Satya Nadella

Nadella said Microsoft's total headcount cost will fall while cost per employee and GPU allocation per researcher rise.

"Our total people cost will go down," he said. "Our cost per head will go up. And my GPU per researcher will go up."

Nadella said Maia 200 already delivers up to 1.7x more tokens per watt and Maia 300 will deliver 40% more tokens per watt than leading GPU systems.

Maia 200 is already delivering up to 1.7 times more tokens per watt across AI experiences in Teams and Copilot. Maia 300 will drive further gains, delivering 40% more tokens per watt compared to leading GPU-based systems.

Nadella said that as AI agents take on higher-value work, consumption-based pricing will unlock business-funded demand tied directly to the value agents deliver, beyond traditional IT-funded budgets.

"Typically, IT funds this, consolidating frontier advances for everyone across the organization," Nadella said of the changing spending models within enterprises. "As agents take on higher-value work, consumption opens up business-funded demand with spending tied directly to the value these agents deliver."

Nadella described Microsoft Execution Containers (MXC) as a policy-driven layer that governs and contains AI agents across Windows and other platforms, with OpenAI, Anthropic and Muse adopting it.

a policy-driven execution layer that governs and contains agents across Windows and other platforms.

Nadella said security grows even more critical as agents gain access to enterprise data, and Microsoft's approach is to separate intelligence from authority, enforcing permissions, monitoring and containment outside the model.

separate intelligence from authority

Satya Nadella highlighted that users paying for Copilot features on a usage basis grew 75% this quarter, signaling a shift toward usage-based monetization.

CEO Satya Nadella emphasized on the earnings call that the number of users paying for Copilot features based on usage increased by 75% this quarter

Nadella said Microsoft unified Chat, Cowork, Autopilot and Code into one Copilot and views it as the new OS for work.

We announced our biggest update to Copilot to date this quarter, bringing together Chat, Cowork, Autopilot and Code into one unified experience. It's one product experience across commercial and consumer and comes with all of Office built in. We think of Copilot as the new OS for work.

Nadella said the business model is expanding from per-seat subscriptions to include consumption, growing TAM and ARPU.

Our business model is expanding from per seat subscriptions to include consumption, expanding our TAM and ARPU. The subscription brings increasingly capable everyday intelligence to every user. Typically, IT funds this, consolidating frontier advances for everyone across the organization.

Nadella said agent-generated pull requests on GitHub rose 6x in six months and agents now open over a third of all PRs.

Agent-generated PRs have increased 6x over the past 6 months, and agents now open more than 1/3 of all PRs on the platform. We are very hard at work on reliability as agentic activity on GitHub soars.

Nadella said Microsoft aims to match leading models within 3-6 months at a fraction of the training cost via its first-party MAI models.

Our goal is to advance the cost-to-outcome frontier, which means delivering capabilities comparable to leading models within 3 to 6 months of their release at a fraction of the training cost, thereby allowing us to transfer those cost benefits to customers.

Nadella said Microsoft's agent security approach separates intelligence from authority, enforcing controls outside the model.

Our approach is to separate intelligence from authority, enforcing identity, permissions, monitoring and containment outside the model itself.

Nadella said local models on new PCs make intelligence effectively unmetered, letting laptops run last year's frontier models.

As more capable models run locally, we see an opportunity to make it more intelligence effectively unmetered. ... With these new devices, you can run last year's frontier models on your laptop and last quarter's model at your desk.

Nadella likened enterprise AI workloads to past database applications, saying shiny-object enthusiasm is waning in favor of real systems.

it does feel like the enterprise workloads. They feel like to me, like the database applications enterprises built in the past that I grew up with is happening now when it comes to AI. And in fact, maybe in some of the shiny object stuff is waning

Nadella said he prefers broad, decade-long enterprise demand over growth reliant on a few large capacity buyers.

the enterprise stuff is not like a quarter-over-quarter. It's a decade-long thing, and that's what I've been waiting for as opposed to like 4 customers who buy a lot of capacity from you and you can show a few quarters of growth. And that's not a business as far as I'm concerned. So I want the breadth.

Nadella said model routers let frontier labs like Anthropic and OpenAI compete for every task, delivering frontier performance faster and cheaper.

you want, in fact, Anthropic and OpenAI, by the way, to compete for your workload, not like once a quarter, but for every task. And that, to me, is what the value of these routers and auto fusion stuff routing does

Nadella said consumption-based monetization will ramp faster in coding than elsewhere because developers are used to pay-as-you-go.

