In the tech-driven world we live in, we often hear about Artificial Intelligence and Cloud Computing, but have you ever wondered about the infrastructure that powers it all? Have you wondered what powers Chat GPT, Azure, and AWS? I will try to explore this complex web of ‘data centers’, the backbone of cloud computing, and also share my thoughts on a different way for tech companies to manage them.
The Cloud: What the heck is it?
Explaining Cloud Computing
I am sure you have heard the term “cloud,” but what does it really mean? In simple terms, cloud computing is like renting digital space and computing power instead of owning and maintaining physical servers. It’s where your photos, documents, and favorite apps are stored and processed.
Behind the Scenes: Data Centers
Have you ever wondered where this “cloud” exists? It’s in massive data centers, vast warehouses filled with servers that process, store, and manage all the data we generate. These data centers are the unsung heroes making cloud services possible.
So, when you come across terms like AWS, Azure, or Google Cloud, what it essentially refers to is the computing infrastructure owned by Amazon (for AWS), Microsoft (for Azure), or Google (for Google Cloud). This compute infrastructure is located within massive warehouses and buildings exclusively owned by these tech giants.
The big debate: Should tech companies own or lease the hardware (in the data centers)?:
Companies like Google, Microsoft, Facebook, Amazon, and others (let’s call them big tech) invest billions in building and maintaining these data centers. While it is crucial for their cloud business, the substantial capital expenditure required poses a significant barrier to capital allocation.
To provide some context, the leading 20 tech companies/hyper-scalers collectively invested USD ~150 billion in data center capEx in 2021. Microsoft allocated around USD ~11 billion for data center capEx in the first quarter of 2023. This trajectory is poised to surge exponentially as the demand intensifies in the ongoing AI race and the broader wave of digitization.
The big question at the core of this discussion revolves around the staggering amount of cash that tech giants are pouring into data center capital expenditures (CapEx). It appears to be a cash-burning engine, challenging the very essence of the asset-light software model in which big tech companies typically thrive.
I argue that while big tech should continue to invest in the software layer, hardware-software integration, and design elements, specialized entities should own the hardware. From a scaling perspective, the current model poses challenges, requiring significant capital investment to expand data center capabilities.
Airline Analogy
Let us think of it like airlines. They don’t own all the planes they fly; they lease them. This allows airlines to focus on what they do best – flying planes – without tying up capital in aircraft ownership. What if tech companies could do something similar with their data centers?
Airlines don’t own their airplanes; they lease them.
Similar to how airlines lease planes from low-risk capital sources, I suggest separating hardware ownership from operational control. Tech companies should retain control and innovation, while low-risk capital handles the infrastructure.
Deconstructing data centers: Unpacking the logic behind ownership restructuring
Sometimes it is easy to look at things by breaking them apart. Data centers can be broken up into five key areas:
| Area | Details | Current state | Proposed state |
| Facility | The physical location housing the data center | Owned | On Lease |
| Hardware | Servers, Racks, RAM, GPUs, CPUs, and other IT assets | Owned | On Lease |
| Design | Configuration and arrangement of hardware, custom silicon/chips/hardware design, and its linkage with software | In-house | In-house |
| Software layer | Integration layer connecting all hardware components into a cohesive cloud offering. Includes models/software for predictive maintenance etc. | In-house | In-house |
| Operations | Day-to-day management, covering aspects like physical security, utility management, and asset replacement | In-house | No preference |
The key source of competitive advantage of the data center success (a.k.a. cloud business for the big tech) lies in the design and the software layer (considering hardware* being a commodity). Rest of the elements are pretty standard across any cloud platform. So owning the design and software layer in-house seems like the right move.
*companies also create specialized silicon/chips for their data centers but I am classifying it under design as it the advantage comes from design of the chips/custom silicon and not the ownership of that once it is put in the data center.
The Proposed Model
Balancing Control and Efficiency
The proposal isn’t about abandoning control but redefining it.
- Tech companies can actively seek entities interested in deploying low-risk capital, such as fixed-income type funds/REITs.
- Tech companies retain ownership of patents, design, and research, but the hardware assets are owned by a separate, low-risk entity.
- By establishing long-term agreements, tech companies can maintain control while an external entity takes ownership of the data center hardware.
It’s like having the cake and eating it too. It is a win-win situation. Low-risk capital discovers a stable investment avenue, and tech companies channel their resources into their expertise: software and design.
Ensuring Integration and Innovation
Despite the separation, I emphasize that the synergy between hardware and software should remain intact. It’s not about dismantling what works but optimizing the system for future scalability and innovation.
Think of the new asset class it opens up for the low-risk investors/fixed-income funds. Envision trillions of dollars of capital finding a purpose in constructing the next era of data centers. I am mesmerized by the possibilities it can open up for the world.
Overcoming Concerns
Addressing loss of control and security: Some might worry about loss of control and security. However, agreements between tech companies and hardware owners can ensure continued control and robust security measures.
Economies of scale and cost considerations: Contrary to concerns about a loss of economies of scale, I believe it could enhance economies by unlocking more global capital for data center investments. From a cost perspective, it’s positioned as a cost-efficient model, allowing tech companies to focus capital on high-growth areas.
Rapid pace of hardware change: The swift evolution of hardware is a common concern. Many point out the rapid upgrades, especially with CPUs giving way to GPUs, and the constant evolution of GPUs. In a landscape marked by these dynamic shifts, how can the proposed model be sustained? The straightforward response lies in the continued use of older hardware. Despite the pace of hardware advancements, the reality is that even the older hardware is still actively employed. So there is no change with the demand/usage with change in asset ownership.
As we venture “Beyond the Clouds,” we discover the intricate world of data centers and the potential for a new model that balances innovation, control, and scalability. In the age of AI and increasing computing demand, it’s a discussion worth having as we chart the future of technology.
Disclaimer: https://vinaysachdeva.com/disclaimer/. The opinions expressed in the blog post are my own and do not reflect the view(s) of my employer.