Listening to the silence between market cycles. Three years ago, I sat in a Seattle coffee shop watching a DeFi protocol's TVL chart spike from $50 million to $1.2 billion in six weeks. The numbers were beautiful—until the yield incentives stopped and the liquidity vanished like morning fog. Today, I see the same pattern in the US electricity grid. The interconnection queue is bloated with AI data center projects that exist only on paper: announcements without binding power purchase agreements, land options without construction timelines, press releases without concrete capital commitments. The silence between the market cycles is the sound of phantom loads.
These projects are not real. They are options on options, bets on bets, the same speculative infrastructure that haunted the 2017 ICO boom where white papers promised to decentralize everything but delivered nothing but losses. The difference is that the electricity grid cannot be forked. Every phantom project that occupies a queue slot delays a real project by months, sometimes years. Every utility that builds a gas plant based on inflated load forecasts will leave ratepayers holding the bag for decades. And every investor who piles into power infrastructure stocks expecting exponential AI demand is buying into a narrative that may not materialize.
The electricity grid is a memory of past decisions, and phantom loads write bad memories. When I audited those 15 ICO smart contracts in 2017, I learned that smart contracts do not enforce honesty—they enforce whatever is written. The grid's interconnection queue is the same: it does not distinguish between a serious project backed by a signed PPA and a speculative placeholder filed by a developer with no financing. The Federal Energy Regulatory Commission's Order 2023 attempted to reform the queue by increasing deposit requirements, but the bar is still too low. In PJM, the largest US grid operator, over 60 GW of generation and storage projects are waiting in line—and only a fraction will ever be built. AI data centers are now adding to that backlog, filing for hundreds of megawatts each, sometimes by companies that have not yet raised a single dollar of construction funding.
Based on my experience mapping liquidity flows during DeFi Summer, I recognize the signals. In 2020, I tracked $500 million in capital moving across Uniswap and Aave, correlating it with Federal Reserve liquidity injections. The pattern was clear: when money was cheap, TVL inflated; when money tightened, the phantom yields evaporated. The same is happening in the power sector. The AI boom narrative has driven a flood of speculative capital into data center development, and the grid interconnection queue is where these speculative projects park their claims. The difference is that the grid does not have a built-in liquidation mechanism. A phantom project can sit in the queue for years, blocking real projects, without any cost until the developer walks away.
The core insight is that the mismatch between AI's exponential narrative and the grid's linear, long-cycle construction is structural. AI model training doubles every few months, but building a transmission line takes a decade. A gas plant takes four to six years. A nuclear reactor takes 15 years. The grid cannot react to hype cycles; it can only plan for steady-state growth. When the hype cycle injects phantom loads into the planning process, the grid overbuilds, and the cost is borne by all ratepayers. This is not a new problem—it is the same as the 'zombie projects' in renewable energy interconnection queues, where wind and solar farms filed for interconnection years ago and never built. But the AI wave adds scale and urgency, because the loads are larger and the timeline is shorter.
The contrarian view is that the phantom project problem is not a sign of an AI bubble, but a signal of a grid planning failure. The narrative that 'AI electricity demand is a hoax' is as dangerous as the narrative that 'AI electricity demand is infinite.' The truth is that real AI demand is growing—but it is growing from a small base and is concentrated in hyper-scale clusters, not distributed across the country. The phantom projects are not evidence of zero demand; they are evidence of poor market design. The same year I saw the ICO audit failures, I also saw the birth of Aave and Compound—real protocols that survived the winter. The energy industry needs a similar evolution: a mechanism to separate signal from noise.
We are the architects of the next era, and the blueprint must include binding commitments. The solution is not to halt all AI data center construction, but to require proof of commitment before granting interconnection. This could mean mandatory PPAs, performance bonds, or milestone-based queue retention. The EU's gas storage regulation already requires a financial guarantee for capacity booking. The US should adopt similar rules for interconnection queues. The cost of a phantom project is not just the wasted queue slot—it is the delay of real projects, the misallocation of capital, and the risk of stranded assets that will be paid for by the public.
The infrastructure is the story. The next cycle will reward those who build verifiable, binding commitments, not those who announce grand visions with no follow-through. Just as the crypto market learned to distinguish between TVL on a real protocol and TVL on a yield farm, the energy market must learn to distinguish between a real data center and a phantom load. The silence between market cycles is the time to build the mechanisms that ensure the next cycle is built on solid ground, not on promises. Listen to the silence. Then build.
