People talk about model parameters, training compute, and tokenomics. They talk about the race to AGI as if it were purely a contest of code and capital. But I keep coming back to a simpler, more stubborn number: 38 gigawatts. That is the power gap Morgan Stanley projects will emerge by 2028 if AI data center demand continues on its current trajectory. For context, that is roughly the equivalent of adding the entire current electricity consumption of a country like Switzerland to the grid, just for servers running inference and training runs.
We like to believe that the digital frontier is weightless, a realm of pure information. But every prompt you type, every image you generate, every autonomous agent you set loose on the internet is ultimately a physical act. It is electrons moving through silicon, generating heat that must be dissipated. And in a bear market, when survival matters more than gains, we must ask the uncomfortable question: what happens when the infrastructure of our digital future runs out of juice?
This is not a story about a technology failure. It is a story about an infrastructure bottleneck that will reshape the competitive landscape of AI, blockchain, and every industry that touches them. It is the collision between an exponential curve of demand and a linear curve of grid expansion. And it is happening right now, beneath the surface of every bullish narrative about AI adoption.
I have spent my career in the intersection of financial engineering and decentralized systems. I have audited whitepapers and governance frameworks. I have seen how trust is built and broken in the digital realm. And from my experience, I can tell you this: the most critical infrastructure is often the least visible. The sequencer running the network, the multi-sig holding the upgrade keys, the power line feeding the data center. The 38GW gap is a governance failure waiting to happen, not because of malicious actors, but because of physics.
Let us break down what this number actually means. When we talk about AI compute, we are talking about GPU clusters. A single NVIDIA H100 draws roughly 700 watts under full load. The newer B200 chips push past 1000 watts. If you deploy 100,000 of these chips, you are already looking at 100 megawatts of IT load alone. Multiply that by the scale of the data centers being planned by Microsoft, Google, Amazon, and Meta, and you quickly understand why the demand curve is vertical.
But the IT load is only the beginning. Data centers have a Power Usage Effectiveness (PUE) ratio, typically between 1.2 and 1.5. For every watt of compute, you need 1.2 to 1.5 watts of total power to keep the facility cool, the network running, and the lights on. So that 38GW gap in IT load translates to a potential 45 to 57GW of actual grid demand. That is a staggering figure. It is not just a number on a slide; it is a physical constraint on our collective digital ambition.
The energy industry is already feeling the pressure. Transformer lead times, the unsung heroes of the electrical grid, have stretched from 40 weeks in 2020 to over 120 weeks today. The manufacturers like Schneider Electric, Eaton, and ABB are seeing order books that stretch years into the future. This is not a supply chain hiccup; it is a fundamental mismatch between the pace of AI innovation and the pace of industrial manufacturing.
In my work with DAOs, I have always emphasized that governance is not just about code, but about the social and physical realities that the code operates within. The same principle applies here. The 38GW gap is a physical manifestation of a governance challenge. Who decides which data centers get the power? Which regions become the new hubs of computational wealth? Who bears the environmental cost of this expansion? These are not technical questions; they are deeply human ones.
The real insight here is that energy is becoming the ultimate centralization force in a world that promised decentralization. We talk about permissionless innovation, but a 120-week lead time on a transformer is a permission gate. It is a barrier to entry that only the largest, most capitalized players can overcome. For a small AI startup or a nascent blockchain project, securing power is now as critical as securing funding. It is the new moat, and it is built on megawatts.
I have seen this dynamic play out in my own experience. In 2020, during DeFi Summer, I co-founded GoverningDAO to help non-technical users understand the risk parameters of lending protocols. The barrier to entry was cognitive. Now, the barrier is physical. A developer in Nairobi with a brilliant idea for a decentralized AI application will face a different, more brutal bottleneck: the lack of affordable, reliable power to run the models. The digital divide is becoming an energy divide.
The contrarian angle that most analysts miss is that this energy crisis might actually be a catalyst for a more efficient, more resilient AI ecosystem. The pressure of the 38GW gap will force innovation in areas we have long neglected. Liquid cooling, which can bring PUE down from 1.4 to below 1.1, is becoming standard for new high-density facilities. Model distillation and quantization are gaining traction because they offer a path to lower power consumption per inference. The economics of scarcity will drive a cultural shift towards doing more with less.
I am reminded of the lessons from the 2022 bear market, when I ran a newsletter called "Resilience & Reality" to help junior developers and investors navigate the FTX collapse. The core message was that trust is earned in bear markets. The same is true for infrastructure. The projects and companies that will thrive in the era of power scarcity are those that acknowledge the constraint and design for it. They will be the ones that secure long-term power purchase agreements, that invest in on-site generation, and that build energy-aware scheduling into their systems.
This is not just a problem for the hyperscalers. It is a problem for every DAO, every Layer 2, every blockchain project that relies on centralized cloud providers for their infrastructure. When AWS raises its prices due to energy costs, the cost is passed down the stack to the end user. The promise of low-fee, decentralized finance becomes harder to keep when the underlying compute is becoming more expensive. The "trustless" system is still dependent on the physical trust of the electrical grid.
So, what is the takeaway? We need to stop treating energy as a mere line item on a data center's operating budget. It is the foundational resource of the digital age. The 38GW gap is a warning shot. It is an invitation to think holistically about the future of AI and blockchain, to integrate energy strategy into governance strategy, and to recognize that our most critical infrastructure is not in the cloud, but in the power lines that feed it.
As I look to the future, I see a world where the most successful organizations, whether they are corporations or DAOs, are those that treat energy procurement with the same rigor as they treat tokenomics or model architecture. They will be the ones that understand that power is not just a utility, but a form of trust. And in a bear market, when every asset is being repriced, the one asset you cannot afford to lose is the trust that your infrastructure will hold. People first, protocol second. Always. And in this case, power is the protocol.