Key Takeaways

  • SaaS business models are undergoing a "re-litigation," forcing companies to unbundle traditional fixed data models, logic, and UI, then rebundle them for new value creation.
  • Agentic AI can unlock 10x more value from a company's "most important database"—often previously captive to specific apps—by exposing it to intelligent agents, as seen with Microsoft's Work IQ.
  • The shift means moving beyond per-user pricing, an "artifact" of budget certainty, towards consumption-based models, despite initial founder reluctance to give away "royalty" on outcomes.
  • Companies should reconsider what to build internally versus what to acquire, especially as 'agent euphoria' drives new development, using a clear marginal cost framework.
  • Apply Satya Nadella's Build vs. Buy Principle for AI Applications to critically evaluate your development and acquisition strategies in the age of agents.

The Satya Nadella's Build vs. Buy Principle for AI Applications

Satya Nadella suggests a clear rule for founders grappling with whether to develop AI applications in-house or acquire them. This framework cuts through the hype, focusing on the cold economics of building versus buying.

Principle: You should always acquire something if the marginal cost of building and maintaining something on your own is higher.

Consideration: Building Cost: The resources, time, and effort required to develop the application internally.

Consideration: Maintenance Cost: Ongoing expenses and effort for security, updates, bug fixes, token burn, and support.

Comparison: Compare the total marginal cost of internal development and maintenance against the cost of acquiring a vendor solution.

Outcome: If internal costs are higher, acquire. If external costs are higher, build.

When This Works (and When It Doesn't)

Nadella's principle is directly applicable when companies face what he calls "agent euphoria" – the excitement to build internal AI applications without fully considering the long-term commitment. It's particularly useful for established enterprises with existing data infrastructure, where a clear cost comparison against an external solution is possible and the goal is optimization rather than inventing a core product category.

This framework shines when the 'something' in question is a utility or a feature that, while important, isn't central to your company's unique value proposition or secret sauce. However, it can fall short for early-stage startups where speed to market, rapid iteration, and building proprietary core IP might initially outweigh a purely marginal cost analysis. If the "build" is your competitive differentiator or creates network effects, the long-term strategic value might justify a higher marginal cost in the short term.

What to Do With This

Imagine you're a founder of a B2B SaaS startup building a customer support automation platform. Your product needs a robust natural language understanding (NLU) module to accurately categorize user queries. You're debating whether to build your own custom NLU model from scratch or integrate a leading third-party NLU API.

Apply Nadella's framework this week:

1. Principle: Should you acquire the NLU API if building and maintaining your own NLU is pricier?

2. Building Cost: Estimate the engineering hours for research, data labeling, model training, and integration. Factor in the cost of hiring or upskilling AI talent. Let's say it's $150,000 in upfront salaries and GPU compute.

3. Maintenance Cost: Project the ongoing expenses: monitoring model drift, retraining with new data, security patches, API updates, and the per-inference cost (token burn) if using your own GPUs. Estimate $5,000 per month.

4. Comparison: Get quotes from 2-3 leading NLU API providers. Calculate their projected monthly consumption costs based on your expected query volume, plus any setup fees. Let's say it's $2,000 per month with a $10,000 setup fee.

5. Outcome: Compare the total cost over, say, three years. If your internal build is $150k + ($5k 36) = $330k, and the API is $10k + ($2k 36) = $82k, the decision is clear: acquire the API. This frees your engineering team to focus on your core competitive advantage, not foundational AI infrastructure that's already commoditized.