I think on the coding side, it will be faster just because I think people are used to paying on this PayGo as the sort of the general market and how it works.

Nadella said he has never been more confident in long-term returns on Microsoft's AI investments.

In closing, based on the demand we see, the strength of our products and the efficiency we are driving across the stack, I've never been more confident in the long-term returns for our investments.

Nadella says accelerating demand drove Microsoft to add 24 data centers across five continents in the quarter, using a fungible fleet to allocate capacity to the highest-value uses.

We are building against a growing book of business with a fungible fleet that allows us to dynamically allocate capacity to the highest value opportunities across compute, data, foundry and co-pilot. Accelerating demand led to us adding 24 new data centers across five continents this quarter.

Nadella says full-stack optimization across data center design, silicon, software and models raised Copilot throughput by over 50% quarter over quarter.

At the same time, we continue to get more from the infrastructure we already have by optimizing across the data center design, silicon systems, software and models. For example, we increased the throughput for co-pilot workloads by over 50% quarter over quarter.

Nadella says Microsoft is helping organisations build and own their own intelligence and earning trust through agent safety and security, turning innovation into growing adoption across every industry.

By helping organisations build and own their own intelligence, and earning trust through agent safety and security, we are translating our innovation into growing adoption and real-world impact across every industry,

Nadella says Microsoft is delivering more capability and cost efficiency across its stack, helping organizations own their intelligence and building trust via agent safety and security, driving adoption across industries.

We are delivering greater capability, choice, control, and cost efficiency for customers across our entire stack," said Satya Nadella, chairman and chief executive officer of Microsoft. “By helping organizations build and own their own intelligence, and earning trust through agent safety and security, we are translating our innovation into growing adoption and real-world impact across every industry.

Nadella says he no longer believes alignment will emerge naturally with scale and now sees it as a harder, distinct problem.

I used to always think that alignment will just fall out. I no longer believe that. So I think this alignment thing may be a harder, different problem,

Nadella argues reinforcement learning alone cannot achieve alignment.

I don’t think you can even RL your way to alignment,

Nadella says more capable but less aligned agents require a reliable zero-trust style system around the model.

If you really want to have agents that are more capable and yet less aligned, then you need to have a trust system that you can rely on,

Nadella warns chain of thought can be unfaithful and that labs hiding it makes the problem worse.

We know that the chain of thought itself can be unfaithful, and in fact it can be made worse when the model company says, ‘Well, it’s unfaithful and I’m going to hide it,’

Nadella warns an unaligned open-weight Chinese model could end up attacking even China.

An unaligned open-weight Chinese model will probably attack even China,

Nadella said the central question of the Super Intelligence era is delivering real-world outcomes, not just the size of the opportunity.

For Microsoft, the central question as we enter this Super Intelligence era is not just how big the opportunity becomes, but how we deliver the real-world outcomes that every country, community, and company demands and deserves,

Nadella said the PHP vulnerability crisis was unlike anything he had seen in a long time, noting the heavy toll on IT and security teams.

I’ve not seen anything like this in a long, long time.” “It was a long day and a long night, I know, for all the IT and security folks,”

Nadella said he had not seen anything like the PHP vulnerability incident in a very long time and stressed the need for software that can be securely maintained as systems grow more complex.

"I've not seen anything like this in a long, long time," Microsoft CEO Satya Nadella said at an event in Germany on Thursday, saying the incident showed the importance of building software that can be maintained securely as systems grow more complex.

Nadella said the PHP vulnerability crisis was unlike anything he had seen in a long time, underscoring the need for digital systems that can be maintained securely.

It was a long day and a long night, I know for all the IT and security folks ... I’ve not seen anything like this in a long, long time,

Nadella says inference makes marginal cost central to software economics, so customers will weigh token costs against outcomes, a 'cost-to-outcome' frontier Microsoft is optimizing.

As a result, customers will increasingly measure the cost of tokens against the outcomes they produce

Nadella says Microsoft is building a platform- and partner-first frontier ecosystem in which others capture more value than Microsoft itself.

...we are building a frontier ecosystem that enables others to build more value on top of and through the ecosystem than is captured inside.

Nadella expresses confidence in long-term returns on AI infrastructure investment, citing demand, efficiency gains and disciplined investment choices.

