In a recent in-depth interview, Microsoft CEO Satya Nadella acknowledged that in the AI era, Microsoft is reclaiming control over the foundational layer through its own large language model development. He explicitly rejected the idea of selling compute power for quick revenue, and predicted that traditional SaaS will fully transition to a hybrid business model combining subscriptions with usage-based billing.
Following the conclusion of Microsoft's annual Build developer conference on June 4th, Nadella participated in a detailed podcast interview with Ben Thompson, founder of the prominent tech business analysis outlet Stratechery.
Facing the dramatically reshaped AI competitive landscape since 2018, the market is highly focused on the direction of Microsoft's massive AI infrastructure capital expenditures, the order and capacity constraints of its Azure cloud business, and how the company will maintain its profit margins and growth narrative in the global AI race.
Nadella did not shy away from the market's most pressing and sensitive topics. Addressing external criticism that Microsoft had "fallen asleep" and become passive due to over-reliance on its partnership with OpenAI, Nadella conceded the competitive environment has changed drastically. "There's OpenAI, there's Anthropic, there's Google, there's a lot of people who have jumped in. We are now competing with a set of people whose names I didn't even know in 2018."
To build a more controllable moat, Microsoft unveiled its self-developed MAI model, built from scratch with a "clean lineage," at the conference. Nadella offered a market-oriented explanation: "Any company has to build a learning machine. What I want to build is essentially a multi-tenant learning system that allows everyone to have their own hill-climbing machine."
Microsoft is attempting to develop a dual-line capability—retaining access to OpenAI's top-tier technology assets while also possessing a fully end-to-end controllable proprietary model. This aims to secure pricing power and control over compute resource allocation in future enterprise services.
Strategic Allocation of Compute, Rejecting "Easy Money"
In January, Microsoft's Azure cloud revenue experienced a slight deviation from expectations due to compute capacity bottlenecks, raising Wall Street concerns about potential strategic missteps in capital expenditure. In this interview, Nadella clarified the company's "triple allocation logic" when compute supply is constrained: hyperscale cloud customers, its own high-margin application businesses, and internal R&D compute.
Confronted with the reality of compute scarcity, Nadella sent a clear signal of commercial discipline to Wall Street: Microsoft will not sacrifice its long-term moat for short-term accounting revenue.
"We made some very hard choices, for example, we are not selling raw GPUs to a bunch of Neolabs," Nadella stated bluntly. "In this day and age, if you want to get short-term Azure revenue, it's very easy. You just go ahead and sell compute to those labs for 'easy money.' But we are not doing that."
Nadella emphasized that the hyperscale cloud business cannot rely solely on a single model company as its primary customer, as these frontier labs will eventually build their own hardware infrastructure. Therefore, Microsoft will allocate compute with extreme restraint, prioritizing the needs of large enterprise customers and directing funding and compute towards internal MAI model development and high-profit proprietary applications like GitHub and M365.
Software is "Not Dead," but Traditional SaaS Business Model is at an Inflection Point
As AI evolves from "Copilot" to "Agents," the market has debated whether traditional software has reached its end. Nadella provided a clear stance: "Software is still alive, but we need to re-architect it for the agent era."
Nadella pointed out that the traditional SaaS model, which bundles data, business logic, and UI into a single subscription, has become obsolete. In the agent era, with numerous AIs continuously calling underlying data 24/7, the marginal cost of compute is skyrocketing.
"When Cowork uses WorkIQ, that's going to be a consumption-based business model. So, what needs to happen is we now need to take what we built, re-architect it for the agent era, and change the business model lever so you have a per-user business model and a consumption business model," Nadella stated. "100% that's the future."
He insightfully noted that in the PC era, software companies relied on hardware performance gains from Moore's Law to mask software bloat. However, inference costs in the AI era are extremely high; without model optimization, software costs will explode. This means future enterprise SaaS customers will become extremely pragmatic: "No customer will consume their allowance or their subscription if it's not creating value for them."
Future Vision: AI Agents as New Applications, Reshaping the Endpoint Ecosystem
Regarding Microsoft's future growth potential, Nadella placed his bets on "Agents." He revealed that the company's future differentiation will not merely be providing infrastructure, but "building agents on top of that infrastructure," with a focus on three large-scale areas: coding, security, and knowledge work.
The interview also shed light on Microsoft's ambition, through projects like Project Solara, to break the existing monopoly of smartphone devices over user access. Nadella envisioned a future of "unmetered intelligence," where native agents run on Windows or edge devices to offset high cloud compute bills.
"The biggest value proposition for a Windows machine in an enterprise setting will be unmetered intelligence. People will say, 'Wow, instead of my cloud bill just going up, I'll own a Windows machine and amortize it that way,'" Nadella added. This combined edge-cloud approach is not only a means for Microsoft to control compute costs but also a core component of its strategy to reshape distribution channels and sustain long-term revenue growth in the AI era.
Comments