Wall Street’s verdict has been harsh. Microsoft’s stock lost more than a third of its value between late 2025 and early 2026 as investors questioned whether record AI infrastructure spending could generate the returns the company promised. Copilot adoption lagged expectations, and critics increasingly framed the tech giant as a company spending faster than it was monetizing.
Microsoft, however, drew an opposite conclusion from the market signals. While much of the industry remains focused on models and chatbots, the company is positioning itself for a larger battle. As enterprise software giants such as Salesforce, ServiceNow, SAP, and Workday defend their applications, Microsoft is building the intelligence layer designed to sit above them all. If AI agents become the primary way work gets done, the company believes value will flow not to the application itself, but to the platform that understands, orchestrates, and governs the work across applications. It is a direct challenge to the enterprise software hierarchy.
“We’re seeing a shift from systems of record to systems of understanding,” Arun Ulag, executive vice president of Azure Data, tells me. “The next generation of enterprise software won’t be defined by where data is stored, but how effectively intelligence can access that data, understand it, and help people act on it.”
Enterprise buyers already possess enormous amounts of information fragmented across ERP platforms, collaboration tools, internal databases, and countless specialized applications. Humans spend much of their working lives stitching those fragments together into decisions. AI still struggles to do the same.
“Most organizations don’t have an AI problem, rather a context problem,” Ulag tells me. “What customers are asking now is how to make agents as trusted and productive as their best employees. The challenge isn’t whether an AI agent can generate an answer, but whether it has the same context, institutional knowledge, and understanding of the business that a person develops over years of working inside an organization.”
He noted that companies onboard AI agents worse than they onboard interns. “You don’t hire an employee and tell them to figure everything out on their own. You give them structure, role clarity, the information they need, and the rules of how the company operates. Agents need exactly the same knowledge foundation humans rely on every day.”
The argument is also the foundation for Microsoft IQ, announced at Build 2026. The platform combines three layers of enterprise intelligence: Work IQ, which understands how people collaborate across Microsoft 365; Fabric IQ, which connects agents to trusted business data and metrics inside Fabric and Power BI; and Foundry IQ, which captures the policies, procedures, and institutional knowledge that govern how a company operates.
Build an agent once, the company says, and it should carry the same understanding of the business everywhere it goes. The company’s confidence stems partly from the scale of the context it already controls. According to Ulag, more than 425,000 organizations run Power BI today, while roughly 20 million semantic models live inside Fabric.
“People collaborate through Teams, communicate through Outlook, and share documents through SharePoint. That’s where decisions are made and where knowledge gets created. Now AI agents can participate in that same environment, understand context, and contribute to it. An agent might recognize a recurring pattern, recommend a new policy, or suggest automating a process the next time it occurs. The result is that intelligence begins to compound,” Ulag explained.
Why Enterprise SaaS Should Worry
For two decades, enterprise software companies built their power around a simple premise: own the interface, own the customer. Employees opened Salesforce to understand customers, ServiceNow to manage workflows, and ERP systems to run the business. The application was the destination.
AI agents threaten to make the destination irrelevant. When a user asks a question, the answer can be assembled from ten different systems simultaneously. The customer doesn’t care where the data originated, but rather if the answer is accurate and actionable. Value accrues to whichever platform can understand, orchestrate, and govern work across all of them.
“What’s changing is that customers are no longer asking which application contains the answer. They’re asking how intelligence can work across all of them. That’s a fundamentally different architecture,” says Ulag. “Even internally, the way we work has changed dramatically in a very short period of time. A year ago, product specifications were largely written by people. Today, an agent can create an initial specification by analyzing customer conversations, telemetry, market research, and competitive information.”
Microsoft is entering the enterprise SaaS/AI fight holding cards no rival matches: cloud, security, business applications, and data platforms, all under one roof. Agentic AI lets it fuse those assets into something closer to an enterprise operating system than a product catalog.
“If customers want Microsoft alone, wonderful. If they want Microsoft together with Snowflake, SAP, ServiceNow, AWS, or Databricks, we want those systems to work well together. Customer interest comes first,” Ulag shared. OneLake stores data in open formats such as Apache Parquet, Delta, and Iceberg by default, and Microsoft will virtualize data residing in competing platforms without forcing customers into an upfront migration project.
However, it is also a Trojan horse. Once the data is reachable and the agents are Microsoft’s, consolidation follows. Ulag shared how a large British enterprise customer migrated its data estate to Fabric and reduced projected five-year spending from $165 million to $45 million.
“We’re seeing more customers ask whether they can simplify their architecture, reduce the number of vendors they work with, and consolidate onto a more integrated platform,” he says. “A lot of organizations are under pressure to create budget for AI. Their budgets aren’t doubling or tripling. So they’re asking where they can reduce complexity and redirect resources. That’s creating a significant opportunity for integrated platforms like Fabric.”
For the customer, it was a dramatic efficiency gain. For the broader software market, it illustrates something else: every dollar saved by consolidation is a dollar that previously flowed to another vendor.
The Bet Behind the Biggest Capex in Tech History
Microsoft keeps spending at a pace that unsettles even some of its own shareholders. But from the company’s perspective, those tens of billions are funding a position the company hopes will become far more valuable: the default platform enterprises rely on when AI moves from experimentation to production. Ulag dismisses the idea that the market’s short-term narrative reflects the company’s long-term position.
“Market perception moves in waves. What gives me confidence is Microsoft’s commitment to innovation across the full stack. Whether it’s building the AI supercomputers that power not only our own models but OpenAI and many others, whether it’s innovation at the Azure software layer. You can see it all the way down to our silicon strategy with Maia and Cobalt, and even into areas like quantum computing,” says Ulag. “Very few companies can innovate across that entire stack, and customers are increasingly seeing the value of having those pieces work together.”
As companies deploy AI agents across sensitive workflows, customer data, intellectual property, compliance requirements, and internal knowledge bases become strategic assets rather than simple inputs. Ulag believes that dynamic will ultimately favor platforms capable of preserving enterprise control rather than absorbing enterprise data.
“The customer’s IQ is the customer’s intellectual property. Customers want AI systems that can leverage that intelligence, but they don’t want it leaking to another vendor, another model provider, or another organization. We believe the future of enterprise AI depends on helping customers activate their own intelligence while maintaining ownership, governance, and control,” Ulag explained. “That’s a fundamentally different approach than treating enterprise knowledge as something that should be centralized somewhere else.”
Salesforce, ServiceNow, SAP, Workday, Oracle, and a wave of AI-native startups are all pursuing rival versions of this future, and none of them intends to surrender the workflow layer quietly. Microsoft is betting, fearlessly and in the open, that the crown goes to whoever controls where that intelligence works.
The Control Layer Verdict
Microsoft no longer wants to be the cloud behind the AI revolution. It wants to be the layer enterprises depend on after the models are deployed: the place where intelligence gets used, governed, and billed. The gambit is ruthless because it looks collaborative — consolidation disguised as choice, openness as a moat. Microsoft frames it as customer-first, multi-vendor-friendly, and technically neutral about which model powers the agents. All three can be true and still concentrate more decision-making authority in one platform than the SaaS era ever did.
