GTC Reflections: Tokenomics and the Rise of the Responsible AI Factory

Author: Michael Welch, CTO at Aligned Data Centers

The Era of Tokenomics: Building the AI Factory Responsibly

If you talk to anyone who attended or tuned into the GTC keynote this year, one takeaway is clear: the legacy data center has evolved  into a dynamic, high-performance AI factory.

Powering this technological leap requires a delicate balance, and building these next-generation AI factories comes with a big responsibility. How do we support growing compute demands while having a positive influence on local communities and protecting scarce electric grid resources? The truth is that one does not have to compromise the other.

Here are a few of my key takeaways from GTC, and how Aligned is engineering the future of AI responsibly:

Moving from Tokenization to “Tokenomics”

To understand the shift, you have to look at the data. When an AI system ingests an image or a document, it breaks that information down into “tokens” – bits of data or fragments of a word. The process of training AI models on all this data is tokenization.

But we are now entering the era of tokenomics.

Tokenomics is all about big picture use of resource. How efficiently can a facility process these tokens? How do you balance the efficiency with the total throughput and speed? What is the cost in terms of power, time, and infrastructure? A deep understanding of these tradeoffs will drive decision making and development plans for AI factories into the future. Anchoring those plans will be the products themselves that are doing the work. NVIDIA’s introduction of the Vera Rubin pods (“SUs” or “Scalable Units” as Nvidia call them) is a perfect example of this shift.

By packaging CPUs, GPUs, and communication layers into one integrated system, these Vera Rubin (VR) pods can efficiently handle large-scale, bandwidth-heavy training workloads and the high-speed demands of inferencing, with the introduction of the new Nvidia Groq 3 LPX inference accelerator.

By intertwining these capabilities, we drive down both the economic and environmental costs of processing tokens, while dramatically increasing the overall performance. At Aligned, we aren’t just building single-use machine learning facilities; we are building adaptive environments capable of supporting this entire, highly efficient AI lifecycle.

The Max P vs. Max Q Debate

In the data center world, we love a good debate about engineering trade-offs. Right now, one of the most important conversations in AI is the “Max P vs. Max Q” dynamic.

I like to explain this with a simple analogy: imagine a car with a governor on the engine.

• Max Q is like cruising at 65 miles per hour on the highway; at that speed, you are at peak efficiency of the car. In an AI factory, this means operating the hardware to prioritize the efficiency at which tokens are processed or generated.
• Max P is taking the governor off and driving 100 miles per hour. You’ll get to your destination faster, but you’ll use more fuel in doing so.

In an era where power is a scarce resource, this choice matters. Our job is to deliver infrastructure that gives our customers the ability to make these decisions intelligently.

Grid Interactivity:  Protecting our greatest asset

One of the ways our commitment to energy stewardship is best demonstrated today is by our approach to grid reliability. True infrastructure responsibility means treating the grid as a shared, dynamic resource and actively enhancing its resilience. Our PDX-01 campus in Hillsboro, Oregon, is that philosophy in action.

When interconnection constraints threatened multi-year delays, Aligned partnered with Portland General Electric (PGE) and Calibrant Energy to pioneer a new path. Together, we deployed an Aligned-funded Battery Energy Storage System (BESS) that bypassed the bottleneck. This innovative solution delivered critical capacity years ahead of schedule while actively enhancing local grid reliability – all at no cost to ratepayers.

During his keynote, Jensen emphasized the critical need to transform AI factories into flexible energy assets. This addresses a reality that utility leaders have rightly pointed out: the U.S. does not necessarily suffer from a base energy supply problem; we face a coincident peak problem. Our electrical grid is engineered to handle a few extreme-demand days, leaving large amounts of capacity underutilized the rest of the year. How can that excess capacity be used to generate tokens?

By building “hybrid” AI factories, we can use co-located generation to get our customers online and generating valuable tokens faster. By tying into utility data in real-time, we can also adjust their operations based on what the community actually needs, potentially even using that exact same infrastructure to supply power back to the grid during times of stress.

For example, on a highly windy day in Texas when excess renewable energy is flooding the grid, a facility can tap into that surplus to maximize token generation. Conversely, it can curtail usage when the local community needs that power back. This ability to actively participate in grid dynamics is a unique benefit that the AI factory approach brings.  

The Bottom Line

The vision shared for the future of AI at GTC is bold and exciting, and it requires a physical home engineered for high density, ultimate thermal efficiency, and relentless reliability.

By mastering tokenomics, engineering for flexible efficiency, and pioneering grid-interactive facilities, we can power the tech of tomorrow while being the great neighbors our communities deserve today.

AI Factory Data Center: Key Questions Answered

What is an AI factory? An AI factory is a high-performance data center designed to handle large-scale AI training and inference workloads, optimized for compute density, cooling, and energy use.

What is tokenomics in AI infrastructure? Tokenomics refers to how efficiently a system processes tokens, factoring in power consumption, throughput, latency, and cost.

What is the difference between Max P and Max Q in AI systems? Max P prioritizes maximum performance and speed, while Max Q focuses on operating at peak efficiency with lower energy consumption.

Why do AI data centers use so much power? AI workloads require intensive compute, especially for training large models. Power use depends on how efficiently infrastructure converts energy into compute output.

What are grid-interactive data centers? Grid-interactive data centers can adjust their energy usage in real time based on grid conditions, helping balance supply and demand.

How can data centers support local communities? Through energy coordination, job creation, and infrastructure investment, while minimizing strain on shared resources like the electrical grid.