Browsers and search engines were at the core of monopoly discussions in the early days of the internet. The question feels more fundamental now. Compute belongs to whom?
It’s no longer about constructing a server warehouse in Oregon or owning a data center in Northern Virginia. These days, control entails more than just managing the complete vertical stack, from software orchestration and cloud infrastructure to chip design and manufacture. The fabric that connects the silicon is just as powerful as the silicon itself.
| Category | Details |
|---|---|
| Core Question | Who controls the AI compute stack? |
| GPU Market Share | Nvidia: ~88%–95% of AI chip market (2025) |
| Critical Manufacturing | ASML: 100% monopoly on EUV lithography machines |
| Cloud Market Share | AWS, Azure, Google Cloud: ~63% combined |
| Key Players | Nvidia, Amazon, Microsoft, Google, Meta, OpenAI, Oracle, xAI |
| Emerging Trend | Decentralized compute networks (Akash, Render) |
| Reference |
You can practically feel the stakes when you go inside a hyperscale data center. Electricity hums in long hallways of metal racks. Perfect rows of blue LEDs flash. Via vents, the air is rapidly cooled, resembling a mechanical heartbeat. This cloud computing isn’t abstract. Its infrastructure is costly and heavy.
Nvidia is at the base of this pyramid. By the end of 2025, Nvidia will hold between 88% and 95% of the market for AI GPUs. Dominance is not a coincidence. Its moat is CUDA, the software environment on which developers have become reliant, rather than just hardware. It becomes expensive and dangerous to switch once you have built your models in CUDA.
Nvidia may have a psychological monopoly in addition to a technical one. Developers have faith in it. Investors have faith in it. Thus, the loop perpetuates itself.
However, Nvidia doesn’t work alone. ASML, a Dutch company with an almost 100% monopoly on Extreme Ultraviolet lithography machines, is the driving force behind every cutting-edge semiconductor. There would be no sub-5nm chips without EUV. There would be no frontier AI without such chips. This is the epitome of vertical integration.
The cloud landlords follow. Together, Google Cloud, Microsoft Azure, and Amazon AWS hold around 63% of the world’s cloud market. They are gatekeepers, not just providers of infrastructure. You probably rent compute from one of them if you wish to train a large-scale AI model.
Last year, as executives presented predicted GPU allocations for AI clients in a glass-walled meeting room in Seattle, a quiet power dynamic was evident. Compute access is limited. It is negotiated. It’s a calculated move.
These businesses are not satisfied with simply renting Nvidia’s chips. They are creating silicon on their own. TPUs from Google. The Inferentia of Amazon. internal AI accelerators at Microsoft. The effort to lessen dependency is costing billions of dollars.
This seems to be more about survival than it is about competition. A new computing aristocracy is forming above them. Oracle, xAI, OpenAI, Microsoft, Alphabet, and Meta are more than just clients. In complex cycles, they partner, co-invest, and occasionally compete. OpenAI is financed by Microsoft and operates on Azure, which reimburses Microsoft for its computational expenses. It’s recursion and capitalism.
Cracks are still showing up. Decentralized alternatives have been fuelled by the cost and scarcity of GPUs. Distributed compute—a sort of “supercloud” built from idle capacity worldwide—is promised by networks like Akash and Render. It sounds like a utopia. It might also be practical.
It’s still unclear if these decentralized approaches can equal hyperscale reliability. However, they stand for resistance to focus.
Meanwhile, everything is complicated by geopolitical tensions. China and Europe are aggressively attempting to lessen their reliance on the massive cloud providers based in the United States. There is increasing regulatory pressure. Shape chip availability is controlled by export. Like telecom networks or oil pipelines, computation has evolved into strategic infrastructure.
It feels like the beginning of a new industrial age to watch this happen. We argue GPUs instead of railroads. We examine fabric barons, the businesses in charge of the optical interconnects that connect thousands of chips, rather than steel barons.
Who actually “owns” computing is still up for debate. For the time being, Nvidia is in charge of the chip bottleneck. The distribution is managed by the cloud hyperscalers. ASML is in charge of manufacturing capacity. Additionally, power is shared across tiers by the top oligarchs.
Maybe ownership is more about leverage than possession. This concentration seems to be comfortable for the market. Investors’ perception of Nvidia’s moat’s durability is reflected in its valuation. Google Cloud, AWS, and Azure are all rapidly increasing their capacity. The amount spent on capital projects can reach the tens of billions.
However, monopolies—even unofficial ones—rarely remain uncontested indefinitely. Technology supremacy fosters disruption, according to history. IBM used to seem invincible. AT&T’s markets changed as well. Rules change. Architectures evolve.
When you’re inside a data center with steel racks and buzzing transformers all around you, you can’t help but wonder if we’re seeing the beginning of a new monopolistic era or the beginning of fragmentation.
Once inconspicuous and inexpensive, compute is now crucial and in short supply.
And the next ten years of power could be defined by whoever controls it.