciating the variability there does in terms of visibility. And any characterization you can give us around what you’re seeing in areas like cost optimization and core workload growth coming back is just helpful context for us in unpacking the numbers. Thank you.
Amy Hood: Thanks, Michael. I may take those a bit in reverse. It’s a little easier to address them. When you think about – we’ve been talking about sort of stabilization and what you saw this quarter, if you break down the Azure number as you saw, which I think I talked a little bit about with Karl was 7 points of contribution from AI, and you could call them the difference ’24 from our core really Azure business. And within that, the activity we saw and the consumption side was really this balance that we were quite used to and have seen throughout the cloud transition. We saw new workload starts and we saw optimizations. And then those optimizations create new budget, and you apply it. And that cycle which is actually quite normal.
We saw it again this quarter in a balanced way. And I think when we talk about stabilization or even what we saw between Q2 and Q3, which is a bit of acceleration in that core, was a lot of the newer project starts relating back to not just AI starts, but lots of other workflows. The companies are still going from on-prem to cloud, Satya mentioned migrations. And some of that, which I know isn’t as exciting as talking about all the AI projects. This is still really foundational work to allow companies to take advantage of the cost savings and the total TCO is still really good. And so I think that balance is really what you saw this quarter, and I do feel like there wasn’t really a big difference, Michael, across industries or across geos. So I would say it was actually pretty consistent is the other maybe texture that I could give you to that question.
And so then when you’re saying do we keep sort of pointing to stabilization, I really do look sort of workload to workload. What are we seeing? Where it starts? And this one actually felt quite balanced and optimization looks like they normally would, which by the way, is super important. It’s something we encourage customers to do. You want to run your workloads as efficiently as you possibly can. It’s critical to customers being able to grow and get value out of that. So I sometimes think we – you all may ask the question more as a negative. And for us, it’s just about a healthy cycle at the customer account level.
Michael Turrin: Consistent core cloud growth is still pretty exciting to us as well. Thank you.
Amy Hood: Thank you.
Brett Iversen: Thanks, Michael. Operator, next question please?
Operator: The next question comes from the line of Kirk Materne with Evercore ISI. Please proceed.
Kirk Materne: Yes. Thanks for taking the question. I’ll add my congrats in the quarter. Hey Satya, I was wondering if you could chime in on a discussion that comes up a lot with investors, which is, is there a sort of data quality problem in the market in terms of being able to take advantage of all these new GenAI capabilities? And I was just curious, if you could comment on, do you see companies making inroads on sort of addressing that? And do you see that as sort of an inhibitor to AI growth at all at this point? Thanks.
Satya Nadella: Yes, it’s a great question because there are two sets of things in order to make sense for successful deployment of these new AI capabilities. I mean if you sort of say this, what is this AI, it does two things, right? There’s a new user experience, there is a natural language interface and second thing is it’s the reasoning engine. And the reasoning engine requires good data, and it’s good requires, good data for grounding, right? So people talk about something called retrieval augmented generation. And in that context, having good grounding data that then help with the reasoning, I think, is helpful. And then, of course, people are also looking to sort of fine-tune or RLHF or essentially take the large model and ground it further.
So all of these tools are now available, the sophistication of how to people can deploy these models across various business processes where there is data and where there is tuning of these models is also getting more widespread, even at system integrators and other developers are there to help enterprises. So all that’s maturing. So we feel good. And this is what I think on the commercial side, these are some of the harder problems to solve broad consumer, right? I mean I think this is a couple of orders of magnitude of improvements in, I’ll call it, our models before we can sort of have more sophisticated open-ended consumer scenarios. Whereas in the enterprise, these are all things we can go tackle. Again, I point to get up, if you think about how it’s got an entire system, right?
It’s just not an AI model. It’s the, AI – the user experience, scaffolding, the editor, the chat, then interpreter and the debugger work along with the continuations of the model to help essentially create these reasoning traces, which help the entire thing work. And effectively, what we are doing with Copilot, Copilot Studio and connectors to all these business systems, think of it as we are creating GitHub Copilot like scenarios for every business system. That’s what I think is going to have both what Amy referenced is business value and better grounding. But you’re absolutely right in saying a lot of work we’re doing with Fabric or Cosmos or PostgreSQL is about preparing that data so that it can be integrated with these AI projects.
Kirk Materne: Thank you.
Brett Iversen: Thanks, Kirk. Operator, we have time for one last question.
Operator: Our last question will come from the line of Alex Zukin with Wolfe Research. Please proceed.
Alex Zukin: Hey guys. Thanks for taking the question. I wanted to ask the AI question but from a Microsoft 365 Copilot perspective. I think you talked a little bit about starting to see some of those impacts positively in the quarter on the office business. I wanted to ask more broadly around that capacity constraint that you alluded to in your prepared remarks, Amy. And kind of how does the easing – how tied are we like as you invest for that CapEx and bring more of the capacity online. How much does that unlock or unlock the ability to deliver both a higher Azure AI number as well as a higher Microsoft 365 Copilot number.
Amy Hood: Thanks for the question. It’s a good opportunity to clarify. And I would not say that there is a capacity constraint on the Copilots. It’s a real priority for us to make sure we optimize the allocation of our capacity to make sure that those per user businesses are able to continue to grow. And so think about that as our priority one. And so then what that does mean is capacity constraints when we have them, you’ll tend to see them on the Azure infrastructure side, the consumption side of the business is a better way of thinking about it
Alex Zukin: Perfect. Thank you.
Brett Iversen: Thanks, Alex. That wraps up the Q&A portion of today’s earnings call. Thank you for joining us today, and we look forward to speaking with all of you soon.
Satya Nadella: Thank you all.
Amy Hood: Thank you.
Operator: This concludes today’s conference. You may now disconnect your lines at this time. Enjoy the rest of your day.
