DiscussionInsightful

An Interview with Microsoft CEO Satya Nadella About Finding Core Competencies

Stratechery's Ben Thompson interviews Microsoft CEO Satya Nadella at Build 2026, covering Microsoft's competitive repositioning in AI, the evolving OpenAI partnership, MAI model ambitions, and the future of software and enterprise computing. Nadella emphasizes Microsoft's focus on building a multi-tenant 'hill-climbing' platform for enterprises rather than competing directly on frontier models alone. Key topics include capital allocation discipline, GitHub Copilot's challenges, Project Solara, and Microsoft's core competency identity.

Summary

The interview opens with Thompson asking Nadella whether he is satisfied with Microsoft's current competitive position. Nadella reframes the question away from zero-sum competition, arguing that Microsoft's opportunity lies in identifying what it is uniquely capable of doing — acting as a trusted platform provider that enables others to create value. He articulates that Microsoft is at its best when it does what the world expects of it, citing the Zune as a cautionary tale of pursuing competitors out of envy rather than genuine capability.

Nadella introduces the concept of 'hill-climbing machines' — the idea that every enterprise needs its own AI learning system that continuously improves against private benchmarks and evals. He argues that a firm's moat in the AI era comes from its tacit knowledge encoded in private evals and reinforcement learning environments, not just API access to frontier models. This framework underpins Microsoft's MAI model strategy: building models with clean lineage that enterprises can deeply customize through their own RL environments, going far beyond surface-level RAG or fine-tuning.

On the OpenAI partnership, Nadella expresses pride in what was built but acknowledges the need to develop independent model capabilities. He notes that Microsoft still has a strong commercial relationship with OpenAI through 2032 but frames MAI models as a necessary parallel track, using OpenAI IP (including reverse knowledge distillation) while building toward model independence. He draws a parallel to the Microsoft-Intel relationship and the Microsoft-SAP relationship as precedents for productive but not exclusive partnerships.

Regarding capital allocation and CapEx, Nadella pushes back on the narrative that Microsoft underinvested in AI infrastructure. He explains a three-bucket framework — hyperscale Azure business, internal application compute (M365, GitHub, Security), and research compute for MAI — and defends the discipline of not chasing 'easy money' by overselling raw GPU capacity to neo-labs. He acknowledges the awkward optics of missing Azure guidance by 0.1% when prioritizing internal compute, but frames it as a necessary long-term tradeoff.

On the software business model, Nadella argues software is not dead but must be rebuilt for the agent era with a hybrid per-seat and consumption pricing model. He sees E7 and bundling as tools to help enterprises budget for consumption while capturing agent value. He notes that outcome-based pricing is essentially royalties — enterprises resist sharing upside — so consumption pricing with optimization incentives is the more realistic model. The MAI/enterprise hill-climbing framework reinforces customer optimization behavior.

GitHub Copilot's competitive challenges are addressed candidly. Nadella acknowledges that the shift to agentic coding — driven largely by Anthropic's model capabilities — caught Microsoft flat-footed, and credits Cursor as a competitive wake-up call, though he contextualizes it historically (Borland vs. Microsoft). He commits to multi-model support in GitHub Copilot including Anthropic, OpenAI, MAI, and open-weight models.

Project Solara is framed as Microsoft's attempt to build a platform for ambient, enterprise-grade agentic devices — not Windows-centric, but an open platform for ODMs to build agent-access hardware for healthcare, enterprise, and other verticals. Nadella explicitly distances Solara from a Windows strategy, framing it as a new platform paradigm suited for the agent era where the center of gravity is cloud agents, not local compute.

The interview closes with a discussion of data center community impact. Nadella agrees that the tech industry must earn the right to build infrastructure by creating real community value, and acknowledges the job displacement concern as the core tension the industry must address more seriously.

Key Insights

  • Nadella argues that Microsoft's competitive advantage lies not in matching frontier model labs but in building a multi-tenant 'hill-climbing' platform that allows every enterprise to develop its own private AI learning system using their own evals, trajectories, and reinforcement learning environments — making the enterprise's tacit knowledge the true moat.
  • Nadella acknowledges that the agentic coding shift driven by Anthropic's model approach genuinely caught Microsoft off guard with GitHub Copilot, framing it not as a Cursor problem but as a fundamental change in what coding assistance meant — from completions and chat to full agent loops.
  • Nadella defends Microsoft's CapEx restraint by explaining a deliberate three-bucket allocation strategy (hyperscale Azure, internal application inference, MAI research) and argues that chasing short-term Azure revenue by selling raw GPUs to neo-labs would undermine long-term enterprise trust and ROIC discipline.
  • Nadella frames outcome-based AI pricing as functionally equivalent to royalties, arguing enterprises are unwilling to share upside, and therefore predicts the dominant enterprise AI pricing model will be hybrid per-seat bundles (for budget predictability) plus consumption for overflow — not pure outcome-based contracts.
  • Nadella claims that Microsoft intentionally did not over-allocate compute to OpenAI's exclusive use, arguing it was 'clear as day' that frontier labs would eventually build their own infrastructure, and that building a hyperscale business dependent on a single model tenant would be structurally unsound — framing the partnership recalibration as foresight rather than reaction.

Topics

Microsoft's core competency and competitive identityHill-climbing machines and enterprise AI customizationMAI models and independence from OpenAICapital allocation and CapEx disciplineHybrid per-seat and consumption software pricingGitHub Copilot and agentic coding competitionProject Solara and ambient enterprise devicesData center community impact and social license

Full transcript available for MurmurCast members

Sign Up to Access

Get AI summaries like this delivered to your inbox daily

Get AI summaries delivered to your inbox

MurmurCast summarizes your YouTube channels, podcasts, and newsletters into one daily email digest.