Microsoft Leans on Neoclouds to Ease AI Crunch

Summary of YouTube Video Transcript

Key Concepts: Neo cloud, AI training/inference, capacity constraints, CapEx vs. OpEx, AI models, Nvidia, CoreWeave, Microsoft's AI strategy.

Microsoft's Neo Cloud Investment and Strategy

Microsoft has invested $33 billion in neo cloud infrastructure. This figure, not explicitly disclosed in quarterly earnings, has been pieced together from various sources. The investment is likely much higher than $33 billion. This neo cloud capacity is primarily used for Microsoft to build its own AI models, reflecting a strategic imperative to catch up with OpenAI and Anthropic.

European and Local Players: Microsoft is utilizing European players like End Scale and Nevius, as well as local players like CoreWeave.

Capacity Constraints: Microsoft faces significant capacity constraints and is leveraging every available resource to bring chips online for its customers, itself, and OpenAI.

Neo Cloud Definition and Purpose

A neo cloud is dedicated to AI training, inference, and storage, differentiating it from infrastructure used for running other software.

Microsoft's Internal AI Model Development

Mustafa Suleyman, formerly of DeepMind and now at Microsoft (via the acquisition of Inflection), leads the consumer AI offering. The first large language model built internally under Suleyman is utilizing CoreWeave's assets, which essentially means using Nvidia hardware.

CapEx vs. OpEx and Financial Implications

Renting neo cloud capacity allows Microsoft CFO Amy Hood to avoid classifying these expenses as Capital Expenditures (CapEx).

CapEx vs. OpEx: Purchasing servers requires depreciation and impacts capital expenditures. Renting neo cloud resources allows Microsoft to classify these costs as Operating Expenses (OpEx).

Investor Perspective: Investors prefer a good balance between CapEx and OpEx.

Flexibility: Renting provides flexibility. If Microsoft determines in five years that it no longer needs the capacity from a specific provider (e.g., Gbps 300), it can switch to another provider (e.g., Vera Rubin) without being burdened by depreciating assets.

Quote: "That's a really important point, right? If you buy a bunch of servers now, you have to depreciate them. Now it's on your capital expenditures, not your operating."

The "Funky" Router Rental Analogy

The video uses an analogy of renting a router to illustrate the seemingly unusual situation of a large company like Microsoft renting infrastructure.

Analogy: "It's like if I was paying somebody to write stories and I was still writing stories."

Explanation: This highlights the fact that Microsoft is building its own AI models while simultaneously renting capacity from neo cloud providers.

Conclusion

Microsoft's $33 billion (and likely growing) investment in neo cloud infrastructure is a strategic move to address capacity constraints, build its own AI models, and manage its financial statements by shifting expenses from CapEx to OpEx. This approach provides flexibility and allows Microsoft to adapt to changing AI infrastructure needs.