Based on the demand we see and the efficiencies we are driving, I am confident in the long-term returns on my investments. That confidence starts with being disciplined about what we build and who we build it for.

Nadella says scaling laws have worked far better than he imagined in 2018, though he sees them as an empirical finding and says alignment remains elusive.

The good news here is the scaling laws have worked and worked magically a lot better and a lot more than I would have ever imagined in 2018

Nadella warns that advanced AI systems inside organizations can act like insider threats, so companies need security, monitoring and control layers outside the models.

You have effectively these intelligences that are inside that have all the capability to be an insider risk

Nadella argues that while everything may look like a commodity, nothing at scale truly is, in response to questions on AI model commoditization.

At the end of the day, everything is a commodity and nothing at scale is a commodity.

Nadella warned that scaling laws are yielding more powerful AI models that are increasingly less trustworthy, so businesses should treat AI systems as new forms of insiders and build safeguards around them.

Our scaling laws are producing even more powerful models that are less and less trustworthy

Nadella said businesses must keep humans in the loop so AI strengthens rather than undermines employees' sense of empowerment and ability to work.

How do you put the human in the loop? How do you really ensure that, ultimately, that sense of empowerment is not lost?

Nadella argues enterprises cannot rely on AI model providers' safety assurances and must separate the supply of intelligence from authority over it via independent oversight systems.

We simply can’t outsource responsibility for what intelligence does on our behalf. A model provider’s assurances do not relieve us of that responsibility. … In other words, we need to separate the supply of intelligence from the authority over it.

Nadella warns that enterprises are granting complex AI agents access to their most sensitive data and authority over mission-critical actions, likening them to potential insider threats.

We are deploying these complex agentic systems and models, with access to our most sensitive data and giving them the ability to take mission-critical actions on our behalf!

Nadella, as quoted by Microsoft's Nadim Abdo, argues that AI agent security controls must sit outside the model and that intelligence supply must be separated from authority over it.

We should not ask a model to be its own security authority. The controls that determine what it can access, what actions it can take, and whether it should continue operating must sit outside the model, where the model cannot bypass or tamper with them.

Nadella announced new Microsoft Sovereign Cloud capabilities in Paris, aimed at giving organizations more choice and control over sensitive workloads across public and private cloud.

Today in Paris, we announced new capabilities across Microsoft Sovereign Cloud, expanding choice and control for organizations running their most sensitive workloads across public and private cloud, and our growing partner ecosystem.

Nadella argues that trustworthy agentic systems come from architecture that minimizes reliance on trusting the model, not from having the most trusted model.

The most trustworthy agentic system will not be the one with the model we trust most. It will be the one designed that enables us to trust the model least.

Nadella argues AI models should be treated as insider risks with security controls outside the model, and says Microsoft is expanding Project Perception to continuously assess agent risk, adapt permissions and contain threats.

I recently wrote about why we need to treat models as insider risks, with security controls sitting outside the model itself. Today, were putting that principle into practice by expanding Project Perception with new capabilities to help continuously assess agent risk, adapt permissions, and contain threats.

Nadella argued that AI systems with access to sensitive data should be treated as potential insiders, with controls and testing independent of the model, and that users cannot rely only on provider assurances.

Last week, Microsoft’s Satya Nadella argued that AI systems with access to sensitive data should be treated as potential insiders, with access controls, logs and testing kept independent of the model. He also said that we couldn’t rely solely on model providers’ assurances and must assume more responsibility.

Nadella warns the industry to wake up and focus on making AI work reliably rather than chasing shiny objects and FOMO.

We, as an industry, have to wake up to this. It’s not about the shiny object and FOMO.

Nadella argues that early success counts for little in this industry, citing Microsoft's 51-year history while calling frontier labs magical businesses.

I look at them and say, hey, this is not an industry where your initial success means anything

Nadella says the industry must wake up to the fact that making AI work is not about chasing shiny objects or FOMO.

"We, as an industry, have to wake up to this," Nadella said. "It's not about the shiny object and FOMO."

Nadella argues that initial success means little in this industry, citing Microsoft's 51-year history when assessing frontier labs.

"I look at them and say, hey, this is not an industry where your initial success means anything," he said.

Nadella argues every frontier AI model should be treated as an insider risk, assumed compromised, and contained from the outset with an emergency brake outside the model.

We must assume a model is compromised and contain it from the start.