The discourse surrounding artificial intelligence frequently centers on the velocity of software innovation, the sophistication of large language models, and the rapid iteration of coding frameworks. However, for the engineers and investors constructing the tangible foundation of this industry, the narrative is defined by different metrics: acres of land, megawatts of power, and the flow of cooling fluids. That was the argument of an AI Wednesday segment from the Intelligence Disrupt team, posted September 30, 2026, which moved the focus from the digital abstraction of AI to the physical realities that constrain its expansion: land, buildings, cooling and power.
The speaker, a commentator known for weekly Bitcoin news segments who hosts the AI-focused show on Wednesdays, argued that the high-performance compute (HPC) industry is far more fragmented than the general public assumes. By his count, the industry breaks into six distinct layers. Understanding this stack is essential for anyone trying to map the current bottlenecks in AI growth. Asked 48 hours before recording, he said, he would have named two, maybe three: the landowners with the space and infrastructure, the application layer building tools like Gemini, and perhaps the energy side. The other three, he admitted, he would not have guessed.
The Six Layers of High-Performance Compute
The first layer is energy, the fundamental power source. The second is the physical real estate: the land, the buildings, and the landlords who own the space. The third is the chipmakers, companies like Nvidia producing the GPUs everything runs on. The fourth layer, which the speaker highlighted as a primary focus, is the GPU cloud. This is often referred to as the AI factory, a specialized segment of the market dedicated to orchestrating hardware resources. The fifth layer involves the model developers, the teams assembling large language models (LLMs) and determining how to train them effectively. The sixth and final layer is the application interface, the user experience that everyday people interact with. Without any one of the six, he said, the entire landscape looks drastically different.
Orchestrating the GPU Cloud
Within this taxonomy, the GPU cloud stands out as a critical orchestrator. The speaker identified CoreWeave as a major player in this fourth layer. The function of this layer is not to create the intelligence, but to create the environment in which intelligence can be scaled. It is a software and networking layer that takes thousands of individual, disconnected accelerator chips and binds them together into a single, unified supercomputer.
This is the software and networking layer that transforms thousands of individual disconnected accelerator chips into a single unified supercomputer
Speaker · 1:35
CoreWeave, according to the speaker, is at the forefront of this sub-sector of the AI industry.
The Landlord's Dilemma: Physical Space as Scarcest Asset
Moving up the stack to the second layer, the speaker pointed to the landowners, the building owners, and the cooling infrastructure providers. This is where Core Scientific fits into the equation. Without the physical space, the speaker argued, the rest of the stack cannot function. The constraint here is not digital; it is geographic and logistical. Land is a finite resource, and suitable land with the necessary power grid connections and cooling capabilities is even rarer.
If you do not have the buildings, if you do not have the land, then you simply cannot put your clusters in there
Speaker · 2:57
The AI boom, the speaker argued, is not constrained by the things you might think it is constrained by. If a company cannot secure the buildings and the land, it cannot deploy its clusters. Without the clusters, it cannot train models or build applications. That makes the relationship between the cloud provider and the landowner critical: the cloud provider needs the space to scale.
Cooling and Power: Hidden Constraints
Beyond the simple presence of land, the infrastructure must support the immense heat and power demands of modern AI hardware. The speaker noted that the cooling layer alone is a complex sub-sector, with entire companies dedicated to specific aspects like air filtration or liquid cooling fluid management. These are not minor details; they are existential requirements. As GPU density increases, so does the thermal load, making industrial-grade cooling a prerequisite for operation. The speaker emphasized that the perspective on these constraints varies significantly depending on location, energy contracts, and existing relationships. A facility in one region may face different challenges than one in another, depending on its location and energy contracts.
Power and physical infrastructure are two of the major roadblocks to growth
Speaker · 3:18
The speaker highlighted that a very real physical scarce asset, land and buildings, is one of the primary constraints. If the physical assets are not present, the digital ambitions cannot be realized.
A Blueprint for Scale
The speaker pointed to Core Scientific and CoreWeave as an example of how the layers fit together. Core Scientific holds the physical assets: energized, gigawatt-scale sites with industrial cooling. CoreWeave, on the other hand, provides the orchestration layer, whose business model, he said, depends on securing massive pipeline capacity from operators like these to scale its specialized GPU cloud.
They have energized gigawatt scale sites with the industrial cooling ready to host the AI workloads right now
Speaker · 3:52
Together, they represent a blueprint for the entire AI stack. Core Scientific is the landowner, holding what the speaker described as the rarest commodity in tech: the physical, tangible element. CoreWeave is the operator, the entity that knows how to link all of that physical capacity together and make it useful. This combination addresses the two major roadblocks to growth: power and physical infrastructure.
They hold that rarest of all commodities in tech, the physical tangible element
Speaker · 4:57
The speaker concluded that without any one of the six layers, the landscape looks drastically different. Without the landlords and the physical infrastructure to host, cool and maintain GPU clusters, he said, "you can't do anything," and someone still has to know how to link it all together.
For those interested in seeing these industries converge, the conversation continues at Intelligence Disrupt 2027, March 22 to 24 in Dallas, Texas, where all six sectors will meet on the show floor to discuss the next phase of this competitive market. The speaker encouraged viewers to secure their tickets.